diff --git a/common/arg.cpp b/common/arg.cpp
index 5bdc25e15..fdbd4f9b7 100644
--- a/common/arg.cpp
+++ b/common/arg.cpp
@@ -1082,7 +1082,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         [](common_params & params) {
             params.verbose_prompt = true;
         }
-    ));
+    ).set_examples({LLAMA_EXAMPLE_COMPLETION, LLAMA_EXAMPLE_CLI, LLAMA_EXAMPLE_EMBEDDING, LLAMA_EXAMPLE_RETRIEVAL}));
     add_opt(common_arg(
         {"--display-prompt"},
         {"--no-display-prompt"},
@@ -2810,6 +2810,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.port = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_PORT"));
+    add_opt(common_arg(
+        {"--reuse-port"},
+        string_format("allow multiple sockets to bind to the same port (default: %s)", params.reuse_port ? "enabled" : "disabled"),
+        [](common_params & params) {
+            params.reuse_port = true;
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_REUSE_PORT"));
     add_opt(common_arg(
         {"--path"}, "PATH",
         string_format("path to serve static files from (default: %s)", params.public_path.c_str()),
@@ -2846,6 +2853,15 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.webui_mcp_proxy = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_WEBUI_MCP_PROXY"));
+    add_opt(common_arg(
+        {"--tools"}, "TOOL1,TOOL2,...",
+        "experimental: whether to enable built-in tools for AI agents - do not enable in untrusted environments (default: no tools)\n"
+        "specify \"all\" to enable all tools\n"
+        "available tools: read_file, file_glob_search, grep_search, exec_shell_command, write_file, edit_file, apply_diff",
+        [](common_params & params, const std::string & value) {
+            params.server_tools = parse_csv_row(value);
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_TOOLS"));
     add_opt(common_arg(
         {"--webui"},
         {"--no-webui"},
diff --git a/common/chat-diff-analyzer.cpp b/common/chat-diff-analyzer.cpp
index e35596b93..414ee892f 100644
--- a/common/chat-diff-analyzer.cpp
+++ b/common/chat-diff-analyzer.cpp
@@ -287,7 +287,7 @@ void analyze_reasoning::compare_reasoning_presence() {
             return p.literal(reasoning_content) + p.space() + p.optional(p.tag("post", (p.marker() + p.space())) + p.rest());
         });
         auto parser_wrapped = build_tagged_peg_parser([&](common_peg_parser_builder &p) {
-            return p.tag("pre", p.marker()) + p.space() + p.literal(reasoning_content) + p.space() + p.tag("post", (p.marker() + p.space())) + p.rest();
+            return p.tag("pre", p.marker() + p.space()) + p.literal(reasoning_content) + p.space() + p.tag("post", (p.marker() + p.space())) + p.rest();
         });
         // try the more aggressive parse first, if it fails, fall back to the delimiter one
         auto result = parser_wrapped.parse_anywhere_and_extract(comparison->output_B);
@@ -297,7 +297,7 @@ void analyze_reasoning::compare_reasoning_presence() {
         if (result.result.success()) {
             if (!result.tags["pre"].empty() && !result.tags["post"].empty()) {
                 mode = reasoning_mode::TAG_BASED;
-                start = trim_whitespace(result.tags["pre"]);
+                start = trim_leading_whitespace(result.tags["pre"]);
                 end   = trim_trailing_whitespace(result.tags["post"]);
             } else if (!result.tags["post"].empty()) {
                 mode = reasoning_mode::TAG_BASED;
@@ -333,7 +333,7 @@ void analyze_reasoning::compare_thinking_enabled() {
     if (left_trimmed.empty() && !diff.right.empty()) {
         if (!right_trimmed.empty() && string_ends_with(comparison->output_B, right_trimmed)) {
             if (start.empty()) {
-                start = right_trimmed;
+                start = trim_leading_whitespace(diff.right);
                 mode  = reasoning_mode::TAG_BASED;
             }
         }
@@ -344,7 +344,7 @@ void analyze_reasoning::compare_thinking_enabled() {
                 if (seg.size() >= 2 && seg[seg.size() - 1].value == left_trimmed && seg[seg.size() - 2].type == segment_type::MARKER) {
                     start = seg[seg.size() - 2].value;
                 }
-                end = left_trimmed;
+                end = trim_trailing_whitespace(diff.left);
                 mode = reasoning_mode::TAG_BASED;
             }
         }
@@ -363,15 +363,23 @@ void analyze_reasoning::compare_thinking_enabled() {
             size_t len = std::min(base.size(), anchor_len);
             std::string anchor = base.substr(base.size() - len);
             auto pos = extended.rfind(anchor);
-            if (pos == std::string::npos || pos + len >= extended.size()) continue;
+            if (pos == std::string::npos || pos + len >= extended.size()) {
+                continue;
+            }
 
             std::string extra = trim_whitespace(extended.substr(pos + len));
-            if (extra.empty()) continue;
+            if (extra.empty()) {
+                continue;
+            }
 
             auto seg = prune_whitespace_segments(segmentize_markers(extra));
             if (seg.size() == 2 && seg[0].type == segment_type::MARKER && seg[1].type == segment_type::MARKER) {
-                if (start.empty()) start = seg[0].value;
-                if (end.empty())   end   = seg[1].value;
+                if (start.empty()) {
+                    start = seg[0].value;
+                }
+                if (end.empty()) {
+                    end   = seg[1].value;
+                }
                 mode = reasoning_mode::TAG_BASED;
                 break;
             }
@@ -423,7 +431,7 @@ void analyze_reasoning::compare_reasoning_scope() {
         LOG_DBG(ANSI_ORANGE "%s: Detected TOOLS_ONLY reasoning mode\n" ANSI_RESET, __func__);
 
         auto parser_wrapped = build_tagged_peg_parser([&](common_peg_parser_builder &p) {
-            return p.tag("pre", p.marker()) + p.space() + p.literal(reasoning_content) + p.space() + p.tag("post", (p.marker() + p.space()));
+            return p.tag("pre", p.marker() + p.space()) + p.literal(reasoning_content) + p.space() + p.tag("post", (p.marker() + p.space()));
         });
         auto result = parser_wrapped.parse_anywhere_and_extract(comparison->output_B);
         if (result.result.success()) {
@@ -516,7 +524,7 @@ analyze_content::analyze_content(const common_chat_template & tmpl, const analyz
         // Take the more promising diff
         std::string pure_content = rdiff.length() > diff_tools.left.length() ? rdiff : diff_tools.left;
         auto parser_wrapped = build_tagged_peg_parser([&](common_peg_parser_builder &p) {
-            return p.tag("pre", p.marker()) + p.space() + p.literal(response) + p.space() + p.tag("post", (p.marker() + p.space())) + p.rest();
+            return p.tag("pre", p.marker() + p.space()) + p.literal(response) + p.space() + p.tag("post", (p.marker() + p.space())) + p.rest();
         });
         auto result = parser_wrapped.parse_anywhere_and_extract(pure_content);
         start = result.tags["pre"];
diff --git a/common/chat.cpp b/common/chat.cpp
index 078e0cee2..2643cd688 100644
--- a/common/chat.cpp
+++ b/common/chat.cpp
@@ -986,6 +986,7 @@ static common_chat_params common_chat_params_init_gpt_oss(const common_chat_temp
     auto has_tools           = inputs.tools.is_array() && !inputs.tools.empty();
     auto has_response_format = !inputs.json_schema.is_null() && inputs.json_schema.is_object();
     auto include_grammar     = has_response_format || (has_tools && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE);
+    auto extract_reasoning   = inputs.reasoning_format != COMMON_REASONING_FORMAT_NONE;
 
     auto parser = build_chat_peg_parser([&](common_chat_peg_builder & p) {
         auto start           = p.rule("start", p.literal("<|start|>assistant"));
@@ -994,7 +995,13 @@ static common_chat_params common_chat_params_init_gpt_oss(const common_chat_temp
         auto channel         = p.literal("<|channel|>") + (p.literal("commentary") | p.literal("analysis"));
         auto constrain_type  = p.chars("[A-Za-z0-9_-]", 1, -1);
 
-        auto analysis = p.rule("analysis", p.literal("<|channel|>analysis<|message|>") + p.reasoning(content) + end);
+        if (extract_reasoning) {
+            p.rule("analysis", p.literal("<|channel|>analysis<|message|>") + p.reasoning(content) + end);
+        } else {
+            p.rule("analysis", p.content(p.literal("<|channel|>analysis<|message|>") + content + end));
+        }
+
+        auto analysis = p.ref("analysis");
         auto preamble = p.rule("preamble", p.literal("<|channel|>commentary<|message|>") + p.content(content) + end);
         auto final_msg = p.rule("final", p.literal("<|channel|>final<|message|>") + p.content(content));
         auto any = p.rule("any", preamble | analysis);
diff --git a/common/common.cpp b/common/common.cpp
index f7d41590c..4ea4eaea1 100644
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -663,6 +663,38 @@ bool string_parse_kv_override(const char * data, std::vector<llama_model_kv_over
     return true;
 }
 
+// simple glob: * matches non-/ chars, ** matches anything including /
+static inline bool glob_match(const char * pattern, const char * str) {
+    if (*pattern == '\0') {
+        return *str == '\0';
+    }
+    if (pattern[0] == '*' && pattern[1] == '*') {
+        const char * p = pattern + 2;
+        if (*p == '/') p++;
+        if (glob_match(p, str)) return true;
+        if (*str != '\0') return glob_match(pattern, str + 1);
+        return false;
+    }
+    if (*pattern == '*') {
+        const char * p = pattern + 1;
+        for (; *str != '\0' && *str != '/'; str++) {
+            if (glob_match(p, str)) return true;
+        }
+        return glob_match(p, str);
+    }
+    if (*pattern == '?' && *str != '\0' && *str != '/') {
+        return glob_match(pattern + 1, str + 1);
+    }
+    if (*pattern == *str) {
+        return glob_match(pattern + 1, str + 1);
+    }
+    return false;
+}
+
+bool glob_match(const std::string & pattern, const std::string & str) {
+    return glob_match(pattern.c_str(), str.c_str());
+}
+
 //
 // Filesystem utils
 //
diff --git a/common/common.h b/common/common.h
index 3e889e739..40f0ea568 100644
--- a/common/common.h
+++ b/common/common.h
@@ -570,6 +570,7 @@ struct common_params {
 
     // server params
     int32_t port                = 8080;          // server listens on this network port
+    bool    reuse_port          = false;         // allow multiple sockets to bind to the same port
     int32_t timeout_read        = 600;           // http read timeout in seconds
     int32_t timeout_write       = timeout_read;  // http write timeout in seconds
     int32_t n_threads_http      = -1;    // number of threads to process HTTP requests (TODO: support threadpool)
@@ -610,6 +611,9 @@ struct common_params {
     bool endpoint_props   = false; // only control POST requests, not GET
     bool endpoint_metrics = false;
 
+    // enable built-in tools
+    std::vector<std::string> server_tools;
+
     // router server configs
     std::string models_dir    = ""; // directory containing models for the router server
     std::string models_preset = ""; // directory containing model presets for the router server
@@ -787,6 +791,8 @@ std::string string_from(const std::vector<int> & values);
 std::string string_from(const struct llama_context * ctx, const std::vector<llama_token> & tokens);
 std::string string_from(const struct llama_context * ctx, const struct llama_batch & batch);
 
+bool glob_match(const std::string & pattern, const std::string & str);
+
 //
 // Filesystem utils
 //
diff --git a/common/hf-cache.cpp b/common/hf-cache.cpp
index 80dcab017..665c9ff06 100644
--- a/common/hf-cache.cpp
+++ b/common/hf-cache.cpp
@@ -26,6 +26,8 @@ namespace nl = nlohmann;
 #include <windows.h>
 #else
 #define HOME_DIR "HOME"
+#include <unistd.h>
+#include <pwd.h>
 #endif
 
 namespace hf_cache {
@@ -51,6 +53,13 @@ static fs::path get_cache_directory() {
                 return entry.path.empty() ? base : base / entry.path;
             }
         }
+#ifndef _WIN32
+        const struct passwd * pw = getpwuid(getuid());
+
+        if (pw->pw_dir && *pw->pw_dir) {
+            return fs::path(pw->pw_dir) / ".cache" / "huggingface" / "hub";
+        }
+#endif
         throw std::runtime_error("Failed to determine HF cache directory");
     }();
 
diff --git a/common/json-schema-to-grammar.cpp b/common/json-schema-to-grammar.cpp
index c7057d651..e2c4d6ce2 100644
--- a/common/json-schema-to-grammar.cpp
+++ b/common/json-schema-to-grammar.cpp
@@ -416,15 +416,30 @@ private:
                     i++;
                 } else if (c == '(') {
                     i++;
-                    if (i < length) {
-                        if (sub_pattern[i] == '?') {
+                    if (i < length && sub_pattern[i] == '?') {
+                        if (i + 1 < length && sub_pattern[i + 1] == ':') {
+                            i += 2; // skip "?:" for non-capturing group, treat as regular group
+                        } else {
+                            // lookahead/lookbehind (?=, ?!, ?<=, ?<!) - not supported
                             _warnings.push_back("Unsupported pattern syntax");
+                            // skip to matching ')' to avoid UB on empty seq
+                            int depth = 1;
+                            while (i < length && depth > 0) {
+                                if (sub_pattern[i] == '\\' && i + 1 < length) {
+                                    i += 2; // skip escaped character
+                                } else {
+                                    if (sub_pattern[i] == '(') depth++;
+                                    else if (sub_pattern[i] == ')') depth--;
+                                    i++;
+                                }
+                            }
+                            continue;
                         }
                     }
                     seq.emplace_back("(" + to_rule(transform()) + ")", false);
                 } else if (c == ')') {
                     i++;
-                    if (start > 0 && sub_pattern[start - 1] != '(') {
+                    if (start > 0 && sub_pattern[start - 1] != '(' && (start < 2 || sub_pattern[start - 2] != '?' || sub_pattern[start - 1] != ':')) {
                         _errors.push_back("Unbalanced parentheses");
                     }
                     return join_seq();
diff --git a/common/reasoning-budget.cpp b/common/reasoning-budget.cpp
index 2ef744278..cc408a686 100644
--- a/common/reasoning-budget.cpp
+++ b/common/reasoning-budget.cpp
@@ -115,9 +115,11 @@ static void common_reasoning_budget_accept(struct llama_sampler * smpl, llama_to
             break;
         }
         case REASONING_BUDGET_FORCING:
-            // force_pos is advanced in apply(), not here.
-            // This ensures the first forced token isn't skipped when the sampler
-            // is initialized directly in FORCING state (e.g. COUNTING + budget=0)
+            ctx->force_pos++;
+            if (ctx->force_pos >= ctx->forced_tokens.size()) {
+                ctx->state = REASONING_BUDGET_DONE;
+                LOG_INF("reasoning-budget: forced sequence complete, done\n");
+            }
             break;
         case REASONING_BUDGET_DONE:
             break;
@@ -144,14 +146,6 @@ static void common_reasoning_budget_apply(struct llama_sampler * smpl, llama_tok
             cur_p->data[i].logit = -INFINITY;
         }
     }
-
-    // advance to next forced token (done here rather than in accept so that
-    // the first forced token isn't skipped when starting in FORCING state)
-    ctx->force_pos++;
-    if (ctx->force_pos >= ctx->forced_tokens.size()) {
-        ctx->state = REASONING_BUDGET_DONE;
-        LOG_INF("reasoning-budget: forced sequence complete, done\n");
-    }
 }
 
 static void common_reasoning_budget_reset(struct llama_sampler * smpl) {
@@ -261,3 +255,10 @@ struct llama_sampler * common_reasoning_budget_init(
         common_reasoning_budget_state    initial_state) {
     return common_reasoning_budget_init_state(vocab, start_tokens, end_tokens, forced_tokens, budget, initial_state);
 }
+
+common_reasoning_budget_state common_reasoning_budget_get_state(const struct llama_sampler * smpl) {
+    if (!smpl) {
+        return REASONING_BUDGET_IDLE;
+    }
+    return ((const common_reasoning_budget_ctx *)smpl->ctx)->state;
+}
diff --git a/common/reasoning-budget.h b/common/reasoning-budget.h
index 130afdea4..ee1a30ed3 100644
--- a/common/reasoning-budget.h
+++ b/common/reasoning-budget.h
@@ -51,3 +51,5 @@ struct llama_sampler * common_reasoning_budget_init(
         const std::vector<llama_token> & forced_tokens,
         int32_t                          budget,
         common_reasoning_budget_state    initial_state);
+
+common_reasoning_budget_state common_reasoning_budget_get_state(const struct llama_sampler * smpl);
diff --git a/common/sampling.cpp b/common/sampling.cpp
index 012e21266..5259c5f3c 100644
--- a/common/sampling.cpp
+++ b/common/sampling.cpp
@@ -7,6 +7,7 @@
 
 #include <algorithm>
 #include <cctype>
+#include <climits>
 #include <cmath>
 #include <cstring>
 #include <unordered_map>
@@ -109,6 +110,7 @@ struct common_sampler {
     common_params_sampling params;
 
     struct llama_sampler * grmr;
+    struct llama_sampler * rbudget;
     struct llama_sampler * chain;
 
     ring_buffer<llama_token> prev;
@@ -188,6 +190,7 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, st
     lparams.no_perf = params.no_perf;
 
     llama_sampler * grmr = nullptr;
+    llama_sampler * rbudget = nullptr;
     llama_sampler * chain = llama_sampler_chain_init(lparams);
 
     std::vector<llama_sampler *> samplers;
@@ -270,7 +273,7 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, st
             }
         }
 
-        if (grmr) {
+        if (grmr && !params.grammar_lazy) {
             try {
                 for (const auto & token : prefill_tokens) {
                     llama_sampler_accept(grmr, token);
@@ -284,15 +287,15 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, st
         }
     }
 
-    // reasoning budget sampler — added first so it can force tokens before other samplers
-    if (params.reasoning_budget_tokens >= 0 && !params.reasoning_budget_forced.empty()) {
-        samplers.push_back(common_reasoning_budget_init(
+    // reasoning budget sampler
+    if (!params.reasoning_budget_start.empty() && !params.reasoning_budget_end.empty()) {
+        rbudget = common_reasoning_budget_init(
             vocab,
             params.reasoning_budget_start,
             params.reasoning_budget_end,
             params.reasoning_budget_forced,
-            params.reasoning_budget_tokens,
-            prefill_tokens));
+            params.reasoning_budget_tokens < 0 ? INT_MAX : params.reasoning_budget_tokens,
+            prefill_tokens);
     }
 
     if (params.has_logit_bias()) {
@@ -383,6 +386,7 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, st
     auto * result = new common_sampler {
         /* .params  = */ params,
         /* .grmr    = */ grmr,
+        /* .rbudget = */ rbudget,
         /* .chain   = */ chain,
         /* .prev    = */ ring_buffer<llama_token>(std::max(32, params.n_prev)),
         /* .cur     = */ {},
@@ -398,11 +402,27 @@ void common_sampler_free(struct common_sampler * gsmpl) {
     }
 
     llama_sampler_free(gsmpl->grmr);
+    llama_sampler_free(gsmpl->rbudget);
     llama_sampler_free(gsmpl->chain);
 
     delete gsmpl;
 }
 
+static bool grammar_should_apply(struct common_sampler * gsmpl) {
+    if (!gsmpl->grmr) {
+        return false;
+    }
+    if (!gsmpl->rbudget) {
+        return true;
+    }
+    if (gsmpl->params.grammar_lazy) {
+        // if grammar is lazy, only apply when reasoning budget is not active
+        const auto state = common_reasoning_budget_get_state(gsmpl->rbudget);
+        return state == REASONING_BUDGET_IDLE || state == REASONING_BUDGET_DONE;
+    }
+    return true;
+}
+
 void common_sampler_accept(struct common_sampler * gsmpl, llama_token token, bool accept_grammar) {
     if (!gsmpl) {
         return;
@@ -410,6 +430,11 @@ void common_sampler_accept(struct common_sampler * gsmpl, llama_token token, boo
 
     const auto tm = gsmpl->tm();
 
+    // grammar_should_apply() checks the reasoning budget state, so calculate this before we accept
+    accept_grammar = accept_grammar && grammar_should_apply(gsmpl);
+
+    llama_sampler_accept(gsmpl->rbudget, token);
+
     if (gsmpl->grmr && accept_grammar) {
         llama_sampler_accept(gsmpl->grmr, token);
     }
@@ -431,6 +456,7 @@ struct common_sampler * common_sampler_clone(common_sampler * gsmpl) {
     return new common_sampler {
         /* .params  = */ gsmpl->params,
         /* .grmr    = */ llama_sampler_clone(gsmpl->grmr),
+        /* .rbudget = */ llama_sampler_clone(gsmpl->rbudget),
         /* .chain   = */ llama_sampler_clone(gsmpl->chain),
         /* .prev    = */ gsmpl->prev,
         /* .cur     = */ gsmpl->cur,
@@ -500,6 +526,7 @@ llama_token common_sampler_sample(struct common_sampler * gsmpl, struct llama_co
     llama_token id = LLAMA_TOKEN_NULL;
 
     auto & grmr  = gsmpl->grmr;
+    auto & rbudget = gsmpl->rbudget;
     auto & chain = gsmpl->chain;
     auto & cur_p = gsmpl->cur_p; // initialized by set_logits
 
@@ -511,7 +538,8 @@ llama_token common_sampler_sample(struct common_sampler * gsmpl, struct llama_co
         if (id != LLAMA_TOKEN_NULL) {
             LOG_DBG("%s: Backend sampler selected token: '%d'. Will not run any CPU samplers\n", __func__, id);
 
-            GGML_ASSERT(!gsmpl->grmr && "using grammar in combination with backend sampling is not supported");
+            GGML_ASSERT(!gsmpl->grmr    && "using grammar in combination with backend sampling is not supported");
+            GGML_ASSERT(!gsmpl->rbudget && "using reasoning budget in combination with backend sampling is not supported");
 
             // TODO: simplify
             gsmpl->cur.resize(1);
@@ -524,7 +552,10 @@ llama_token common_sampler_sample(struct common_sampler * gsmpl, struct llama_co
 
     gsmpl->set_logits(ctx, idx);
 
-    if (grammar_first) {
+    // apply reasoning budget first
+    llama_sampler_apply(rbudget, &cur_p);
+
+    if (grammar_first && grammar_should_apply(gsmpl)) {
         llama_sampler_apply(grmr, &cur_p);
     }
 
@@ -532,7 +563,7 @@ llama_token common_sampler_sample(struct common_sampler * gsmpl, struct llama_co
 
     id = cur_p.data[cur_p.selected].id;
 
-    if (grammar_first) {
+    if (grammar_first || !grammar_should_apply(gsmpl)) {
         return id;
     }
 
@@ -553,7 +584,12 @@ llama_token common_sampler_sample(struct common_sampler * gsmpl, struct llama_co
     // if the token is not valid, sample again, but first apply the grammar sampler and then the sampling chain
     gsmpl->set_logits(ctx, idx);
 
-    llama_sampler_apply(grmr,  &cur_p);
+    llama_sampler_apply(rbudget,  &cur_p);
+
+    if (grammar_should_apply(gsmpl)) {
+        llama_sampler_apply(grmr,  &cur_p);
+    }
+
     llama_sampler_apply(chain, &cur_p);
 
     GGML_ASSERT(cur_p.selected != -1 && "no selected token during sampling - check your sampling configuration");
diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
index 82d1004c6..bcf98cfae 100755
--- a/convert_hf_to_gguf.py
+++ b/convert_hf_to_gguf.py
@@ -7150,6 +7150,8 @@ class DeepseekOCRVisionModel(MmprojModel):
             return gguf.GGMLQuantizationType.F32
         if ".rel_pos_h" in name or '.rel_pos_w' in name:
             return gguf.GGMLQuantizationType.F32
+        if ".neck." in name or ".net_" in name:
+            return gguf.GGMLQuantizationType.F32
         return super().tensor_force_quant(name, new_name, bid, n_dims)
 
     def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
diff --git a/ggml/src/ggml-cuda/common.cuh b/ggml/src/ggml-cuda/common.cuh
index abb5957e9..057029ba3 100644
--- a/ggml/src/ggml-cuda/common.cuh
+++ b/ggml/src/ggml-cuda/common.cuh
@@ -809,7 +809,13 @@ static __device__ __forceinline__ float ggml_cuda_e8m0_to_fp32(uint8_t x) {
 static __device__ __forceinline__ float ggml_cuda_ue4m3_to_fp32(uint8_t x) {
 #ifdef FP8_AVAILABLE
     const uint32_t bits = x * (x != 0x7F && x != 0xFF); // Convert NaN to 0.0f to match CPU implementation.
+#if defined(GGML_USE_HIP) && defined(CDNA3)
+    // ROCm dose not support fp8 in software on devices with fp8 hardware,
+    // but CDNA3 supports only e4m3_fnuz (no inf).
+    const __hip_fp8_e4m3_fnuz xf = *reinterpret_cast<const __hip_fp8_e4m3_fnuz *>(&bits);
+#else
     const __nv_fp8_e4m3 xf = *reinterpret_cast<const __nv_fp8_e4m3 *>(&bits);
+#endif // defined(GGML_USE_HIP) && defined(GGML_USE_HIP)
     return static_cast<float>(xf) / 2;
 #else
     NO_DEVICE_CODE;
diff --git a/ggml/src/ggml-metal/ggml-metal-device.m b/ggml/src/ggml-metal/ggml-metal-device.m
index b93316ce6..2e537c0cc 100644
--- a/ggml/src/ggml-metal/ggml-metal-device.m
+++ b/ggml/src/ggml-metal/ggml-metal-device.m
@@ -696,7 +696,7 @@ ggml_metal_device_t ggml_metal_device_init(int device) {
                     "    auto tB = B.slice((int)tgid.x, 0); \n"
                     " \n"
                     "    matmul2d< \n"
-                    "        matmul2d_descriptor(8, 8, dynamic_extent), \n"
+                    "        matmul2d_descriptor(16, 16, dynamic_extent), \n"
                     "        execution_simdgroups<4>> mm; \n"
                     " \n"
                     "    auto cT = mm.get_destination_cooperative_tensor<decltype(tA), decltype(tB), float>(); \n"
@@ -746,7 +746,7 @@ ggml_metal_device_t ggml_metal_device_init(int device) {
                     "    auto tB = B.slice((int)tgid.x, 0); \n"
                     " \n"
                     "    matmul2d< \n"
-                    "        matmul2d_descriptor(8, 8, dynamic_extent), \n"
+                    "        matmul2d_descriptor(16, 16, dynamic_extent), \n"
                     "        execution_simdgroups<4>> mm; \n"
                     " \n"
                     "    auto cT = mm.get_destination_cooperative_tensor<decltype(tA), decltype(tB), float>(); \n"
diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
index 3fc08a304..c77e9777e 100644
--- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -1128,6 +1128,16 @@ struct vk_op_glu_push_constants {
     uint32_t mode;  // 0: default, 1: swapped, 2: split
     float alpha; // for swiglu_oai
     float limit;
+    uint32_t nb01;
+    uint32_t nb02;
+    uint32_t nb03;
+    uint32_t ne01;
+    uint32_t ne02;
+    uint32_t nb11;
+    uint32_t nb12;
+    uint32_t nb13;
+    uint32_t ne11;
+    uint32_t ne12;
 };
 
 struct vk_op_unary_push_constants {
@@ -5074,7 +5084,7 @@ static vk_device ggml_vk_get_device(size_t idx) {
         } else {
             device_queue_create_infos.push_back({vk::DeviceQueueCreateFlags(), compute_queue_family_index, 1, priorities});
         }
-        vk::DeviceCreateInfo device_create_info;
+        vk::DeviceCreateInfo device_create_info{};
         std::vector<const char *> device_extensions;
         vk::PhysicalDeviceFeatures device_features = device->physical_device.getFeatures();
 
@@ -5443,12 +5453,10 @@ static vk_device ggml_vk_get_device(size_t idx) {
 #endif
         device->name = GGML_VK_NAME + std::to_string(idx);
 
-        device_create_info = {
-            vk::DeviceCreateFlags(),
-            device_queue_create_infos,
-            {},
-            device_extensions
-        };
+        device_create_info
+            .setFlags(vk::DeviceCreateFlags())
+            .setQueueCreateInfos(device_queue_create_infos)
+            .setPEnabledExtensionNames(device_extensions);
         device_create_info.setPNext(&device_features2);
         device->device = device->physical_device.createDevice(device_create_info);
 
@@ -11086,8 +11094,6 @@ static void ggml_vk_glu(ggml_backend_vk_context * ctx, vk_context& subctx, const
     const float alpha = op_params_f[2];
     const float limit = op_params_f[3];
 
-    GGML_ASSERT(ggml_is_contiguous(src0));
-
     if (!split) {
         GGML_ASSERT(src0->ne[0] / 2 == dst->ne[0]);
     } else {
@@ -11105,7 +11111,17 @@ static void ggml_vk_glu(ggml_backend_vk_context * ctx, vk_context& subctx, const
             (uint32_t)dst->ne[0],
             mode,
             alpha,
-            limit
+            limit,
+            (uint32_t)(src0->nb[1] / src0->nb[0]),
+            (uint32_t)(src0->nb[2] / src0->nb[0]),
+            (uint32_t)(src0->nb[3] / src0->nb[0]),
+            (uint32_t)src0->ne[1],
+            (uint32_t)src0->ne[2],
+            (uint32_t)(dst->nb[1] / dst->nb[0]),
+            (uint32_t)(dst->nb[2] / dst->nb[0]),
+            (uint32_t)(dst->nb[3] / dst->nb[0]),
+            (uint32_t)dst->ne[1],
+            (uint32_t)dst->ne[2]
         });
 }
 
@@ -15255,8 +15271,7 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
                 case GGML_GLU_OP_SWIGLU_OAI:
                 case GGML_GLU_OP_GEGLU_ERF:
                 case GGML_GLU_OP_GEGLU_QUICK:
-                    return ggml_is_contiguous(op->src[0]) &&
-                           (op->src[0]->type == GGML_TYPE_F32 || op->src[0]->type == GGML_TYPE_F16) &&
+                    return (op->src[0]->type == GGML_TYPE_F32 || op->src[0]->type == GGML_TYPE_F16) &&
                            (op->type == GGML_TYPE_F32 || op->type == GGML_TYPE_F16) &&
                            (op->src[0]->type == op->type);
                 default:
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.glsl b/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.glsl
index 216898934..95298922d 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.glsl
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.glsl
@@ -16,4 +16,14 @@ layout (push_constant) uniform parameter
     uint mode;
     float alpha;
     float limit;
+    uint nb01;
+    uint nb02;
+    uint nb03;
+    uint ne01;
+    uint ne02;
+    uint nb11;
+    uint nb12;
+    uint nb13;
+    uint ne11;
+    uint ne12;
 } p;
diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/glu_main.glsl b/ggml/src/ggml-vulkan/vulkan-shaders/glu_main.glsl
index 85cf65a9e..359461306 100644
--- a/ggml/src/ggml-vulkan/vulkan-shaders/glu_main.glsl
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/glu_main.glsl
@@ -8,22 +8,32 @@ void main() {
     const uint row = i / p.ne20;
     const uint col = i - row * p.ne20;
 
+    const uint i3 = row / (p.ne01 * p.ne02);
+    const uint i2 = (row % (p.ne01 * p.ne02)) / p.ne01;
+    const uint i1 = row % p.ne01;
+    const uint src_idx = i3 * p.nb03 + i2 * p.nb02 + i1 * p.nb01 + col;
+
+    const uint dst_i3 = row / (p.ne11 * p.ne12);
+    const uint dst_i2 = (row % (p.ne11 * p.ne12)) / p.ne11;
+    const uint dst_i1 = row % p.ne11;
+    const uint dst_idx = dst_i3 * p.nb13 + dst_i2 * p.nb12 + dst_i1 * p.nb11 + col;
+
     if (p.mode == 0) {
         // Default
         const uint offset = p.ne00 / 2;
-        const uint idx = row * p.ne00 + col;
+        const uint idx = src_idx;
 
-        data_d[row * offset + col] = D_TYPE(op(float(data_a[idx]), float(data_a[idx + offset])));
+        data_d[dst_idx] = D_TYPE(op(float(data_a[idx]), float(data_a[idx + offset])));
     } else if (p.mode == 1) {
         // Swapped
         const uint offset = p.ne00 / 2;
-        const uint idx = row * p.ne00 + col;
+        const uint idx = src_idx;
 
-        data_d[row * offset + col] = D_TYPE(op(float(data_a[idx + offset]), float(data_a[idx])));
+        data_d[dst_idx] = D_TYPE(op(float(data_a[idx + offset]), float(data_a[idx])));
     } else {
         // Split
-        const uint idx = row * p.ne00 + col;
+        const uint idx = src_idx;
 
-        data_d[idx] = D_TYPE(op(float(data_a[idx]), float(data_b[idx])));
+        data_d[dst_idx] = D_TYPE(op(float(data_a[idx]), float(data_b[idx])));
     }
 }
diff --git a/src/llama-quant.cpp b/src/llama-quant.cpp
index 56c9e8197..b309574ae 100644
--- a/src/llama-quant.cpp
+++ b/src/llama-quant.cpp
@@ -345,9 +345,12 @@ static bool tensor_allows_quantization(const llama_model_quantize_params * param
 
     // do not quantize specific multimodal tensors
     quantize &= name.find(".position_embd") == std::string::npos;
-    quantize &= name.find("sam.patch_embd") == std::string::npos;
     quantize &= name.find("sam.pos_embd")   == std::string::npos;
+    quantize &= name.find("sam.neck.")      == std::string::npos;
+    quantize &= name.find("sam.net_")       == std::string::npos;
     quantize &= name.find(".rel_pos")       == std::string::npos;
+    quantize &= name.find(".patch_embd")    == std::string::npos;
+    quantize &= name.find(".patch_merger")  == std::string::npos;
 
     return quantize;
 }
diff --git a/tools/completion/completion.cpp b/tools/completion/completion.cpp
index fa8e99b75..431d04bd0 100644
--- a/tools/completion/completion.cpp
+++ b/tools/completion/completion.cpp
@@ -147,13 +147,19 @@ int main(int argc, char ** argv) {
 
     ctx   = llama_init->context();
     model = llama_init->model();
-    smpl  = llama_init->sampler(0);
 
     if (ctx == NULL) {
         LOG_ERR("%s: error: unable to create context\n", __func__);
         return 1;
     }
 
+    if (model == NULL) {
+        LOG_ERR("%s: error: unable to load model\n", __func__);
+        return 1;
+    }
+
+    smpl = llama_init->sampler(0);
+
     llama_memory_t mem = llama_get_memory(ctx);
     const llama_vocab * vocab = llama_model_get_vocab(model);
 
@@ -696,7 +702,7 @@ int main(int argc, char ** argv) {
                 if (!common_prompt_batch_decode(ctx, embd, n_past, params.n_batch, path_session, save_now)) {
                     return 1;
                 }
-                session_tokens.insert(session_tokens.end(), embd.begin(), embd.begin());
+                session_tokens.insert(session_tokens.end(), embd.begin(), embd.end());
                 n_session_consumed = session_tokens.size();
                 session_do_save = false;
 
diff --git a/tools/mtmd/clip.cpp b/tools/mtmd/clip.cpp
index 062797aa7..60ed4188c 100644
--- a/tools/mtmd/clip.cpp
+++ b/tools/mtmd/clip.cpp
@@ -1456,6 +1456,16 @@ struct clip_model_loader {
 
             // sanity check
             {
+                if (hparams.image_size < 0) {
+                    // note: some models having hparams.image_size == 0, which means the image size is dynamic
+                    throw std::runtime_error(string_format("%s: image_size (%d) cannot be negative\n", __func__, hparams.image_size));
+                }
+                if (hparams.patch_size <= 0) {
+                    throw std::runtime_error(string_format("%s: patch_size (%d) must be greater than 0\n", __func__, hparams.patch_size));
+                }
+                if (hparams.n_embd <= 0) {
+                    throw std::runtime_error(string_format("%s: n_embd (%d) must be greater than 0\n", __func__, hparams.n_embd));
+                }
                 if (hparams.image_max_pixels < hparams.image_min_pixels) {
                     throw std::runtime_error(string_format("%s: image_max_pixels (%d) is less than image_min_pixels (%d)\n", __func__, hparams.image_max_pixels, hparams.image_min_pixels));
                 }
@@ -2852,7 +2862,6 @@ void setup_init_vision_shim_kcpp(struct clip_ctx * ctx_v) {
     }
 
     GGML_ASSERT(image_preproc != nullptr);
-
     //end of lcpp code block
      // =====
 }
diff --git a/tools/mtmd/mtmd-audio.cpp b/tools/mtmd/mtmd-audio.cpp
index 447f61aaa..e68387c27 100644
--- a/tools/mtmd/mtmd-audio.cpp
+++ b/tools/mtmd/mtmd-audio.cpp
@@ -13,23 +13,20 @@
 
 constexpr bool DEBUG = false;
 
-void mtmd_audio_cache::fill_sin_cos_table(int n) {
+void mtmd_audio_cache::fill_sin_cos_table(uint32_t n) {
     sin_vals.resize(n);
     cos_vals.resize(n);
-    for (int i = 0; i < n; i++) {
+    for (uint32_t i = 0; i < n; i++) {
         double theta = (2 * M_PI * i) / n;
         sin_vals[i]  = sinf(theta);
         cos_vals[i]  = cosf(theta);
     }
 }
 
-void mtmd_audio_cache::fill_hann_window(int length, bool periodic) {
+void mtmd_audio_cache::fill_hann_window(uint32_t length, bool periodic) {
     hann_window.resize(length);
-    int offset = -1;
-    if (periodic) {
-        offset = 0;
-    }
-    for (int i = 0; i < length; i++) {
+    int offset = periodic ? 0 : -1;
+    for (uint32_t i = 0; i < length; i++) {
         hann_window[i] = 0.5 * (1.0 - cosf((2.0 * M_PI * i) / (length + offset)));
     }
 }
@@ -165,6 +162,7 @@ static void dft_impl(const mtmd_audio_cache & cache, const float * in, int N, fl
 //              false = input is complex-valued (interleaved real/imag, stride 2)
 template <bool Inverse, bool RealInput>
 static void fft_impl(const mtmd_audio_cache & cache, float * in, int N, float * out) {
+    GGML_ASSERT(N > 0);
     const int n_sin_cos_vals = cache.sin_vals.size();
 
     if (N == 1) {
@@ -407,6 +405,8 @@ static bool log_mel_spectrogram(
     }
 
 
+    GGML_ASSERT(params.n_fft_bins > 0);
+    GGML_ASSERT(params.hop_length > 0);
     out.n_mel = params.n_mel;
     out.n_len = (n_samples - frame_size) / frame_step + 1;
     // TODO: handle these checks better
@@ -438,6 +438,7 @@ static bool log_mel_spectrogram(
 
     const int effective_n_len = n_samples_in / frame_step;
     if (params.norm_per_feature) {
+        GGML_ASSERT(effective_n_len > 1);
         for (int i = 0; i < out.n_mel; i++) {
             double mean = 0;
             for (int j = 0; j < effective_n_len; ++j) {
@@ -639,6 +640,7 @@ mtmd_audio_streaming_istft::mtmd_audio_streaming_istft(int n_fft, int hop_length
     padding_to_remove((n_fft - hop_length) / 2),
     ifft_in(n_fft * 2 * 4, 0.0f),  // extra space for recursive IFFT
     ifft_out(n_fft * 2 * 4, 0.0f) {
+    GGML_ASSERT(n_fft > 0 && hop_length > 0 && hop_length <= n_fft);
     cache.fill_sin_cos_table(n_fft);
     cache.fill_hann_window(n_fft, true);
 }
diff --git a/tools/mtmd/mtmd-audio.h b/tools/mtmd/mtmd-audio.h
index 016c7392e..53857a2eb 100644
--- a/tools/mtmd/mtmd-audio.h
+++ b/tools/mtmd/mtmd-audio.h
@@ -33,9 +33,9 @@ struct mtmd_audio_cache {
 
     mtmd_audio_mel_filters filters;
 
-    void fill_sin_cos_table(int n);
+    void fill_sin_cos_table(uint32_t n);
 
-    void fill_hann_window(int length, bool periodic);
+    void fill_hann_window(uint32_t length, bool periodic);
 
     // Build mel filterbank matrix [n_mel × n_fft_bins] at runtime.
     // n_fft_bins must be (N_fft / 2 + 1). Example: if N_fft=512 -> n_fft_bins=257.
diff --git a/tools/mtmd/mtmd-helper.cpp b/tools/mtmd/mtmd-helper.cpp
index 718ce67a1..2dfd3c256 100644
--- a/tools/mtmd/mtmd-helper.cpp
+++ b/tools/mtmd/mtmd-helper.cpp
@@ -127,6 +127,7 @@ struct decode_embd_batch {
     std::vector<int8_t>         logits;
     llama_batch batch;
     decode_embd_batch(float * embd, int32_t n_tokens, int n_pos_per_embd, int n_mmproj_embd) : n_pos_per_embd(n_pos_per_embd), n_mmproj_embd(n_mmproj_embd) {
+        GGML_ASSERT(n_tokens > 0 && n_pos_per_embd > 0 && n_mmproj_embd > 0);
         pos     .resize(n_tokens * n_pos_per_embd);
         n_seq_id.resize(n_tokens);
         seq_ids .resize(n_tokens + 1);
@@ -157,6 +158,7 @@ struct decode_embd_batch {
     // M-RoPE for image
     void set_position_mrope_2d(llama_pos pos_0, int nx, int ny, llama_seq_id seq_id) {
         GGML_ASSERT(n_pos_per_embd == 4);
+        GGML_ASSERT(nx > 0 && ny > 0 && nx * ny == batch.n_tokens);
         seq_id_0[0] = seq_id;
         for (int y = 0; y < ny; y++) {
             for (int x = 0; x < nx; x++) {
@@ -192,6 +194,7 @@ struct decode_embd_batch {
     }
 
     llama_batch get_view(int offset, int n_tokens) {
+        GGML_ASSERT(offset >= 0 && n_tokens > 0 && offset + n_tokens <= batch.n_tokens);
         llama_pos * pos_ptr;
         pos_view.clear();
         pos_view.reserve(n_tokens * n_pos_per_embd);
@@ -235,6 +238,7 @@ int32_t mtmd_helper_decode_image_chunk(
         llama_seq_id seq_id,
         int32_t n_batch,
         llama_pos * new_n_past) {
+    GGML_ASSERT(n_batch > 0);
     auto chunk_type = mtmd_input_chunk_get_type(chunk);
     const char * name = chunk_type == MTMD_INPUT_CHUNK_TYPE_IMAGE ? "image" : "audio";
     if (chunk_type == MTMD_INPUT_CHUNK_TYPE_TEXT) {
@@ -312,6 +316,7 @@ int32_t mtmd_helper_eval_chunk_single(mtmd_context * ctx,
         int32_t n_batch,
         bool logits_last,
         llama_pos * new_n_past) {
+    GGML_ASSERT(n_batch > 0);
     int32_t ret;
     llama_batch text_batch = llama_batch_init(n_batch, 0, 1);
     auto chunk_type = mtmd_input_chunk_get_type(chunk);
@@ -508,6 +513,11 @@ mtmd_bitmap * mtmd_helper_bitmap_init_from_file(mtmd_context * ctx, const char *
     fseek(f, 0, SEEK_END);
     long file_size = ftell(f);
     fseek(f, 0, SEEK_SET);
+    if (file_size < 0) {
+        LOG_ERR("Failed to get file size of %s\n", fname);
+        fclose(f);
+        return nullptr;
+    }
     buf.resize(file_size);
 
     size_t n_read = fread(buf.data(), 1, file_size, f);
diff --git a/tools/mtmd/mtmd-image.cpp b/tools/mtmd/mtmd-image.cpp
index b44643719..a2166622b 100644
--- a/tools/mtmd/mtmd-image.cpp
+++ b/tools/mtmd/mtmd-image.cpp
@@ -99,6 +99,8 @@ struct img_tool {
     }
 
     static void crop(const clip_image_u8 & image, clip_image_u8 & dst, int x, int y, int w, int h) {
+        GGML_ASSERT(x >= 0 && y >= 0 && w > 0 && h > 0);
+        GGML_ASSERT(x + w <= image.nx && y + h <= image.ny);
         dst.nx = w;
         dst.ny = h;
         dst.buf.resize(3 * w * h);
@@ -196,6 +198,7 @@ struct img_tool {
 private:
     // Bilinear resize function
     static void resize_bilinear(const clip_image_u8 & src, clip_image_u8 & dst, int target_width, int target_height) {
+        GGML_ASSERT(src.nx >= 2 && src.ny >= 2);
         dst.nx = target_width;
         dst.ny = target_height;
         dst.buf.resize(3 * target_width * target_height);
@@ -207,8 +210,8 @@ private:
             for (int x = 0; x < target_width; x++) {
                 float px = x_ratio * x;
                 float py = y_ratio * y;
-                int x_floor = static_cast<int>(px);
-                int y_floor = static_cast<int>(py);
+                int x_floor = std::min(static_cast<int>(px), src.nx - 2);
+                int y_floor = std::min(static_cast<int>(py), src.ny - 2);
                 float x_lerp = px - x_floor;
                 float y_lerp = py - y_floor;
 
@@ -347,6 +350,7 @@ private:
         // Returns: kernel size (ksize) - number of input pixels that contribute to each output pixel
         auto precompute_weights = [&](int inSize, int outSize,
                                      std::vector<int> & bounds, std::vector<int32_t> & weights) -> int {
+            GGML_ASSERT(inSize > 0 && outSize > 0);
             double support, scale, filterscale;
             double center, ww, ss;
             int xx, x, ksize, xmin, xmax, xcnt;
diff --git a/tools/mtmd/mtmd.cpp b/tools/mtmd/mtmd.cpp
index d078120f7..9c400ce10 100644
--- a/tools/mtmd/mtmd.cpp
+++ b/tools/mtmd/mtmd.cpp
@@ -641,6 +641,11 @@ struct mtmd_tokenizer {
                 add_text(ctx->img_beg, true); // add image begin token
             }
 
+            // sanity check
+            GGML_ASSERT(bitmap->nx > 0 && bitmap->ny > 0);
+            GGML_ASSERT(bitmap->data.size() == (size_t)bitmap->nx * bitmap->ny * 3);
+            GGML_ASSERT(ctx->image_preproc != nullptr);
+
             // convert mtmd_bitmap to clip_image_u8
             clip_image_u8_ptr img_u8(clip_image_u8_init());
             img_u8->nx = bitmap->nx;
@@ -649,7 +654,6 @@ struct mtmd_tokenizer {
             std::memcpy(img_u8->buf.data(), bitmap->data.data(), img_u8->nx * img_u8->ny * 3);
 
             // preprocess image
-            GGML_ASSERT(ctx->image_preproc != nullptr);
             clip_image_f32_batch batch_f32;
             bool ok = ctx->image_preproc->preprocess(*img_u8, batch_f32);
             if (!ok) {
@@ -773,6 +777,11 @@ struct mtmd_tokenizer {
                 add_text(ctx->aud_beg, true); // add audio begin token
             }
 
+            // sanity check
+            GGML_ASSERT(ctx->audio_preproc != nullptr);
+            GGML_ASSERT(bitmap->data.size() > sizeof(float));
+            GGML_ASSERT(bitmap->data.size() % sizeof(float) == 0);
+
             // preprocess audio
             std::vector<mtmd_audio_mel> mel_spec_chunks;
             const float * samples = (const float *)bitmap->data.data();
diff --git a/tools/server/README-dev.md b/tools/server/README-dev.md
index 5f82e35d6..1dd00b72e 100644
--- a/tools/server/README-dev.md
+++ b/tools/server/README-dev.md
@@ -125,6 +125,61 @@ The framework automatically starts a `llama-server` instance, sends requests, an
 
 For detailed instructions, see the [test documentation](./tests/README.md).
 
+### API for tools
+
+This endpoint is intended to be used internally by the Web UI and subject to change or to be removed in the future.
+
+**GET /tools**
+
+Get a list of tools, each tool has these fields:
+- `tool` (string): the ID name of the tool, to be used in POST call. Example: `read_file`
+- `display_name` (string): the name to be displayed on UI. Example: `Read file`
+- `type` (string): always be `"builtin"` for now
+- `permissions` (object): a mapping string --> boolean that indicates the permission required by this tool. This is useful for the UI to ask the user before calling the tool. For now, the only permission supported is `"write"`
+- `definition` (object): the OAI-compat definition of this tool
+
+**POST /tools**
+
+Invoke a tool call, request body is a JSON object with:
+- `tool` (string): the name of the tool
+- `params` (object): a mapping from argument name (string) to argument value
+
+Returns JSON object. There are two response formats:
+
+Format 1: Plain text. The text will be placed into a field called `plain_text_response`, example:
+
+```json
+{
+    "plain_text_response": "this is a text response"
+}
+```
+
+The client should extract this value and place it inside message content (note: content is no longer a JSON), example
+
+```json
+{
+    "role": "tool",
+    "content": "this is a text response"
+}
+```
+
+Format 2: Normal JSON response, example:
+
+```json
+{
+    "error": "cannot open this file"
+}
+```
+
+That requires `JSON.stringify` when formatted to message content:
+
+```json
+{
+    "role": "tool",
+    "content": "{\"error\":\"cannot open this file\"}"
+}
+```
+
 ### Notable Related PRs
 
 - Initial server implementation: https://github.com/ggml-org/llama.cpp/pull/1443
diff --git a/tools/server/public/index.html.gz b/tools/server/public/index.html.gz
index 14d115fa5..5e7326a58 100644
Binary files a/tools/server/public/index.html.gz and b/tools/server/public/index.html.gz differ
diff --git a/tools/server/server-common.cpp b/tools/server/server-common.cpp
index e01c8c53d..ed5e306fc 100644
--- a/tools/server/server-common.cpp
+++ b/tools/server/server-common.cpp
@@ -1110,7 +1110,7 @@ json oaicompat_chat_params_parse(
             reasoning_budget = json_value(body, "thinking_budget_tokens", -1);
         }
 
-        if (reasoning_budget >= 0 && !chat_params.thinking_end_tag.empty()) {
+        if (!chat_params.thinking_end_tag.empty()) {
             llama_params["reasoning_budget_tokens"] = reasoning_budget;
             llama_params["reasoning_budget_start_tag"] = chat_params.thinking_start_tag;
             llama_params["reasoning_budget_end_tag"] = chat_params.thinking_end_tag;
diff --git a/tools/server/server-context.cpp b/tools/server/server-context.cpp
index b79a5270b..6f737d94d 100644
--- a/tools/server/server-context.cpp
+++ b/tools/server/server-context.cpp
@@ -2493,7 +2493,7 @@ private:
                     bool has_mtmd = false;
 
                     // check if we should process the image
-                    if (slot.prompt.n_tokens() < slot.task->n_tokens() && input_tokens[slot.prompt.n_tokens()] == LLAMA_TOKEN_NULL) {
+                    while (slot.prompt.n_tokens() < slot.task->n_tokens() && input_tokens[slot.prompt.n_tokens()] == LLAMA_TOKEN_NULL) {
                         // process the image
                         size_t n_tokens_out = 0;
                         int32_t res = input_tokens.process_chunk(ctx, mctx, slot.prompt.n_tokens(), slot.prompt.tokens.pos_next(), slot.id, n_tokens_out);
diff --git a/tools/server/server-http.cpp b/tools/server/server-http.cpp
index 3466512d0..2262577e1 100644
--- a/tools/server/server-http.cpp
+++ b/tools/server/server-http.cpp
@@ -8,9 +8,11 @@
 #include <string>
 #include <thread>
 
+#ifdef LLAMA_BUILD_WEBUI
 // auto generated files (see README.md for details)
 #include "index.html.gz.hpp"
 #include "loading.html.hpp"
+#endif
 
 //
 // HTTP implementation using cpp-httplib
@@ -110,6 +112,16 @@ bool server_http_context::init(const common_params & params) {
     // set timeouts and change hostname and port
     srv->set_read_timeout (params.timeout_read);
     srv->set_write_timeout(params.timeout_write);
+    srv->set_socket_options([reuse_port = params.reuse_port](socket_t sock) {
+        httplib::set_socket_opt(sock, SOL_SOCKET, SO_REUSEADDR, 1);
+        if (reuse_port) {
+#ifdef SO_REUSEPORT
+            httplib::set_socket_opt(sock, SOL_SOCKET, SO_REUSEPORT, 1);
+#else
+            LOG_WRN("%s: SO_REUSEPORT is not supported\n", __func__);
+#endif
+        }
+    });
 
     if (params.api_keys.size() == 1) {
         auto key = params.api_keys[0];
@@ -181,11 +193,14 @@ bool server_http_context::init(const common_params & params) {
     auto middleware_server_state = [this](const httplib::Request & req, httplib::Response & res) {
         bool ready = is_ready.load();
         if (!ready) {
+#ifdef LLAMA_BUILD_WEBUI
             auto tmp = string_split<std::string>(req.path, '.');
             if (req.path == "/" || tmp.back() == "html") {
                 res.status = 503;
                 res.set_content(reinterpret_cast<const char*>(loading_html), loading_html_len, "text/html; charset=utf-8");
-            } else {
+            } else
+#endif
+            {
                 // no endpoints is allowed to be accessed when the server is not ready
                 // this is to prevent any data races or inconsistent states
                 res.status = 503;
@@ -255,6 +270,7 @@ bool server_http_context::init(const common_params & params) {
                 return 1;
             }
         } else {
+#ifdef LLAMA_BUILD_WEBUI
             // using embedded static index.html
             srv->Get(params.api_prefix + "/", [](const httplib::Request & req, httplib::Response & res) {
                 if (req.get_header_value("Accept-Encoding").find("gzip") == std::string::npos) {
@@ -268,6 +284,7 @@ bool server_http_context::init(const common_params & params) {
                 }
                 return false;
             });
+#endif
         }
     }
     return true;
diff --git a/tools/server/server-task.cpp b/tools/server/server-task.cpp
index 7d543b929..3018ac90f 100644
--- a/tools/server/server-task.cpp
+++ b/tools/server/server-task.cpp
@@ -478,19 +478,17 @@ task_params server_task::params_from_json_cmpl(
     // Parse reasoning budget sampler parameters
     {
         const int32_t budget = json_value(data, "reasoning_budget_tokens", (int32_t) -1);
-        if (budget >= 0) {
-            const auto start_tag = json_value(data, "reasoning_budget_start_tag", std::string());
-            const auto end_tag   = json_value(data, "reasoning_budget_end_tag", std::string());
-            const auto message   = json_value(data, "reasoning_budget_message", std::string());
-            params.sampling.reasoning_budget_tokens = budget;
+        const auto start_tag = json_value(data, "reasoning_budget_start_tag", std::string());
+        const auto end_tag   = json_value(data, "reasoning_budget_end_tag", std::string());
+        const auto message   = json_value(data, "reasoning_budget_message", std::string());
+        params.sampling.reasoning_budget_tokens = budget;
 
-            if (!start_tag.empty()) {
-                params.sampling.reasoning_budget_start = common_tokenize(vocab, start_tag, false, true);
-            }
-            if (!end_tag.empty()) {
-                params.sampling.reasoning_budget_end = common_tokenize(vocab, end_tag, false, true);
-                params.sampling.reasoning_budget_forced = common_tokenize(vocab, message + end_tag, false, true);
-            }
+        if (!start_tag.empty()) {
+            params.sampling.reasoning_budget_start = common_tokenize(vocab, start_tag, false, true);
+        }
+        if (!end_tag.empty()) {
+            params.sampling.reasoning_budget_end = common_tokenize(vocab, end_tag, false, true);
+            params.sampling.reasoning_budget_forced = common_tokenize(vocab, message + end_tag, false, true);
 
             SRV_DBG("reasoning budget: tokens=%d, generation_prompt='%s', start=%zu toks, end=%zu toks, forced=%zu toks\n",
                 budget, params.sampling.generation_prompt.c_str(),
diff --git a/tools/server/server-tools.cpp b/tools/server/server-tools.cpp
new file mode 100644
index 000000000..81e360de4
--- /dev/null
+++ b/tools/server/server-tools.cpp
@@ -0,0 +1,768 @@
+#include "server-tools.h"
+
+#include <sheredom/subprocess.h>
+
+#include <filesystem>
+#include <fstream>
+#include <regex>
+#include <thread>
+#include <chrono>
+#include <atomic>
+#include <cstring>
+#include <climits>
+
+namespace fs = std::filesystem;
+
+//
+// internal helpers
+//
+
+static std::vector<char *> to_cstr_vec(const std::vector<std::string> & v) {
+    std::vector<char *> r;
+    r.reserve(v.size() + 1);
+    for (const auto & s : v) {
+        r.push_back(const_cast<char *>(s.c_str()));
+    }
+    r.push_back(nullptr);
+    return r;
+}
+
+struct run_proc_result {
+    std::string output;
+    int  exit_code = -1;
+    bool timed_out = false;
+};
+
+static run_proc_result run_process(
+        const std::vector<std::string> & args,
+        size_t max_output,
+        int timeout_secs) {
+    run_proc_result res;
+
+    subprocess_s proc;
+    auto argv = to_cstr_vec(args);
+
+    int options = subprocess_option_no_window
+                | subprocess_option_combined_stdout_stderr
+                | subprocess_option_inherit_environment
+                | subprocess_option_search_user_path;
+
+    if (subprocess_create(argv.data(), options, &proc) != 0) {
+        res.output = "failed to spawn process";
+        return res;
+    }
+
+    std::atomic<bool> done{false};
+    std::atomic<bool> timed_out{false};
+
+    std::thread timeout_thread([&]() {
+        auto deadline = std::chrono::steady_clock::now() + std::chrono::seconds(timeout_secs);
+        while (!done.load()) {
+            if (std::chrono::steady_clock::now() >= deadline) {
+                timed_out.store(true);
+                subprocess_terminate(&proc);
+                return;
+            }
+            std::this_thread::sleep_for(std::chrono::milliseconds(100));
+        }
+    });
+
+    FILE * f = subprocess_stdout(&proc);
+    std::string output;
+    bool truncated = false;
+    if (f) {
+        char buf[4096];
+        while (fgets(buf, sizeof(buf), f) != nullptr) {
+            if (!truncated) {
+                size_t len = strlen(buf);
+                if (output.size() + len <= max_output) {
+                    output.append(buf, len);
+                } else {
+                    output.append(buf, max_output - output.size());
+                    truncated = true;
+                }
+            }
+        }
+    }
+
+    done.store(true);
+    if (timeout_thread.joinable()) {
+        timeout_thread.join();
+    }
+
+    subprocess_join(&proc, &res.exit_code);
+    subprocess_destroy(&proc);
+
+    res.output    = output;
+    res.timed_out = timed_out.load();
+    if (truncated) {
+        res.output += "\n[output truncated]";
+    }
+    return res;
+}
+
+json server_tool::to_json() {
+    return {
+        {"display_name", display_name},
+        {"tool", name},
+        {"type", "builtin"},
+        {"permissions", json{
+            {"write", permission_write}
+        }},
+        {"definition", get_definition()},
+    };
+}
+
+//
+// read_file: read a file with optional line range and line-number prefix
+//
+
+static constexpr size_t SERVER_TOOL_READ_FILE_MAX_SIZE = 16 * 1024; // 16 KB
+
+struct server_tool_read_file : server_tool {
+    server_tool_read_file() {
+        name = "read_file";
+        display_name = "Read file";
+        permission_write = false;
+    }
+
+    json get_definition() override {
+        return {
+            {"type", "function"},
+            {"function", {
+                {"name", name},
+                {"description", "Read the contents of a file. Optionally specify a 1-based line range. "
+                                "If append_loc is true, each line is prefixed with its line number (e.g. \"1\u2192 ...\")."},
+                {"parameters", {
+                    {"type", "object"},
+                    {"properties", {
+                        {"path",       {{"type", "string"},  {"description", "Path to the file"}}},
+                        {"start_line", {{"type", "integer"}, {"description", "First line to read, 1-based (default: 1)"}}},
+                        {"end_line",   {{"type", "integer"}, {"description", "Last line to read, 1-based inclusive (default: end of file)"}}},
+                        {"append_loc", {{"type", "boolean"}, {"description", "Prefix each line with its line number"}}},
+                    }},
+                    {"required", json::array({"path"})},
+                }},
+            }},
+        };
+    }
+
+    json invoke(json params) override {
+        std::string path  = params.at("path").get<std::string>();
+        int  start_line   = json_value(params, "start_line", 1);
+        int  end_line     = json_value(params, "end_line",  -1); // -1 = no limit
+        bool append_loc   = json_value(params, "append_loc", false);
+
+        std::error_code ec;
+        uintmax_t file_size = fs::file_size(path, ec);
+        if (ec) {
+            return {{"error", "cannot stat file: " + ec.message()}};
+        }
+        if (file_size > SERVER_TOOL_READ_FILE_MAX_SIZE && end_line == -1) {
+            return {{"error", string_format(
+                "file too large (%zu bytes, max %zu). Use start_line/end_line to read a portion.",
+                (size_t)file_size, SERVER_TOOL_READ_FILE_MAX_SIZE)}};
+        }
+
+        std::ifstream f(path);
+        if (!f) {
+            return {{"error", "failed to open file: " + path}};
+        }
+
+        std::string result;
+        std::string line;
+        int lineno = 0;
+
+        while (std::getline(f, line)) {
+            lineno++;
+            if (lineno < start_line) continue;
+            if (end_line != -1 && lineno > end_line) break;
+
+            std::string out_line;
+            if (append_loc) {
+                out_line = std::to_string(lineno) + "\u2192 " + line + "\n";
+            } else {
+                out_line = line + "\n";
+            }
+
+            if (result.size() + out_line.size() > SERVER_TOOL_READ_FILE_MAX_SIZE) {
+                result += "[output truncated]";
+                break;
+            }
+            result += out_line;
+        }
+
+        return {{"plain_text_response", result}};
+    }
+};
+
+//
+// file_glob_search: find files matching a glob pattern under a base directory
+//
+
+static constexpr size_t SERVER_TOOL_FILE_SEARCH_MAX_RESULTS = 100;
+
+struct server_tool_file_glob_search : server_tool {
+    server_tool_file_glob_search() {
+        name = "file_glob_search";
+        display_name = "File search";
+        permission_write = false;
+    }
+
+    json get_definition() override {
+        return {
+            {"type", "function"},
+            {"function", {
+                {"name", name},
+                {"description", "Recursively search for files matching a glob pattern under a directory."},
+                {"parameters", {
+                    {"type", "object"},
+                    {"properties", {
+                        {"path",    {{"type", "string"}, {"description", "Base directory to search in"}}},
+                        {"include", {{"type", "string"}, {"description", "Glob pattern for files to include (e.g. \"**/*.cpp\"). Default: **"}}},
+                        {"exclude", {{"type", "string"}, {"description", "Glob pattern for files to exclude"}}},
+                    }},
+                    {"required", json::array({"path"})},
+                }},
+            }},
+        };
+    }
+
+    json invoke(json params) override {
+        std::string base    = params.at("path").get<std::string>();
+        std::string include = json_value(params, "include", std::string("**"));
+        std::string exclude = json_value(params, "exclude", std::string(""));
+
+        std::ostringstream output_text;
+        size_t count = 0;
+
+        std::error_code ec;
+        for (const auto & entry : fs::recursive_directory_iterator(base,
+                fs::directory_options::skip_permission_denied, ec)) {
+            if (!entry.is_regular_file()) continue;
+
+            std::string rel = fs::relative(entry.path(), base, ec).string();
+            if (ec) continue;
+            std::replace(rel.begin(), rel.end(), '\\', '/');
+
+            if (!glob_match(include, rel)) continue;
+            if (!exclude.empty() && glob_match(exclude, rel)) continue;
+
+            output_text << entry.path().string() << "\n";
+            if (++count >= SERVER_TOOL_FILE_SEARCH_MAX_RESULTS) {
+                break;
+            }
+        }
+
+        output_text << "\n---\nTotal matches: " << count << "\n";
+
+        return {{"plain_text_response", output_text.str()}};
+    }
+};
+
+//
+// grep_search: search for a regex pattern in files
+//
+
+static constexpr size_t SERVER_TOOL_GREP_SEARCH_MAX_RESULTS = 100;
+
+struct server_tool_grep_search : server_tool {
+    server_tool_grep_search() {
+        name = "grep_search";
+        display_name = "Grep search";
+        permission_write = false;
+    }
+
+    json get_definition() override {
+        return {
+            {"type", "function"},
+            {"function", {
+                {"name", name},
+                {"description", "Search for a regex pattern in files under a path. Returns matching lines."},
+                {"parameters", {
+                    {"type", "object"},
+                    {"properties", {
+                        {"path",                {{"type", "string"},  {"description", "File or directory to search in"}}},
+                        {"pattern",             {{"type", "string"},  {"description", "Regular expression pattern to search for"}}},
+                        {"include",             {{"type", "string"},  {"description", "Glob pattern to filter files (default: **)"}}},
+                        {"exclude",             {{"type", "string"},  {"description", "Glob pattern to exclude files"}}},
+                        {"return_line_numbers", {{"type", "boolean"}, {"description", "If true, include line numbers in results"}}},
+                    }},
+                    {"required", json::array({"path", "pattern"})},
+                }},
+            }},
+        };
+    }
+
+    json invoke(json params) override {
+        std::string path    = params.at("path").get<std::string>();
+        std::string pat_str = params.at("pattern").get<std::string>();
+        std::string include = json_value(params, "include", std::string("**"));
+        std::string exclude = json_value(params, "exclude", std::string(""));
+        bool show_lineno    = json_value(params, "return_line_numbers", false);
+
+        std::regex pattern;
+        try {
+            pattern = std::regex(pat_str);
+        } catch (const std::regex_error & e) {
+            return {{"error", std::string("invalid regex: ") + e.what()}};
+        }
+
+        std::ostringstream output_text;
+        size_t total = 0;
+
+        auto search_file = [&](const fs::path & fpath) {
+            std::ifstream f(fpath);
+            if (!f) return;
+            std::string line;
+            int lineno = 0;
+            while (std::getline(f, line) && total < SERVER_TOOL_GREP_SEARCH_MAX_RESULTS) {
+                lineno++;
+                if (std::regex_search(line, pattern)) {
+                    output_text << fpath.string() << ":";
+                    if (show_lineno) {
+                        output_text << lineno << ":";
+                    }
+                    output_text << line << "\n";
+                    total++;
+                }
+            }
+        };
+
+        std::error_code ec;
+        if (fs::is_regular_file(path, ec)) {
+            search_file(path);
+        } else if (fs::is_directory(path, ec)) {
+            for (const auto & entry : fs::recursive_directory_iterator(path,
+                    fs::directory_options::skip_permission_denied, ec)) {
+                if (!entry.is_regular_file()) continue;
+                if (total >= SERVER_TOOL_GREP_SEARCH_MAX_RESULTS) break;
+
+                std::string rel = fs::relative(entry.path(), path, ec).string();
+                if (ec) continue;
+                std::replace(rel.begin(), rel.end(), '\\', '/');
+
+                if (!glob_match(include, rel)) continue;
+                if (!exclude.empty() && glob_match(exclude, rel)) continue;
+
+                search_file(entry.path());
+            }
+        } else {
+            return {{"error", "path does not exist: " + path}};
+        }
+
+        output_text << "\n\n---\nTotal matches: " << total << "\n";
+
+        return {{"plain_text_response", output_text.str()}};
+    }
+};
+
+//
+// exec_shell_command: run an arbitrary shell command
+//
+
+static constexpr size_t SERVER_TOOL_EXEC_SHELL_COMMAND_MAX_OUTPUT_SIZE = 16 * 1024; // 16 KB
+static constexpr int    SERVER_TOOL_EXEC_SHELL_COMMAND_MAX_TIMEOUT     = 60;        // seconds
+
+struct server_tool_exec_shell_command : server_tool {
+    server_tool_exec_shell_command() {
+        name = "exec_shell_command";
+        display_name = "Execute shell command";
+        permission_write = true;
+    }
+
+    json get_definition() override {
+        return {
+            {"type", "function"},
+            {"function", {
+                {"name", name},
+                {"description", "Execute a shell command and return its output (stdout and stderr combined)."},
+                {"parameters", {
+                    {"type", "object"},
+                    {"properties", {
+                        {"command",         {{"type", "string"},  {"description", "Shell command to execute"}}},
+                        {"timeout",         {{"type", "integer"}, {"description", string_format("Timeout in seconds (default 10, max %d)", SERVER_TOOL_EXEC_SHELL_COMMAND_MAX_TIMEOUT)}}},
+                        {"max_output_size", {{"type", "integer"}, {"description", string_format("Maximum output size in bytes (default %zu)", SERVER_TOOL_EXEC_SHELL_COMMAND_MAX_OUTPUT_SIZE)}}},
+                    }},
+                    {"required", json::array({"command"})},
+                }},
+            }},
+        };
+    }
+
+    json invoke(json params) override {
+        std::string command   = params.at("command").get<std::string>();
+        int    timeout        = json_value(params, "timeout",         10);
+        size_t max_output     = (size_t) json_value(params, "max_output_size", (int) SERVER_TOOL_EXEC_SHELL_COMMAND_MAX_OUTPUT_SIZE);
+
+        timeout    = std::min(timeout,    SERVER_TOOL_EXEC_SHELL_COMMAND_MAX_TIMEOUT);
+        max_output = std::min(max_output, SERVER_TOOL_EXEC_SHELL_COMMAND_MAX_OUTPUT_SIZE);
+
+#ifdef _WIN32
+        std::vector<std::string> args = {"cmd", "/c", command};
+#else
+        std::vector<std::string> args = {"sh", "-c", command};
+#endif
+
+        auto res = run_process(args, max_output, timeout);
+
+        std::string text_output = res.output;
+        text_output += string_format("\n[exit code: %d]", res.exit_code);
+        if (res.timed_out) {
+            text_output += " [exit due to timed out]";
+        }
+
+        return {{"plain_text_response", text_output}};
+    }
+};
+
+//
+// write_file: create or overwrite a file
+//
+
+struct server_tool_write_file : server_tool {
+    server_tool_write_file() {
+        name = "write_file";
+        display_name = "Write file";
+        permission_write = true;
+    }
+
+    json get_definition() override {
+        return {
+            {"type", "function"},
+            {"function", {
+                {"name", name},
+                {"description", "Write content to a file, creating it (including parent directories) if it does not exist. May use with edit_file for more complex edits."},
+                {"parameters", {
+                    {"type", "object"},
+                    {"properties", {
+                        {"path",    {{"type", "string"}, {"description", "Path of the file to write"}}},
+                        {"content", {{"type", "string"}, {"description", "Content to write"}}},
+                    }},
+                    {"required", json::array({"path", "content"})},
+                }},
+            }},
+        };
+    }
+
+    json invoke(json params) override {
+        std::string path    = params.at("path").get<std::string>();
+        std::string content = params.at("content").get<std::string>();
+
+        std::error_code ec;
+        fs::path fpath(path);
+        if (fpath.has_parent_path()) {
+            fs::create_directories(fpath.parent_path(), ec);
+            if (ec) {
+                return {{"error", "failed to create directories: " + ec.message()}};
+            }
+        }
+
+        std::ofstream f(path, std::ios::binary);
+        if (!f) {
+            return {{"error", "failed to open file for writing: " + path}};
+        }
+        f << content;
+        if (!f) {
+            return {{"error", "failed to write file: " + path}};
+        }
+
+        return {{"result", "file written successfully"}, {"path", path}, {"bytes", content.size()}};
+    }
+};
+
+//
+// edit_file: edit file content via line-based changes
+//
+
+struct server_tool_edit_file : server_tool {
+    server_tool_edit_file() {
+        name = "edit_file";
+        display_name = "Edit file";
+        permission_write = true;
+    }
+
+    json get_definition() override {
+        return {
+            {"type", "function"},
+            {"function", {
+                {"name", name},
+                {"description",
+                    "Edit a file by applying a list of line-based changes. "
+                    "Each change targets a 1-based inclusive line range and has a mode: "
+                    "\"replace\" (replace lines with content), "
+                    "\"delete\" (remove lines, content must be empty string), "
+                    "\"append\" (insert content after line_end). "
+                    "Set line_start to -1 to target the end of file (line_end is ignored in that case). "
+                    "Changes must not overlap. They are applied in reverse line order automatically."},
+                {"parameters", {
+                    {"type", "object"},
+                    {"properties", {
+                        {"path",    {{"type", "string"}, {"description", "Path to the file to edit"}}},
+                        {"changes", {
+                            {"type", "array"},
+                            {"description", "List of changes to apply"},
+                            {"items", {
+                                {"type", "object"},
+                                {"properties", {
+                                    {"mode",       {{"type", "string"},  {"description", "\"replace\", \"delete\", or \"append\""}}},
+                                    {"line_start", {{"type", "integer"}, {"description", "First line of the range (1-based); use -1 for end of file"}}},
+                                    {"line_end",   {{"type", "integer"}, {"description", "Last line of the range (1-based, inclusive); ignored when line_start is -1"}}},
+                                    {"content",    {{"type", "string"},  {"description", "Content to insert; must be empty string for delete mode"}}},
+                                }},
+                                {"required", json::array({"mode", "line_start", "line_end", "content"})},
+                            }},
+                        }},
+                    }},
+                    {"required", json::array({"path", "changes"})},
+                }},
+            }},
+        };
+    }
+
+    json invoke(json params) override {
+        std::string path = params.at("path").get<std::string>();
+        const json & changes = params.at("changes");
+
+        if (!changes.is_array()) {
+            return {{"error", "\"changes\" must be an array"}};
+        }
+
+        // read file into lines
+        std::ifstream fin(path);
+        if (!fin) {
+            return {{"error", "failed to open file: " + path}};
+        }
+        std::vector<std::string> lines;
+        {
+            std::string line;
+            while (std::getline(fin, line)) {
+                lines.push_back(line);
+            }
+        }
+        fin.close();
+
+        // validate and collect changes, then sort descending by line_start
+        struct change_entry {
+            std::string mode;
+            int line_start; // 1-based
+            int line_end;   // 1-based inclusive
+            std::string content;
+        };
+        std::vector<change_entry> entries;
+        entries.reserve(changes.size());
+
+        for (const auto & ch : changes) {
+            change_entry e;
+            e.mode       = ch.at("mode").get<std::string>();
+            e.line_start = ch.at("line_start").get<int>();
+            e.line_end   = ch.at("line_end").get<int>();
+            e.content    = ch.at("content").get<std::string>();
+
+            if (e.mode != "replace" && e.mode != "delete" && e.mode != "append") {
+                return {{"error", "invalid mode \"" + e.mode + "\"; must be replace, delete, or append"}};
+            }
+            if (e.mode == "delete" && !e.content.empty()) {
+                return {{"error", "content must be empty string for delete mode"}};
+            }
+            int n = (int) lines.size();
+            if (e.line_start == -1) {
+                // -1 means end of file; line_end is ignored — normalize to point past last line
+                e.line_start = n + 1;
+                e.line_end   = n + 1;
+            } else {
+                if (e.line_start < 1 || e.line_end < e.line_start) {
+                    return {{"error", string_format("invalid line range [%d, %d]", e.line_start, e.line_end)}};
+                }
+                if (e.line_end > n) {
+                    return {{"error", string_format("line_end %d exceeds file length %d", e.line_end, n)}};
+                }
+            }
+            entries.push_back(std::move(e));
+        }
+
+        // sort descending so earlier-indexed changes don't shift later ones
+        std::sort(entries.begin(), entries.end(), [](const change_entry & a, const change_entry & b) {
+            return a.line_start > b.line_start;
+        });
+
+        // apply changes (0-based indices internally)
+        for (const auto & e : entries) {
+            int idx_start = e.line_start - 1; // 0-based
+            int idx_end   = e.line_end   - 1; // 0-based inclusive
+
+            // split content into lines (preserve trailing newline awareness)
+            std::vector<std::string> new_lines;
+            if (!e.content.empty()) {
+                std::istringstream ss(e.content);
+                std::string ln;
+                while (std::getline(ss, ln)) {
+                    new_lines.push_back(ln);
+                }
+                // if content ends with \n, getline consumed it — no extra empty line needed
+                // if content does NOT end with \n, last line is still captured correctly
+            }
+
+            if (e.mode == "replace") {
+                // erase [idx_start, idx_end] and insert new_lines
+                lines.erase(lines.begin() + idx_start, lines.begin() + idx_end + 1);
+                lines.insert(lines.begin() + idx_start, new_lines.begin(), new_lines.end());
+            } else if (e.mode == "delete") {
+                lines.erase(lines.begin() + idx_start, lines.begin() + idx_end + 1);
+            } else { // append
+                // idx_end + 1 may equal lines.size() when line_start == -1 (end of file)
+                lines.insert(lines.begin() + idx_end + 1, new_lines.begin(), new_lines.end());
+            }
+        }
+
+        // write file back
+        std::ofstream fout(path, std::ios::binary);
+        if (!fout) {
+            return {{"error", "failed to open file for writing: " + path}};
+        }
+        for (size_t i = 0; i < lines.size(); i++) {
+            fout << lines[i];
+            if (i + 1 < lines.size()) {
+                fout << "\n";
+            }
+        }
+        if (!lines.empty()) {
+            fout << "\n";
+        }
+        if (!fout) {
+            return {{"error", "failed to write file: " + path}};
+        }
+
+        return {{"result", "file edited successfully"}, {"path", path}, {"lines", (int) lines.size()}};
+    }
+};
+
+//
+// apply_diff: apply a unified diff via git apply
+//
+
+struct server_tool_apply_diff : server_tool {
+    server_tool_apply_diff() {
+        name = "apply_diff";
+        display_name = "Apply diff";
+        permission_write = true;
+    }
+
+    json get_definition() override {
+        return {
+            {"type", "function"},
+            {"function", {
+                {"name", name},
+                {"description", "Apply a unified diff to edit one or more files using git apply. Use this instead of edit_file when the changes are complex."},
+                {"parameters", {
+                    {"type", "object"},
+                    {"properties", {
+                        {"diff", {{"type", "string"}, {"description", "Unified diff content in git diff format"}}},
+                    }},
+                    {"required", json::array({"diff"})},
+                }},
+            }},
+        };
+    }
+
+    json invoke(json params) override {
+        std::string diff = params.at("diff").get<std::string>();
+
+        // write diff to a temporary file
+        static std::atomic<int> counter{0};
+        std::string tmp_path = (fs::temp_directory_path() /
+            ("llama_patch_" + std::to_string(++counter) + ".patch")).string();
+
+        {
+            std::ofstream f(tmp_path, std::ios::binary);
+            if (!f) {
+                return {{"error", "failed to create temp patch file"}};
+            }
+            f << diff;
+        }
+
+        auto res = run_process({"git", "apply", tmp_path}, 4096, 10);
+
+        std::error_code ec;
+        fs::remove(tmp_path, ec);
+
+        if (res.exit_code != 0) {
+            return {{"error", "git apply failed (exit " + std::to_string(res.exit_code) + "): " + res.output}};
+        }
+        return {{"result", "patch applied successfully"}};
+    }
+};
+
+//
+// public API
+//
+
+static std::vector<std::unique_ptr<server_tool>> build_tools() {
+    std::vector<std::unique_ptr<server_tool>> tools;
+    tools.push_back(std::make_unique<server_tool_read_file>());
+    tools.push_back(std::make_unique<server_tool_file_glob_search>());
+    tools.push_back(std::make_unique<server_tool_grep_search>());
+    tools.push_back(std::make_unique<server_tool_exec_shell_command>());
+    tools.push_back(std::make_unique<server_tool_write_file>());
+    tools.push_back(std::make_unique<server_tool_edit_file>());
+    tools.push_back(std::make_unique<server_tool_apply_diff>());
+    return tools;
+}
+
+void server_tools::setup(const std::vector<std::string> & enabled_tools) {
+    if (!enabled_tools.empty()) {
+        std::unordered_set<std::string> enabled_set(enabled_tools.begin(), enabled_tools.end());
+        auto all_tools = build_tools();
+
+        tools.clear();
+        for (auto & t : all_tools) {
+            if (enabled_set.count(t->name) > 0 || enabled_set.count("all") > 0) {
+                tools.push_back(std::move(t));
+            }
+        }
+    }
+
+    handle_get = [this](const server_http_req &) -> server_http_res_ptr {
+        auto res = std::make_unique<server_http_res>();
+        try {
+            json result = json::array();
+            for (const auto & t : tools) {
+                result.push_back(t->to_json());
+            }
+            res->data = safe_json_to_str(result);
+        } catch (const std::exception & e) {
+            SRV_ERR("got exception: %s\n", e.what());
+            res->status = 500;
+            res->data   = safe_json_to_str(format_error_response(e.what(), ERROR_TYPE_SERVER));
+        }
+        return res;
+    };
+
+    handle_post = [this](const server_http_req & req) -> server_http_res_ptr {
+        auto res = std::make_unique<server_http_res>();
+        try {
+            json body = json::parse(req.body);
+            std::string tool_name = body.at("tool").get<std::string>();
+            json params = body.value("params", json::object());
+            json result = invoke(tool_name, params);
+            res->data   = safe_json_to_str(result);
+        } catch (const json::exception & e) {
+            res->status = 400;
+            res->data   = safe_json_to_str(format_error_response(e.what(), ERROR_TYPE_INVALID_REQUEST));
+        } catch (const std::exception & e) {
+            SRV_ERR("got exception: %s\n", e.what());
+            res->status = 500;
+            res->data   = safe_json_to_str(format_error_response(e.what(), ERROR_TYPE_SERVER));
+        }
+        return res;
+    };
+}
+
+json server_tools::invoke(const std::string & name, const json & params) {
+    for (auto & t : tools) {
+        if (t->name == name) {
+            return t->invoke(params);
+        }
+    }
+    return {{"error", "unknown tool: " + name}};
+}
diff --git a/tools/server/server-tools.h b/tools/server/server-tools.h
new file mode 100644
index 000000000..444ef5f80
--- /dev/null
+++ b/tools/server/server-tools.h
@@ -0,0 +1,26 @@
+#pragma once
+
+#include "server-common.h"
+#include "server-http.h"
+
+struct server_tool {
+    std::string name;
+    std::string display_name;
+    bool permission_write = false;
+
+    virtual ~server_tool() = default;
+    virtual json get_definition() = 0;
+    virtual json invoke(json params) = 0;
+
+    json to_json();
+};
+
+struct server_tools {
+    std::vector<std::unique_ptr<server_tool>> tools;
+
+    void setup(const std::vector<std::string> & enabled_tools);
+    json invoke(const std::string & name, const json & params);
+
+    server_http_context::handler_t handle_get;
+    server_http_context::handler_t handle_post;
+};
diff --git a/tools/server/server.cpp b/tools/server/server.cpp
index 73e3aa44e..ef54a46b1 100644
--- a/tools/server/server.cpp
+++ b/tools/server/server.cpp
@@ -2,6 +2,7 @@
 #include "server-http.h"
 #include "server-models.h"
 #include "server-cors-proxy.h"
+#include "server-tools.h"
 
 #include "arg.h"
 #include "common.h"
@@ -124,6 +125,7 @@ int main(int argc, char ** argv) {
 
     // register API routes
     server_routes routes(params, ctx_server);
+    server_tools tools;
 
     bool is_router_server = params.model.path.empty();
     std::optional<server_models_routes> models_routes{};
@@ -211,6 +213,16 @@ int main(int argc, char ** argv) {
         ctx_http.get ("/cors-proxy",      ex_wrapper(proxy_handler_get));
         ctx_http.post("/cors-proxy",      ex_wrapper(proxy_handler_post));
     }
+    // EXPERIMENTAL built-in tools
+    if (!params.server_tools.empty()) {
+        tools.setup(params.server_tools);
+        SRV_WRN("%s", "-----------------\n");
+        SRV_WRN("%s", "Built-in tools are enabled, do not expose server to untrusted environments\n");
+        SRV_WRN("%s", "This feature is EXPERIMENTAL and may be changed in the future\n");
+        SRV_WRN("%s", "-----------------\n");
+        ctx_http.get ("/tools",           ex_wrapper(tools.handle_get));
+        ctx_http.post("/tools",           ex_wrapper(tools.handle_post));
+    }
 
     //
     // Start the server
diff --git a/tools/server/tests/utils.py b/tools/server/tests/utils.py
index c6fe11261..8f62210bc 100644
--- a/tools/server/tests/utils.py
+++ b/tools/server/tests/utils.py
@@ -288,7 +288,15 @@ class ServerProcess:
             server_instances.remove(self)
         if self.process:
             print(f"Stopping server with pid={self.process.pid}")
-            self.process.kill()
+            self.process.terminate()
+            try:
+                self.process.wait(timeout=5)
+            except subprocess.TimeoutExpired:
+                print(f"Server pid={self.process.pid} did not terminate in time, killing")
+                self.process.kill()
+                self.process.wait(timeout=5)
+            except Exception as e:
+                print(f"Error waiting for server: {e}")
             self.process = None
 
     def make_request(
diff --git a/tools/server/webui/src/lib/actions/fade-in-view.svelte.ts b/tools/server/webui/src/lib/actions/fade-in-view.svelte.ts
new file mode 100644
index 000000000..5c726b7e4
--- /dev/null
+++ b/tools/server/webui/src/lib/actions/fade-in-view.svelte.ts
@@ -0,0 +1,54 @@
+/**
+ * Svelte action that fades in an element when it enters the viewport.
+ * Uses IntersectionObserver for efficient viewport detection.
+ *
+ * If skipIfVisible is set and the element is already visible in the viewport
+ * when the action attaches (e.g. a markdown block promoted from unstable
+ * during streaming), the fade is skipped entirely to avoid a flash.
+ */
+export function fadeInView(
+	node: HTMLElement,
+	options: { duration?: number; y?: number; skipIfVisible?: boolean } = {}
+) {
+	const { duration = 300, y = 0, skipIfVisible = false } = options;
+
+	if (skipIfVisible) {
+		const rect = node.getBoundingClientRect();
+		const isAlreadyVisible =
+			rect.top < window.innerHeight &&
+			rect.bottom > 0 &&
+			rect.left < window.innerWidth &&
+			rect.right > 0;
+
+		if (isAlreadyVisible) {
+			return;
+		}
+	}
+
+	node.style.opacity = '0';
+	node.style.transform = `translateY(${y}px)`;
+	node.style.transition = `opacity ${duration}ms ease-out, transform ${duration}ms ease-out`;
+
+	$effect(() => {
+		const observer = new IntersectionObserver(
+			(entries) => {
+				for (const entry of entries) {
+					if (entry.isIntersecting) {
+						requestAnimationFrame(() => {
+							node.style.opacity = '1';
+							node.style.transform = 'translateY(0)';
+						});
+						observer.disconnect();
+					}
+				}
+			},
+			{ threshold: 0.05 }
+		);
+
+		observer.observe(node);
+
+		return () => {
+			observer.disconnect();
+		};
+	});
+}
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatForm/ChatFormActions/ChatFormActions.svelte b/tools/server/webui/src/lib/components/app/chat/ChatForm/ChatFormActions/ChatFormActions.svelte
index b51dd682e..95c3c5da1 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatForm/ChatFormActions/ChatFormActions.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatForm/ChatFormActions/ChatFormActions.svelte
@@ -10,9 +10,9 @@
 		ModelsSelector,
 		ModelsSelectorSheet
 	} from '$lib/components/app';
-	import { DialogChatSettings } from '$lib/components/app/dialogs';
 	import { SETTINGS_SECTION_TITLES } from '$lib/constants';
 	import { mcpStore } from '$lib/stores/mcp.svelte';
+	import { getChatSettingsDialogContext } from '$lib/contexts';
 	import { FileTypeCategory } from '$lib/enums';
 	import { getFileTypeCategory } from '$lib/utils';
 	import { config } from '$lib/stores/settings.svelte';
@@ -169,7 +169,7 @@
 		selectorModelRef?.open();
 	}
 
-	let showChatSettingsDialogWithMcpSection = $state(false);
+	const chatSettingsDialog = getChatSettingsDialogContext();
 
 	let hasMcpPromptsSupport = $derived.by(() => {
 		const perChatOverrides = conversationsStore.getAllMcpServerOverrides();
@@ -197,7 +197,7 @@
 				{onSystemPromptClick}
 				{onMcpPromptClick}
 				{onMcpResourcesClick}
-				onMcpSettingsClick={() => (showChatSettingsDialogWithMcpSection = true)}
+				onMcpSettingsClick={() => chatSettingsDialog.open(SETTINGS_SECTION_TITLES.MCP)}
 			/>
 		{:else}
 			<ChatFormActionAttachmentsDropdown
@@ -210,13 +210,13 @@
 				{onSystemPromptClick}
 				{onMcpPromptClick}
 				{onMcpResourcesClick}
-				onMcpSettingsClick={() => (showChatSettingsDialogWithMcpSection = true)}
+				onMcpSettingsClick={() => chatSettingsDialog.open(SETTINGS_SECTION_TITLES.MCP)}
 			/>
 		{/if}
 
 		<McpServersSelector
 			{disabled}
-			onSettingsClick={() => (showChatSettingsDialogWithMcpSection = true)}
+			onSettingsClick={() => chatSettingsDialog.open(SETTINGS_SECTION_TITLES.MCP)}
 		/>
 	</div>
 
@@ -265,9 +265,3 @@
 		/>
 	{/if}
 </div>
-
-<DialogChatSettings
-	open={showChatSettingsDialogWithMcpSection}
-	onOpenChange={(open) => (showChatSettingsDialogWithMcpSection = open)}
-	initialSection={SETTINGS_SECTION_TITLES.MCP}
-/>
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessage.svelte b/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessage.svelte
index 52a935510..b48c71aec 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessage.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessage.svelte
@@ -180,6 +180,10 @@
 		chatActions.continueAssistantMessage(message);
 	}
 
+	function handleForkConversation(options: { name: string; includeAttachments: boolean }) {
+		chatActions.forkConversation(message, options);
+	}
+
 	function handleNavigateToSibling(siblingId: string) {
 		chatActions.navigateToSibling(siblingId);
 	}
@@ -285,6 +289,7 @@
 		onCopy={handleCopy}
 		onDelete={handleDelete}
 		onEdit={handleEdit}
+		onForkConversation={handleForkConversation}
 		onNavigateToSibling={handleNavigateToSibling}
 		onShowDeleteDialogChange={handleShowDeleteDialogChange}
 		{showDeleteDialog}
@@ -303,6 +308,7 @@
 		onCopy={handleCopy}
 		onDelete={handleDelete}
 		onEdit={handleEdit}
+		onForkConversation={handleForkConversation}
 		onNavigateToSibling={handleNavigateToSibling}
 		onRegenerate={handleRegenerate}
 		onShowDeleteDialogChange={handleShowDeleteDialogChange}
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageActions.svelte b/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageActions.svelte
index 97b34e92c..fec1a506e 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageActions.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageActions.svelte
@@ -1,12 +1,16 @@
 <script lang="ts">
-	import { Edit, Copy, RefreshCw, Trash2, ArrowRight } from '@lucide/svelte';
+	import { Edit, Copy, RefreshCw, Trash2, ArrowRight, GitBranch } from '@lucide/svelte';
 	import {
 		ActionIcon,
 		ChatMessageBranchingControls,
 		DialogConfirmation
 	} from '$lib/components/app';
 	import { Switch } from '$lib/components/ui/switch';
+	import { Checkbox } from '$lib/components/ui/checkbox';
+	import Input from '$lib/components/ui/input/input.svelte';
+	import Label from '$lib/components/ui/label/label.svelte';
 	import { MessageRole } from '$lib/enums';
+	import { activeConversation } from '$lib/stores/conversations.svelte';
 
 	interface Props {
 		role: MessageRole.USER | MessageRole.ASSISTANT;
@@ -24,6 +28,7 @@
 		onEdit?: () => void;
 		onRegenerate?: () => void;
 		onContinue?: () => void;
+		onForkConversation?: (options: { name: string; includeAttachments: boolean }) => void;
 		onDelete: () => void;
 		onConfirmDelete: () => void;
 		onNavigateToSibling?: (siblingId: string) => void;
@@ -42,6 +47,7 @@
 		onConfirmDelete,
 		onContinue,
 		onDelete,
+		onForkConversation,
 		onNavigateToSibling,
 		onShowDeleteDialogChange,
 		onRegenerate,
@@ -53,10 +59,27 @@
 		onRawOutputToggle
 	}: Props = $props();
 
+	let showForkDialog = $state(false);
+	let forkName = $state('');
+	let forkIncludeAttachments = $state(true);
+
 	function handleConfirmDelete() {
 		onConfirmDelete();
 		onShowDeleteDialogChange(false);
 	}
+
+	function handleOpenForkDialog() {
+		const conv = activeConversation();
+
+		forkName = `Fork of ${conv?.name ?? 'Conversation'}`;
+		forkIncludeAttachments = true;
+		showForkDialog = true;
+	}
+
+	function handleConfirmFork() {
+		onForkConversation?.({ name: forkName.trim(), includeAttachments: forkIncludeAttachments });
+		showForkDialog = false;
+	}
 </script>
 
 <div class="relative {justify === 'start' ? 'mt-2' : ''} flex h-6 items-center justify-between">
@@ -86,6 +109,10 @@
 				<ActionIcon icon={ArrowRight} tooltip="Continue" onclick={onContinue} />
 			{/if}
 
+			{#if onForkConversation}
+				<ActionIcon icon={GitBranch} tooltip="Fork conversation" onclick={handleOpenForkDialog} />
+			{/if}
+
 			<ActionIcon icon={Trash2} tooltip="Delete" onclick={onDelete} />
 		</div>
 	</div>
@@ -116,3 +143,42 @@
 	onConfirm={handleConfirmDelete}
 	onCancel={() => onShowDeleteDialogChange(false)}
 />
+
+<DialogConfirmation
+	bind:open={showForkDialog}
+	title="Fork Conversation"
+	description="Create a new conversation branching from this message."
+	confirmText="Fork"
+	cancelText="Cancel"
+	icon={GitBranch}
+	onConfirm={handleConfirmFork}
+	onCancel={() => (showForkDialog = false)}
+>
+	<div class="flex flex-col gap-4 py-2">
+		<div class="flex flex-col gap-2">
+			<Label for="fork-name">Title</Label>
+
+			<Input
+				id="fork-name"
+				class="text-foreground"
+				placeholder="Enter fork name"
+				type="text"
+				bind:value={forkName}
+			/>
+		</div>
+
+		<div class="flex items-center gap-2">
+			<Checkbox
+				id="fork-attachments"
+				checked={forkIncludeAttachments}
+				onCheckedChange={(checked) => {
+					forkIncludeAttachments = checked === true;
+				}}
+			/>
+
+			<Label for="fork-attachments" class="cursor-pointer text-sm font-normal">
+				Include all attachments
+			</Label>
+		</div>
+	</div>
+</DialogConfirmation>
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageAssistant.svelte b/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageAssistant.svelte
index 553c3fd94..97aee0300 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageAssistant.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageAssistant.svelte
@@ -3,14 +3,12 @@
 		ChatMessageAgenticContent,
 		ChatMessageActions,
 		ChatMessageStatistics,
-		MarkdownContent,
 		ModelBadge,
 		ModelsSelector
 	} from '$lib/components/app';
 	import { getMessageEditContext } from '$lib/contexts';
 	import { useProcessingState } from '$lib/hooks/use-processing-state.svelte';
 	import { isLoading, isChatStreaming } from '$lib/stores/chat.svelte';
-	import { agenticStreamingToolCall } from '$lib/stores/agentic.svelte';
 	import { autoResizeTextarea, copyToClipboard, isIMEComposing } from '$lib/utils';
 	import { tick } from 'svelte';
 	import { fade } from 'svelte/transition';
@@ -41,6 +39,7 @@
 		onContinue?: () => void;
 		onDelete: () => void;
 		onEdit?: () => void;
+		onForkConversation?: (options: { name: string; includeAttachments: boolean }) => void;
 		onNavigateToSibling?: (siblingId: string) => void;
 		onRegenerate: (modelOverride?: string) => void;
 		onShowDeleteDialogChange: (show: boolean) => void;
@@ -60,6 +59,7 @@
 		onCopy,
 		onDelete,
 		onEdit,
+		onForkConversation,
 		onNavigateToSibling,
 		onRegenerate,
 		onShowDeleteDialogChange,
@@ -87,13 +87,7 @@
 	const hasAgenticMarkers = $derived(
 		messageContent?.includes(AGENTIC_TAGS.TOOL_CALL_START) ?? false
 	);
-	const hasStreamingToolCall = $derived(
-		isChatStreaming() && agenticStreamingToolCall(message.convId) !== null
-	);
 	const hasReasoningMarkers = $derived(messageContent?.includes(REASONING_TAGS.START) ?? false);
-	const isStructuredContent = $derived(
-		hasAgenticMarkers || hasReasoningMarkers || hasStreamingToolCall
-	);
 	const processingState = useProcessingState();
 
 	let currentConfig = $derived(config());
@@ -256,15 +250,13 @@
 	{:else if message.role === MessageRole.ASSISTANT}
 		{#if showRawOutput}
 			<pre class="raw-output">{messageContent || ''}</pre>
-		{:else if isStructuredContent}
+		{:else}
 			<ChatMessageAgenticContent
 				content={messageContent || ''}
 				isStreaming={isChatStreaming()}
 				highlightTurns={highlightAgenticTurns}
 				{message}
 			/>
-		{:else}
-			<MarkdownContent content={messageContent || ''} attachments={message.extra} />
 		{/if}
 	{:else}
 		<div class="text-sm whitespace-pre-wrap">
@@ -355,6 +347,7 @@
 			onContinue={currentConfig.enableContinueGeneration && !hasReasoningMarkers
 				? onContinue
 				: undefined}
+			{onForkConversation}
 			{onDelete}
 			{onConfirmDelete}
 			{onNavigateToSibling}
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageUser.svelte b/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageUser.svelte
index 9b9d9a3e9..f2f6288d6 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageUser.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessageUser.svelte
@@ -21,6 +21,7 @@
 		onEdit: () => void;
 		onDelete: () => void;
 		onConfirmDelete: () => void;
+		onForkConversation?: (options: { name: string; includeAttachments: boolean }) => void;
 		onShowDeleteDialogChange: (show: boolean) => void;
 		onNavigateToSibling?: (siblingId: string) => void;
 		onCopy: () => void;
@@ -35,6 +36,7 @@
 		onEdit,
 		onDelete,
 		onConfirmDelete,
+		onForkConversation,
 		onShowDeleteDialogChange,
 		onNavigateToSibling,
 		onCopy
@@ -114,6 +116,7 @@
 					{onCopy}
 					{onDelete}
 					{onEdit}
+					{onForkConversation}
 					{onNavigateToSibling}
 					{onShowDeleteDialogChange}
 					{siblingInfo}
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessages.svelte b/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessages.svelte
index 23143c955..5ff53644c 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessages.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatMessages/ChatMessages.svelte
@@ -1,4 +1,5 @@
 <script lang="ts">
+	import { fadeInView } from '$lib/actions/fade-in-view.svelte';
 	import { ChatMessage } from '$lib/components/app';
 	import { setChatActionsContext } from '$lib/contexts';
 	import { MessageRole } from '$lib/enums';
@@ -78,6 +79,13 @@
 			onUserAction?.();
 			await chatStore.continueAssistantMessage(message.id);
 			refreshAllMessages();
+		},
+
+		forkConversation: async (
+			message: DatabaseMessage,
+			options: { name: string; includeAttachments: boolean }
+		) => {
+			await conversationsStore.forkConversation(message.id, options);
 		}
 	});
 
@@ -140,13 +148,18 @@
 	});
 </script>
 
-<div class="flex h-full flex-col space-y-10 pt-24 {className}" style="height: auto; ">
+<div
+	class="flex h-full flex-col space-y-10 pt-24 {className}"
+	style="height: auto; min-height: calc(100dvh - 14rem);"
+>
 	{#each displayMessages as { message, isLastAssistantMessage, siblingInfo } (message.id)}
-		<ChatMessage
-			class="mx-auto w-full max-w-[48rem]"
-			{message}
-			{isLastAssistantMessage}
-			{siblingInfo}
-		/>
+		<div use:fadeInView>
+			<ChatMessage
+				class="mx-auto w-full max-w-[48rem]"
+				{message}
+				{isLastAssistantMessage}
+				{siblingInfo}
+			/>
+		</div>
 	{/each}
 </div>
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatScreen/ChatScreen.svelte b/tools/server/webui/src/lib/components/app/chat/ChatScreen/ChatScreen.svelte
index 984cd7efa..c55e7ca00 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatScreen/ChatScreen.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatScreen/ChatScreen.svelte
@@ -12,7 +12,6 @@
 	} from '$lib/components/app';
 	import * as Alert from '$lib/components/ui/alert';
 	import * as AlertDialog from '$lib/components/ui/alert-dialog';
-	import { INITIAL_SCROLL_DELAY } from '$lib/constants';
 	import { KeyboardKey } from '$lib/enums';
 	import { createAutoScrollController } from '$lib/hooks/use-auto-scroll.svelte';
 	import {
@@ -48,7 +47,7 @@
 	let showFileErrorDialog = $state(false);
 	let uploadedFiles = $state<ChatUploadedFile[]>([]);
 
-	const autoScroll = createAutoScrollController();
+	const autoScroll = createAutoScrollController({ isColumnReverse: true });
 
 	let fileErrorData = $state<{
 		generallyUnsupported: File[];
@@ -310,13 +309,15 @@
 
 	afterNavigate(() => {
 		if (!disableAutoScroll) {
-			setTimeout(() => autoScroll.scrollToBottom('instant'), INITIAL_SCROLL_DELAY);
+			autoScroll.enable();
 		}
 	});
 
 	onMount(() => {
+		autoScroll.startObserving();
+
 		if (!disableAutoScroll) {
-			setTimeout(() => autoScroll.scrollToBottom('instant'), INITIAL_SCROLL_DELAY);
+			autoScroll.enable();
 		}
 
 		const pendingDraft = chatStore.consumePendingDraft();
@@ -333,10 +334,6 @@
 	$effect(() => {
 		autoScroll.setDisabled(disableAutoScroll);
 	});
-
-	$effect(() => {
-		autoScroll.updateInterval(isCurrentConversationLoading);
-	});
 </script>
 
 {#if isDragOver}
@@ -351,7 +348,7 @@
 	<div
 		bind:this={chatScrollContainer}
 		aria-label="Chat interface with file drop zone"
-		class="flex h-full flex-col overflow-y-auto px-4 md:px-6"
+		class="flex h-full flex-col-reverse overflow-y-auto px-4 md:px-6"
 		ondragenter={handleDragEnter}
 		ondragleave={handleDragLeave}
 		ondragover={handleDragOver}
@@ -359,57 +356,59 @@
 		onscroll={handleScroll}
 		role="main"
 	>
-		<ChatMessages
-			class="mb-16 md:mb-24"
-			messages={activeMessages()}
-			onUserAction={() => {
-				autoScroll.enable();
-				autoScroll.scrollToBottom();
-			}}
-		/>
+		<div class="flex flex-col">
+			<ChatMessages
+				class="mb-16 md:mb-24"
+				messages={activeMessages()}
+				onUserAction={() => {
+					autoScroll.enable();
+					autoScroll.scrollToBottom();
+				}}
+			/>
 
-		<div
-			class="pointer-events-none sticky right-0 bottom-4 left-0 mt-auto"
-			in:slide={{ duration: 150, axis: 'y' }}
-		>
-			<ChatScreenProcessingInfo />
+			<div
+				class="pointer-events-none sticky right-0 bottom-4 left-0 mt-auto"
+				in:slide={{ duration: 150, axis: 'y' }}
+			>
+				<ChatScreenProcessingInfo />
 
-			{#if hasPropsError}
-				<div
-					class="pointer-events-auto mx-auto mb-4 max-w-[48rem] px-1"
-					in:fly={{ y: 10, duration: 250 }}
-				>
-					<Alert.Root variant="destructive">
-						<AlertTriangle class="h-4 w-4" />
-						<Alert.Title class="flex items-center justify-between">
-							<span>Server unavailable</span>
-							<button
-								onclick={() => serverStore.fetch()}
-								disabled={isServerLoading}
-								class="flex items-center gap-1.5 rounded-lg bg-destructive/20 px-2 py-1 text-xs font-medium hover:bg-destructive/30 disabled:opacity-50"
-							>
-								<RefreshCw class="h-3 w-3 {isServerLoading ? 'animate-spin' : ''}" />
-								{isServerLoading ? 'Retrying...' : 'Retry'}
-							</button>
-						</Alert.Title>
-						<Alert.Description>{serverError()}</Alert.Description>
-					</Alert.Root>
+				{#if hasPropsError}
+					<div
+						class="pointer-events-auto mx-auto mb-4 max-w-[48rem] px-1"
+						in:fly={{ y: 10, duration: 250 }}
+					>
+						<Alert.Root variant="destructive">
+							<AlertTriangle class="h-4 w-4" />
+							<Alert.Title class="flex items-center justify-between">
+								<span>Server unavailable</span>
+								<button
+									onclick={() => serverStore.fetch()}
+									disabled={isServerLoading}
+									class="flex items-center gap-1.5 rounded-lg bg-destructive/20 px-2 py-1 text-xs font-medium hover:bg-destructive/30 disabled:opacity-50"
+								>
+									<RefreshCw class="h-3 w-3 {isServerLoading ? 'animate-spin' : ''}" />
+									{isServerLoading ? 'Retrying...' : 'Retry'}
+								</button>
+							</Alert.Title>
+							<Alert.Description>{serverError()}</Alert.Description>
+						</Alert.Root>
+					</div>
+				{/if}
+
+				<div class="conversation-chat-form pointer-events-auto rounded-t-3xl">
+					<ChatScreenForm
+						disabled={hasPropsError || isEditing()}
+						{initialMessage}
+						isLoading={isCurrentConversationLoading}
+						onFileRemove={handleFileRemove}
+						onFileUpload={handleFileUpload}
+						onSend={handleSendMessage}
+						onStop={() => chatStore.stopGeneration()}
+						onSystemPromptAdd={handleSystemPromptAdd}
+						showHelperText={false}
+						bind:uploadedFiles
+					/>
 				</div>
-			{/if}
-
-			<div class="conversation-chat-form pointer-events-auto rounded-t-3xl">
-				<ChatScreenForm
-					disabled={hasPropsError || isEditing()}
-					{initialMessage}
-					isLoading={isCurrentConversationLoading}
-					onFileRemove={handleFileRemove}
-					onFileUpload={handleFileUpload}
-					onSend={handleSendMessage}
-					onStop={() => chatStore.stopGeneration()}
-					onSystemPromptAdd={handleSystemPromptAdd}
-					showHelperText={false}
-					bind:uploadedFiles
-				/>
 			</div>
 		</div>
 	</div>
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatScreen/ChatScreenHeader.svelte b/tools/server/webui/src/lib/components/app/chat/ChatScreen/ChatScreenHeader.svelte
index 2730d5e73..0a920be6f 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatScreen/ChatScreenHeader.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatScreen/ChatScreenHeader.svelte
@@ -1,16 +1,11 @@
 <script lang="ts">
 	import { Settings } from '@lucide/svelte';
-	import { DialogChatSettings } from '$lib/components/app';
 	import { Button } from '$lib/components/ui/button';
 	import { useSidebar } from '$lib/components/ui/sidebar';
+	import { getChatSettingsDialogContext } from '$lib/contexts';
 
 	const sidebar = useSidebar();
-
-	let settingsOpen = $state(false);
-
-	function toggleSettings() {
-		settingsOpen = true;
-	}
+	const chatSettingsDialog = getChatSettingsDialogContext();
 </script>
 
 <header
@@ -22,12 +17,10 @@
 		<Button
 			variant="ghost"
 			size="icon-lg"
-			onclick={toggleSettings}
+			onclick={() => chatSettingsDialog.open()}
 			class="rounded-full backdrop-blur-lg"
 		>
 			<Settings class="h-4 w-4" />
 		</Button>
 	</div>
 </header>
-
-<DialogChatSettings open={settingsOpen} onOpenChange={(open) => (settingsOpen = open)} />
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatSettings/ChatSettings.svelte b/tools/server/webui/src/lib/components/app/chat/ChatSettings/ChatSettings.svelte
index 44d59e2b3..995dd1fdd 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatSettings/ChatSettings.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatSettings/ChatSettings.svelte
@@ -296,6 +296,11 @@
 					label: 'Disable reasoning content parsing',
 					type: SettingsFieldType.CHECKBOX
 				},
+				{
+					key: SETTINGS_KEYS.EXCLUDE_REASONING_FROM_CONTEXT,
+					label: 'Exclude reasoning from context',
+					type: SettingsFieldType.CHECKBOX
+				},
 				{
 					key: SETTINGS_KEYS.SHOW_RAW_OUTPUT_SWITCH,
 					label: 'Enable raw output toggle',
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebar.svelte b/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebar.svelte
index b2c0baebb..1d05ea727 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebar.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebar.svelte
@@ -1,13 +1,18 @@
 <script lang="ts">
 	import { goto } from '$app/navigation';
 	import { page } from '$app/state';
-	import { Trash2 } from '@lucide/svelte';
+	import { Trash2, Pencil } from '@lucide/svelte';
 	import { ChatSidebarConversationItem, DialogConfirmation } from '$lib/components/app';
+	import { Checkbox } from '$lib/components/ui/checkbox';
+	import Label from '$lib/components/ui/label/label.svelte';
 	import ScrollArea from '$lib/components/ui/scroll-area/scroll-area.svelte';
 	import * as Sidebar from '$lib/components/ui/sidebar';
-	import * as AlertDialog from '$lib/components/ui/alert-dialog';
 	import Input from '$lib/components/ui/input/input.svelte';
-	import { conversationsStore, conversations } from '$lib/stores/conversations.svelte';
+	import {
+		conversationsStore,
+		conversations,
+		buildConversationTree
+	} from '$lib/stores/conversations.svelte';
 	import { chatStore } from '$lib/stores/chat.svelte';
 	import { getPreviewText } from '$lib/utils';
 	import ChatSidebarActions from './ChatSidebarActions.svelte';
@@ -18,6 +23,7 @@
 	let isSearchModeActive = $state(false);
 	let searchQuery = $state('');
 	let showDeleteDialog = $state(false);
+	let deleteWithForks = $state(false);
 	let showEditDialog = $state(false);
 	let selectedConversation = $state<DatabaseConversation | null>(null);
 	let editedName = $state('');
@@ -35,10 +41,30 @@
 		return conversations();
 	});
 
+	let conversationTree = $derived(buildConversationTree(filteredConversations));
+
+	let selectedConversationHasDescendants = $derived.by(() => {
+		if (!selectedConversation) return false;
+
+		const allConvs = conversations();
+		const queue = [selectedConversation.id];
+
+		while (queue.length > 0) {
+			const parentId = queue.pop()!;
+
+			for (const c of allConvs) {
+				if (c.forkedFromConversationId === parentId) return true;
+			}
+		}
+
+		return false;
+	});
+
 	async function handleDeleteConversation(id: string) {
 		const conversation = conversations().find((conv) => conv.id === id);
 		if (conversation) {
 			selectedConversation = conversation;
+			deleteWithForks = false;
 			showDeleteDialog = true;
 		}
 	}
@@ -54,11 +80,14 @@
 
 	function handleConfirmDelete() {
 		if (selectedConversation) {
+			const convId = selectedConversation.id;
+			const withForks = deleteWithForks;
 			showDeleteDialog = false;
 
 			setTimeout(() => {
-				conversationsStore.deleteConversation(selectedConversation.id);
-				selectedConversation = null;
+				conversationsStore.deleteConversation(convId, {
+					deleteWithForks: withForks
+				});
 			}, 100); // Wait for animation to finish
 		}
 	}
@@ -110,7 +139,7 @@
 </script>
 
 <ScrollArea class="h-[100vh]">
-	<Sidebar.Header class=" top-0 z-10 gap-6 bg-sidebar/50 px-4 py-4 pb-2 backdrop-blur-lg md:sticky">
+	<Sidebar.Header class=" top-0 z-10 gap-4 bg-sidebar/50 p-4 pb-2 backdrop-blur-lg md:sticky">
 		<a href="#/" onclick={handleMobileSidebarItemClick}>
 			<h1 class="inline-flex items-center gap-1 px-2 text-xl font-semibold">llama.cpp UI</h1>
 		</a>
@@ -118,7 +147,7 @@
 		<ChatSidebarActions {handleMobileSidebarItemClick} bind:isSearchModeActive bind:searchQuery />
 	</Sidebar.Header>
 
-	<Sidebar.Group class="mt-4 space-y-2 p-0 px-4">
+	<Sidebar.Group class="mt-2 space-y-2 p-0 px-4">
 		{#if (filteredConversations.length > 0 && isSearchModeActive) || !isSearchModeActive}
 			<Sidebar.GroupLabel>
 				{isSearchModeActive ? 'Search results' : 'Conversations'}
@@ -127,15 +156,17 @@
 
 		<Sidebar.GroupContent>
 			<Sidebar.Menu>
-				{#each filteredConversations as conversation (conversation.id)}
-					<Sidebar.MenuItem class="mb-1">
+				{#each conversationTree as { conversation, depth } (conversation.id)}
+					<Sidebar.MenuItem class="mb-1 p-0">
 						<ChatSidebarConversationItem
 							conversation={{
 								id: conversation.id,
 								name: conversation.name,
 								lastModified: conversation.lastModified,
-								currNode: conversation.currNode
+								currNode: conversation.currNode,
+								forkedFromConversationId: conversation.forkedFromConversationId
 							}}
+							{depth}
 							{handleMobileSidebarItemClick}
 							isActive={currentChatId === conversation.id}
 							onSelect={selectConversation}
@@ -146,7 +177,7 @@
 					</Sidebar.MenuItem>
 				{/each}
 
-				{#if filteredConversations.length === 0}
+				{#if conversationTree.length === 0}
 					<div class="px-2 py-4 text-center">
 						<p class="mb-4 p-4 text-sm text-muted-foreground">
 							{searchQuery.length > 0
@@ -177,35 +208,40 @@
 		showDeleteDialog = false;
 		selectedConversation = null;
 	}}
-/>
+>
+	{#if selectedConversationHasDescendants}
+		<div class="flex items-center gap-2 py-2">
+			<Checkbox id="delete-with-forks" bind:checked={deleteWithForks} />
 
-<AlertDialog.Root bind:open={showEditDialog}>
-	<AlertDialog.Content>
-		<AlertDialog.Header>
-			<AlertDialog.Title>Edit Conversation Name</AlertDialog.Title>
-			<AlertDialog.Description>
-				<Input
-					class="mt-4 text-foreground"
-					onkeydown={(e) => {
-						if (e.key === 'Enter') {
-							e.preventDefault();
-							handleConfirmEdit();
-						}
-					}}
-					placeholder="Enter a new name"
-					type="text"
-					bind:value={editedName}
-				/>
-			</AlertDialog.Description>
-		</AlertDialog.Header>
-		<AlertDialog.Footer>
-			<AlertDialog.Cancel
-				onclick={() => {
-					showEditDialog = false;
-					selectedConversation = null;
-				}}>Cancel</AlertDialog.Cancel
-			>
-			<AlertDialog.Action onclick={handleConfirmEdit}>Save</AlertDialog.Action>
-		</AlertDialog.Footer>
-	</AlertDialog.Content>
-</AlertDialog.Root>
+			<Label for="delete-with-forks" class="text-sm">Also delete all forked conversations</Label>
+		</div>
+	{/if}
+</DialogConfirmation>
+
+<DialogConfirmation
+	bind:open={showEditDialog}
+	title="Edit Conversation Name"
+	description=""
+	confirmText="Save"
+	cancelText="Cancel"
+	icon={Pencil}
+	onConfirm={handleConfirmEdit}
+	onCancel={() => {
+		showEditDialog = false;
+		selectedConversation = null;
+	}}
+	onKeydown={(e) => {
+		if (e.key === 'Enter') {
+			e.preventDefault();
+			e.stopImmediatePropagation();
+			handleConfirmEdit();
+		}
+	}}
+>
+	<Input
+		class="text-foreground"
+		placeholder="Enter a new name"
+		type="text"
+		bind:value={editedName}
+	/>
+</DialogConfirmation>
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebarActions.svelte b/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebarActions.svelte
index 30d1f9d4b..e0d379f6d 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebarActions.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebarActions.svelte
@@ -3,6 +3,9 @@
 	import { KeyboardShortcutInfo } from '$lib/components/app';
 	import { Button } from '$lib/components/ui/button';
 	import { Input } from '$lib/components/ui/input';
+	import { McpLogo } from '$lib/components/app';
+	import { SETTINGS_SECTION_TITLES } from '$lib/constants';
+	import { getChatSettingsDialogContext } from '$lib/contexts';
 
 	interface Props {
 		handleMobileSidebarItemClick: () => void;
@@ -18,6 +21,8 @@
 
 	let searchInput: HTMLInputElement | null = $state(null);
 
+	const chatSettingsDialog = getChatSettingsDialogContext();
+
 	function handleSearchModeDeactivate() {
 		isSearchModeActive = false;
 		searchQuery = '';
@@ -30,7 +35,7 @@
 	});
 </script>
 
-<div class="space-y-0.5">
+<div class="my-1 space-y-1">
 	{#if isSearchModeActive}
 		<div class="relative">
 			<Search class="absolute top-2.5 left-2 h-4 w-4 text-muted-foreground" />
@@ -50,13 +55,14 @@
 		</div>
 	{:else}
 		<Button
-			class="w-full justify-between hover:[&>kbd]:opacity-100"
+			class="w-full justify-between backdrop-blur-none! hover:[&>kbd]:opacity-100"
 			href="?new_chat=true#/"
 			onclick={handleMobileSidebarItemClick}
 			variant="ghost"
 		>
 			<div class="flex items-center gap-2">
 				<SquarePen class="h-4 w-4" />
+
 				New chat
 			</div>
 
@@ -64,7 +70,7 @@
 		</Button>
 
 		<Button
-			class="w-full justify-between hover:[&>kbd]:opacity-100"
+			class="w-full justify-between backdrop-blur-none! hover:[&>kbd]:opacity-100"
 			onclick={() => {
 				isSearchModeActive = true;
 			}}
@@ -72,10 +78,25 @@
 		>
 			<div class="flex items-center gap-2">
 				<Search class="h-4 w-4" />
-				Search conversations
+
+				Search
 			</div>
 
 			<KeyboardShortcutInfo keys={['cmd', 'k']} />
 		</Button>
+
+		<Button
+			class="w-full justify-between backdrop-blur-none! hover:[&>kbd]:opacity-100"
+			onclick={() => {
+				chatSettingsDialog.open(SETTINGS_SECTION_TITLES.MCP);
+			}}
+			variant="ghost"
+		>
+			<div class="flex items-center gap-2">
+				<McpLogo class="h-4 w-4" />
+
+				MCP Servers
+			</div>
+		</Button>
 	{/if}
 </div>
diff --git a/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebarConversationItem.svelte b/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebarConversationItem.svelte
index 5c48909cd..e22cb4f82 100644
--- a/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebarConversationItem.svelte
+++ b/tools/server/webui/src/lib/components/app/chat/ChatSidebar/ChatSidebarConversationItem.svelte
@@ -1,13 +1,23 @@
 <script lang="ts">
-	import { Trash2, Pencil, MoreHorizontal, Download, Loader2, Square } from '@lucide/svelte';
+	import {
+		Trash2,
+		Pencil,
+		MoreHorizontal,
+		Download,
+		Loader2,
+		Square,
+		GitBranch
+	} from '@lucide/svelte';
 	import { DropdownMenuActions } from '$lib/components/app';
 	import * as Tooltip from '$lib/components/ui/tooltip';
+	import { FORK_TREE_DEPTH_PADDING } from '$lib/constants';
 	import { getAllLoadingChats } from '$lib/stores/chat.svelte';
 	import { conversationsStore } from '$lib/stores/conversations.svelte';
 	import { onMount } from 'svelte';
 
 	interface Props {
 		isActive?: boolean;
+		depth?: number;
 		conversation: DatabaseConversation;
 		handleMobileSidebarItemClick?: () => void;
 		onDelete?: (id: string) => void;
@@ -23,7 +33,8 @@
 		onEdit,
 		onSelect,
 		onStop,
-		isActive = false
+		isActive = false,
+		depth = 0
 	}: Props = $props();
 
 	let renderActionsDropdown = $state(false);
@@ -88,14 +99,34 @@
 
 <!-- svelte-ignore a11y_mouse_events_have_key_events -->
 <button
-	class="group flex min-h-9 w-full cursor-pointer items-center justify-between space-x-3 rounded-lg px-3 py-1.5 text-left transition-colors hover:bg-foreground/10 {isActive
+	class="group flex min-h-9 w-full cursor-pointer items-center justify-between space-x-3 rounded-lg py-1.5 text-left transition-colors hover:bg-foreground/10 {isActive
 		? 'bg-foreground/5 text-accent-foreground'
-		: ''}"
+		: ''} px-3"
 	onclick={handleSelect}
 	onmouseover={handleMouseOver}
 	onmouseleave={handleMouseLeave}
 >
-	<div class="flex min-w-0 flex-1 items-center gap-2">
+	<div
+		class="flex min-w-0 flex-1 items-center gap-2"
+		style:padding-left="{depth * FORK_TREE_DEPTH_PADDING}px"
+	>
+		{#if depth > 0}
+			<Tooltip.Root>
+				<Tooltip.Trigger>
+					<a
+						href="#/chat/{conversation.forkedFromConversationId}"
+						class="flex shrink-0 items-center text-muted-foreground transition-colors hover:text-foreground"
+					>
+						<GitBranch class="h-3.5 w-3.5" />
+					</a>
+				</Tooltip.Trigger>
+
+				<Tooltip.Content>
+					<p>See parent conversation</p>
+				</Tooltip.Content>
+			</Tooltip.Root>
+		{/if}
+
 		{#if isLoading}
 			<Tooltip.Root>
 				<Tooltip.Trigger>
diff --git a/tools/server/webui/src/lib/components/app/content/MarkdownContent.svelte b/tools/server/webui/src/lib/components/app/content/MarkdownContent.svelte
index 0b10e1400..9976ffa7c 100644
--- a/tools/server/webui/src/lib/components/app/content/MarkdownContent.svelte
+++ b/tools/server/webui/src/lib/components/app/content/MarkdownContent.svelte
@@ -36,6 +36,7 @@
 	import { createAutoScrollController } from '$lib/hooks/use-auto-scroll.svelte';
 	import type { DatabaseMessageExtra } from '$lib/types/database';
 	import { config } from '$lib/stores/settings.svelte';
+	import { fadeInView } from '$lib/actions/fade-in-view.svelte';
 
 	interface Props {
 		attachments?: DatabaseMessageExtra[];
@@ -598,7 +599,7 @@
 		: ''}"
 >
 	{#each renderedBlocks as block (block.id)}
-		<div class="markdown-block" data-block-id={block.id}>
+		<div class="markdown-block" data-block-id={block.id} use:fadeInView={{ skipIfVisible: true }}>
 			<!-- eslint-disable-next-line no-at-html-tags -->
 			{@html block.html}
 		</div>
@@ -651,7 +652,6 @@
 />
 
 <style>
-	.markdown-block,
 	.markdown-block--unstable {
 		display: contents;
 	}
diff --git a/tools/server/webui/src/lib/components/app/dialogs/DialogConfirmation.svelte b/tools/server/webui/src/lib/components/app/dialogs/DialogConfirmation.svelte
index d8aa66f3e..41fdf7d41 100644
--- a/tools/server/webui/src/lib/components/app/dialogs/DialogConfirmation.svelte
+++ b/tools/server/webui/src/lib/components/app/dialogs/DialogConfirmation.svelte
@@ -1,6 +1,6 @@
 <script lang="ts">
 	import * as AlertDialog from '$lib/components/ui/alert-dialog';
-	import type { Component } from 'svelte';
+	import type { Component, Snippet } from 'svelte';
 	import { KeyboardKey } from '$lib/enums';
 
 	interface Props {
@@ -14,6 +14,7 @@
 		onConfirm: () => void;
 		onCancel: () => void;
 		onKeydown?: (event: KeyboardEvent) => void;
+		children?: Snippet;
 	}
 
 	let {
@@ -26,7 +27,8 @@
 		icon,
 		onConfirm,
 		onCancel,
-		onKeydown
+		onKeydown,
+		children
 	}: Props = $props();
 
 	function handleKeydown(event: KeyboardEvent) {
@@ -60,6 +62,10 @@
 			</AlertDialog.Description>
 		</AlertDialog.Header>
 
+		{#if children}
+			{@render children()}
+		{/if}
+
 		<AlertDialog.Footer>
 			<AlertDialog.Cancel onclick={onCancel}>{cancelText}</AlertDialog.Cancel>
 			<AlertDialog.Action
diff --git a/tools/server/webui/src/lib/constants/agentic.ts b/tools/server/webui/src/lib/constants/agentic.ts
index 7ff9e4e52..ac31d5126 100644
--- a/tools/server/webui/src/lib/constants/agentic.ts
+++ b/tools/server/webui/src/lib/constants/agentic.ts
@@ -50,6 +50,8 @@ export const AGENTIC_REGEX = {
 	PARTIAL_MARKER: /<<<[A-Za-z_]*$/,
 	// Matches reasoning content blocks (including tags)
 	REASONING_BLOCK: /<<<reasoning_content_start>>>[\s\S]*?<<<reasoning_content_end>>>/g,
+	// Captures the reasoning text between start/end tags
+	REASONING_EXTRACT: /<<<reasoning_content_start>>>([\s\S]*?)<<<reasoning_content_end>>>/,
 	// Matches an opening reasoning tag and any remaining content (unterminated)
 	REASONING_OPEN: /<<<reasoning_content_start>>>[\s\S]*$/,
 	// Matches a complete agentic tool call display block (start to end marker)
diff --git a/tools/server/webui/src/lib/constants/auto-scroll.ts b/tools/server/webui/src/lib/constants/auto-scroll.ts
index 098f435d6..ca9ba5a9e 100644
--- a/tools/server/webui/src/lib/constants/auto-scroll.ts
+++ b/tools/server/webui/src/lib/constants/auto-scroll.ts
@@ -1,3 +1,2 @@
 export const AUTO_SCROLL_INTERVAL = 100;
-export const INITIAL_SCROLL_DELAY = 50;
 export const AUTO_SCROLL_AT_BOTTOM_THRESHOLD = 10;
diff --git a/tools/server/webui/src/lib/constants/context-keys.ts b/tools/server/webui/src/lib/constants/context-keys.ts
new file mode 100644
index 000000000..ba61d25e8
--- /dev/null
+++ b/tools/server/webui/src/lib/constants/context-keys.ts
@@ -0,0 +1,3 @@
+export const CONTEXT_KEY_MESSAGE_EDIT = 'chat-message-edit';
+export const CONTEXT_KEY_CHAT_ACTIONS = 'chat-actions';
+export const CONTEXT_KEY_CHAT_SETTINGS_DIALOG = 'chat-settings-dialog';
diff --git a/tools/server/webui/src/lib/constants/index.ts b/tools/server/webui/src/lib/constants/index.ts
index f3593c03b..70bed50a6 100644
--- a/tools/server/webui/src/lib/constants/index.ts
+++ b/tools/server/webui/src/lib/constants/index.ts
@@ -10,6 +10,7 @@ export * from './cache';
 export * from './chat-form';
 export * from './code-blocks';
 export * from './code';
+export * from './context-keys';
 export * from './css-classes';
 export * from './favicon';
 export * from './floating-ui-constraints';
diff --git a/tools/server/webui/src/lib/constants/settings-config.ts b/tools/server/webui/src/lib/constants/settings-config.ts
index c1440b09f..196866ece 100644
--- a/tools/server/webui/src/lib/constants/settings-config.ts
+++ b/tools/server/webui/src/lib/constants/settings-config.ts
@@ -10,6 +10,7 @@ export const SETTING_CONFIG_DEFAULT: Record<string, string | number | boolean |
 	theme: ColorMode.SYSTEM,
 	showThoughtInProgress: false,
 	disableReasoningParsing: false,
+	excludeReasoningFromContext: false,
 	showRawOutputSwitch: false,
 	keepStatsVisible: false,
 	showMessageStats: true,
@@ -106,6 +107,8 @@ export const SETTING_CONFIG_INFO: Record<string, string> = {
 	showThoughtInProgress: 'Expand thought process by default when generating messages.',
 	disableReasoningParsing:
 		'Send reasoning_format=none to prevent server-side extraction of reasoning tokens into separate field',
+	excludeReasoningFromContext:
+		'Strip reasoning content from previous messages before sending to the model. When unchecked, reasoning is sent back via the reasoning_content field so the model can see its own chain-of-thought across turns.',
 	showRawOutputSwitch:
 		'Show toggle button to display messages as plain text instead of Markdown-formatted content',
 	keepStatsVisible: 'Keep processing statistics visible after generation finishes.',
diff --git a/tools/server/webui/src/lib/constants/settings-keys.ts b/tools/server/webui/src/lib/constants/settings-keys.ts
index 120910357..c8b4b503a 100644
--- a/tools/server/webui/src/lib/constants/settings-keys.ts
+++ b/tools/server/webui/src/lib/constants/settings-keys.ts
@@ -54,6 +54,7 @@ export const SETTINGS_KEYS = {
 	SHOW_TOOL_CALL_IN_PROGRESS: 'showToolCallInProgress',
 	// Developer
 	DISABLE_REASONING_PARSING: 'disableReasoningParsing',
+	EXCLUDE_REASONING_FROM_CONTEXT: 'excludeReasoningFromContext',
 	SHOW_RAW_OUTPUT_SWITCH: 'showRawOutputSwitch',
 	CUSTOM: 'custom'
 } as const;
diff --git a/tools/server/webui/src/lib/constants/ui.ts b/tools/server/webui/src/lib/constants/ui.ts
index a75b30f2f..be9c50d07 100644
--- a/tools/server/webui/src/lib/constants/ui.ts
+++ b/tools/server/webui/src/lib/constants/ui.ts
@@ -1 +1,2 @@
+export const FORK_TREE_DEPTH_PADDING = 8;
 export const SYSTEM_MESSAGE_PLACEHOLDER = 'System message';
diff --git a/tools/server/webui/src/lib/contexts/chat-actions.context.ts b/tools/server/webui/src/lib/contexts/chat-actions.context.ts
index eba0fec02..e9050fa27 100644
--- a/tools/server/webui/src/lib/contexts/chat-actions.context.ts
+++ b/tools/server/webui/src/lib/contexts/chat-actions.context.ts
@@ -1,4 +1,5 @@
 import { getContext, setContext } from 'svelte';
+import { CONTEXT_KEY_CHAT_ACTIONS } from '$lib/constants';
 
 export interface ChatActionsContext {
 	copy: (message: DatabaseMessage) => void;
@@ -21,9 +22,13 @@ export interface ChatActionsContext {
 	) => void;
 	regenerateWithBranching: (message: DatabaseMessage, modelOverride?: string) => void;
 	continueAssistantMessage: (message: DatabaseMessage) => void;
+	forkConversation: (
+		message: DatabaseMessage,
+		options: { name: string; includeAttachments: boolean }
+	) => void;
 }
 
-const CHAT_ACTIONS_KEY = Symbol.for('chat-actions');
+const CHAT_ACTIONS_KEY = Symbol.for(CONTEXT_KEY_CHAT_ACTIONS);
 
 export function setChatActionsContext(ctx: ChatActionsContext): ChatActionsContext {
 	return setContext(CHAT_ACTIONS_KEY, ctx);
diff --git a/tools/server/webui/src/lib/contexts/chat-settings-dialog.context.ts b/tools/server/webui/src/lib/contexts/chat-settings-dialog.context.ts
new file mode 100644
index 000000000..3f98c3c0e
--- /dev/null
+++ b/tools/server/webui/src/lib/contexts/chat-settings-dialog.context.ts
@@ -0,0 +1,19 @@
+import { getContext, setContext } from 'svelte';
+import type { SettingsSectionTitle } from '$lib/constants';
+import { CONTEXT_KEY_CHAT_SETTINGS_DIALOG } from '$lib/constants';
+
+export interface ChatSettingsDialogContext {
+	open: (initialSection?: SettingsSectionTitle) => void;
+}
+
+const CHAT_SETTINGS_DIALOG_KEY = Symbol.for(CONTEXT_KEY_CHAT_SETTINGS_DIALOG);
+
+export function setChatSettingsDialogContext(
+	ctx: ChatSettingsDialogContext
+): ChatSettingsDialogContext {
+	return setContext(CHAT_SETTINGS_DIALOG_KEY, ctx);
+}
+
+export function getChatSettingsDialogContext(): ChatSettingsDialogContext {
+	return getContext(CHAT_SETTINGS_DIALOG_KEY);
+}
diff --git a/tools/server/webui/src/lib/contexts/index.ts b/tools/server/webui/src/lib/contexts/index.ts
index 73ff6f96f..6c69070d8 100644
--- a/tools/server/webui/src/lib/contexts/index.ts
+++ b/tools/server/webui/src/lib/contexts/index.ts
@@ -11,3 +11,9 @@ export {
 	setChatActionsContext,
 	type ChatActionsContext
 } from './chat-actions.context';
+
+export {
+	getChatSettingsDialogContext,
+	setChatSettingsDialogContext,
+	type ChatSettingsDialogContext
+} from './chat-settings-dialog.context';
diff --git a/tools/server/webui/src/lib/contexts/message-edit.context.ts b/tools/server/webui/src/lib/contexts/message-edit.context.ts
index 7af116daa..2d344cca5 100644
--- a/tools/server/webui/src/lib/contexts/message-edit.context.ts
+++ b/tools/server/webui/src/lib/contexts/message-edit.context.ts
@@ -1,4 +1,5 @@
 import { getContext, setContext } from 'svelte';
+import { CONTEXT_KEY_MESSAGE_EDIT } from '$lib/constants';
 
 export interface MessageEditState {
 	readonly isEditing: boolean;
@@ -22,7 +23,7 @@ export interface MessageEditActions {
 
 export type MessageEditContext = MessageEditState & MessageEditActions;
 
-const MESSAGE_EDIT_KEY = Symbol.for('chat-message-edit');
+const MESSAGE_EDIT_KEY = Symbol.for(CONTEXT_KEY_MESSAGE_EDIT);
 
 /**
  * Sets the message edit context. Call this in the parent component (ChatMessage.svelte).
diff --git a/tools/server/webui/src/lib/hooks/use-auto-scroll.svelte.ts b/tools/server/webui/src/lib/hooks/use-auto-scroll.svelte.ts
index afd0aad83..204d71a90 100644
--- a/tools/server/webui/src/lib/hooks/use-auto-scroll.svelte.ts
+++ b/tools/server/webui/src/lib/hooks/use-auto-scroll.svelte.ts
@@ -1,8 +1,8 @@
 import { AUTO_SCROLL_AT_BOTTOM_THRESHOLD, AUTO_SCROLL_INTERVAL } from '$lib/constants';
 
 export interface AutoScrollOptions {
-	/** Whether auto-scroll is disabled globally (e.g., from settings) */
 	disabled?: boolean;
+	isColumnReverse?: boolean;
 }
 
 /**
@@ -12,6 +12,7 @@ export interface AutoScrollOptions {
  * - Auto-scrolls to bottom during streaming/loading
  * - Stops auto-scroll when user manually scrolls up
  * - Resumes auto-scroll when user scrolls back to bottom
+ * - Supports both normal and column-reverse scroll containers
  */
 export class AutoScrollController {
 	private _autoScrollEnabled = $state(true);
@@ -21,9 +22,14 @@ export class AutoScrollController {
 	private _scrollTimeout: ReturnType<typeof setTimeout> | undefined;
 	private _container: HTMLElement | undefined;
 	private _disabled: boolean;
+	private _isColumnReverse: boolean;
+	private _mutationObserver: MutationObserver | null = null;
+	private _rafPending = false;
+	private _observerEnabled = false;
 
 	constructor(options: AutoScrollOptions = {}) {
 		this._disabled = options.disabled ?? false;
+		this._isColumnReverse = options.isColumnReverse ?? false;
 	}
 
 	get autoScrollEnabled(): boolean {
@@ -38,7 +44,12 @@ export class AutoScrollController {
 	 * Binds the controller to a scrollable container element.
 	 */
 	setContainer(container: HTMLElement | undefined): void {
+		this._doStopObserving();
 		this._container = container;
+
+		if (this._observerEnabled && container && !this._disabled) {
+			this._doStartObserving();
+		}
 	}
 
 	/**
@@ -49,6 +60,9 @@ export class AutoScrollController {
 		if (disabled) {
 			this._autoScrollEnabled = false;
 			this.stopInterval();
+			this._doStopObserving();
+		} else if (this._observerEnabled && this._container && !this._mutationObserver) {
+			this._doStartObserving();
 		}
 	}
 
@@ -59,10 +73,23 @@ export class AutoScrollController {
 		if (this._disabled || !this._container) return;
 
 		const { scrollTop, scrollHeight, clientHeight } = this._container;
-		const distanceFromBottom = scrollHeight - scrollTop - clientHeight;
+
+		let distanceFromBottom: number;
+		let isScrollingUp: boolean;
+
+		if (this._isColumnReverse) {
+			// column-reverse: scrollTop=0 at bottom, negative when scrolled up
+			distanceFromBottom = Math.abs(scrollTop);
+			isScrollingUp = scrollTop < this._lastScrollTop;
+		} else {
+			// normal: scrollTop=0 at top, increases when scrolled down
+			distanceFromBottom = scrollHeight - clientHeight - scrollTop;
+			isScrollingUp = scrollTop < this._lastScrollTop;
+		}
+
 		const isAtBottom = distanceFromBottom < AUTO_SCROLL_AT_BOTTOM_THRESHOLD;
 
-		if (scrollTop < this._lastScrollTop && !isAtBottom) {
+		if (isScrollingUp && !isAtBottom) {
 			this._userScrolledUp = true;
 			this._autoScrollEnabled = false;
 		} else if (isAtBottom && this._userScrolledUp) {
@@ -90,10 +117,12 @@ export class AutoScrollController {
 	scrollToBottom(behavior: ScrollBehavior = 'smooth'): void {
 		if (this._disabled || !this._container) return;
 
-		this._container.scrollTo({
-			top: this._container.scrollHeight,
-			behavior
-		});
+		if (this._isColumnReverse) {
+			// column-reverse: scrollTop=0 is the bottom
+			this._container.scrollTo({ top: 0, behavior });
+		} else {
+			this._container.scrollTo({ top: this._container.scrollHeight, behavior });
+		}
 	}
 
 	/**
@@ -150,11 +179,69 @@ export class AutoScrollController {
 	 */
 	destroy(): void {
 		this.stopInterval();
+		this._doStopObserving();
+
 		if (this._scrollTimeout) {
 			clearTimeout(this._scrollTimeout);
 			this._scrollTimeout = undefined;
 		}
 	}
+
+	/**
+	 * Starts a MutationObserver on the container that auto-scrolls to bottom
+	 * on content changes. More responsive than interval-based polling.
+	 */
+	startObserving(): void {
+		this._observerEnabled = true;
+
+		if (this._container && !this._disabled && !this._mutationObserver) {
+			this._doStartObserving();
+		}
+	}
+
+	/**
+	 * Stops the MutationObserver.
+	 */
+	stopObserving(): void {
+		this._observerEnabled = false;
+		this._doStopObserving();
+	}
+
+	private _doStartObserving(): void {
+		if (!this._container || this._mutationObserver) return;
+
+		const isReverse = this._isColumnReverse;
+
+		this._mutationObserver = new MutationObserver(() => {
+			if (!this._autoScrollEnabled || this._rafPending) return;
+			this._rafPending = true;
+			requestAnimationFrame(() => {
+				this._rafPending = false;
+				if (this._autoScrollEnabled && this._container) {
+					if (isReverse) {
+						// column-reverse: scrollTop=0 is the bottom
+						this._container.scrollTop = 0;
+					} else {
+						this._container.scrollTop = this._container.scrollHeight;
+					}
+				}
+			});
+		});
+
+		this._mutationObserver.observe(this._container, {
+			childList: true,
+			subtree: true,
+			characterData: true
+		});
+	}
+
+	private _doStopObserving(): void {
+		if (this._mutationObserver) {
+			this._mutationObserver.disconnect();
+			this._mutationObserver = null;
+		}
+		this._rafPending = false;
+	}
 }
 
 /**
diff --git a/tools/server/webui/src/lib/services/chat.service.ts b/tools/server/webui/src/lib/services/chat.service.ts
index eb7f090b6..735291328 100644
--- a/tools/server/webui/src/lib/services/chat.service.ts
+++ b/tools/server/webui/src/lib/services/chat.service.ts
@@ -57,6 +57,46 @@ export class ChatService {
 	 *
 	 */
 
+	/**
+	 * Extracts reasoning text from content that contains internal reasoning tags.
+	 * Returns the concatenated reasoning content or undefined if none found.
+	 */
+	private static extractReasoningFromContent(
+		content: ApiChatMessageData['content'] | null | undefined
+	): string | undefined {
+		if (!content) return undefined;
+
+		const extractFromString = (text: string): string => {
+			const parts: string[] = [];
+			// Use a fresh regex instance to avoid shared lastIndex state
+			const re = new RegExp(AGENTIC_REGEX.REASONING_EXTRACT.source);
+			let match = re.exec(text);
+			while (match) {
+				parts.push(match[1]);
+				// advance past the matched portion and retry
+				text = text.slice(match.index + match[0].length);
+				match = re.exec(text);
+			}
+			return parts.join('');
+		};
+
+		if (typeof content === 'string') {
+			const result = extractFromString(content);
+			return result || undefined;
+		}
+
+		if (!Array.isArray(content)) return undefined;
+
+		const parts: string[] = [];
+		for (const part of content) {
+			if (part.type === ContentPartType.TEXT && part.text) {
+				const result = extractFromString(part.text);
+				if (result) parts.push(result);
+			}
+		}
+		return parts.length > 0 ? parts.join('') : undefined;
+	}
+
 	/**
 	 * Sends a chat completion request to the llama.cpp server.
 	 * Supports both streaming and non-streaming responses with comprehensive parameter configuration.
@@ -111,7 +151,8 @@ export class ChatService {
 			custom,
 			timings_per_token,
 			// Config options
-			disableReasoningParsing
+			disableReasoningParsing,
+			excludeReasoningFromContext
 		} = options;
 
 		const normalizedMessages: ApiChatMessageData[] = messages
@@ -159,14 +200,24 @@ export class ChatService {
 		}
 
 		const requestBody: ApiChatCompletionRequest = {
-			messages: normalizedMessages.map((msg: ApiChatMessageData) => ({
-				role: msg.role,
-				// Strip reasoning tags/content from the prompt to avoid polluting KV cache.
-				// TODO: investigate backend expectations for reasoning tags and add a toggle if needed.
-				content: ChatService.stripReasoningContent(msg.content),
-				tool_calls: msg.tool_calls,
-				tool_call_id: msg.tool_call_id
-			})),
+			messages: normalizedMessages.map((msg: ApiChatMessageData) => {
+				// Always strip internal reasoning/agentic tags from content
+				const cleanedContent = ChatService.stripReasoningContent(msg.content);
+				const mapped: ApiChatCompletionRequest['messages'][0] = {
+					role: msg.role,
+					content: cleanedContent,
+					tool_calls: msg.tool_calls,
+					tool_call_id: msg.tool_call_id
+				};
+				// When preserving reasoning, extract it from raw content and send as separate field
+				if (!excludeReasoningFromContext) {
+					const reasoning = ChatService.extractReasoningFromContent(msg.content);
+					if (reasoning) {
+						mapped.reasoning_content = reasoning;
+					}
+				}
+				return mapped;
+			}),
 			stream,
 			return_progress: stream ? true : undefined,
 			tools: tools && tools.length > 0 ? tools : undefined
diff --git a/tools/server/webui/src/lib/services/database.service.ts b/tools/server/webui/src/lib/services/database.service.ts
index 53ec80c27..8f7b81fe9 100644
--- a/tools/server/webui/src/lib/services/database.service.ts
+++ b/tools/server/webui/src/lib/services/database.service.ts
@@ -1,5 +1,6 @@
 import Dexie, { type EntityTable } from 'dexie';
-import { findDescendantMessages, uuid } from '$lib/utils';
+import { findDescendantMessages, uuid, filterByLeafNodeId } from '$lib/utils';
+import type { McpServerOverride } from '$lib/types/database';
 
 class LlamacppDatabase extends Dexie {
 	conversations!: EntityTable<DatabaseConversation, string>;
@@ -173,8 +174,47 @@ export class DatabaseService {
 	 *
 	 * @param id - Conversation ID
 	 */
-	static async deleteConversation(id: string): Promise<void> {
+	static async deleteConversation(
+		id: string,
+		options?: { deleteWithForks?: boolean }
+	): Promise<void> {
 		await db.transaction('rw', [db.conversations, db.messages], async () => {
+			if (options?.deleteWithForks) {
+				// Recursively collect all descendant IDs
+				const idsToDelete: string[] = [];
+				const queue = [id];
+
+				while (queue.length > 0) {
+					const parentId = queue.pop()!;
+					const children = await db.conversations
+						.filter((c) => c.forkedFromConversationId === parentId)
+						.toArray();
+
+					for (const child of children) {
+						idsToDelete.push(child.id);
+						queue.push(child.id);
+					}
+				}
+
+				for (const forkId of idsToDelete) {
+					await db.conversations.delete(forkId);
+					await db.messages.where('convId').equals(forkId).delete();
+				}
+			} else {
+				// Reparent direct children to deleted conv's parent
+				const conv = await db.conversations.get(id);
+				const newParent = conv?.forkedFromConversationId;
+				const directChildren = await db.conversations
+					.filter((c) => c.forkedFromConversationId === id)
+					.toArray();
+
+				for (const child of directChildren) {
+					await db.conversations.update(child.id, {
+						forkedFromConversationId: newParent ?? undefined
+					});
+				}
+			}
+
 			await db.conversations.delete(id);
 			await db.messages.where('convId').equals(id).delete();
 		});
@@ -364,4 +404,88 @@ export class DatabaseService {
 			return { imported: importedCount, skipped: skippedCount };
 		});
 	}
+
+	/**
+	 *
+	 *
+	 * Forking
+	 *
+	 *
+	 */
+
+	/**
+	 * Forks a conversation at a specific message, creating a new conversation
+	 * containing all messages from the root up to (and including) the target message.
+	 *
+	 * @param sourceConvId - The source conversation ID
+	 * @param atMessageId - The message ID to fork at (the new conversation ends here)
+	 * @param options - Fork options (name and whether to include attachments)
+	 * @returns The newly created conversation
+	 */
+	static async forkConversation(
+		sourceConvId: string,
+		atMessageId: string,
+		options: { name: string; includeAttachments: boolean }
+	): Promise<DatabaseConversation> {
+		return await db.transaction('rw', [db.conversations, db.messages], async () => {
+			const sourceConv = await db.conversations.get(sourceConvId);
+			if (!sourceConv) {
+				throw new Error(`Source conversation ${sourceConvId} not found`);
+			}
+
+			const allMessages = await db.messages.where('convId').equals(sourceConvId).toArray();
+
+			const pathMessages = filterByLeafNodeId(allMessages, atMessageId, true) as DatabaseMessage[];
+			if (pathMessages.length === 0) {
+				throw new Error(`Could not resolve message path to ${atMessageId}`);
+			}
+
+			const idMap = new Map<string, string>();
+
+			for (const msg of pathMessages) {
+				idMap.set(msg.id, uuid());
+			}
+
+			const newConvId = uuid();
+			const clonedMessages: DatabaseMessage[] = pathMessages.map((msg) => {
+				const newId = idMap.get(msg.id)!;
+				const newParent = msg.parent ? (idMap.get(msg.parent) ?? null) : null;
+				const newChildren = msg.children
+					.filter((childId: string) => idMap.has(childId))
+					.map((childId: string) => idMap.get(childId)!);
+
+				return {
+					...msg,
+					id: newId,
+					convId: newConvId,
+					parent: newParent,
+					children: newChildren,
+					extra: options.includeAttachments ? msg.extra : undefined
+				};
+			});
+
+			const lastClonedMessage = clonedMessages[clonedMessages.length - 1];
+			const newConv: DatabaseConversation = {
+				id: newConvId,
+				name: options.name,
+				lastModified: Date.now(),
+				currNode: lastClonedMessage.id,
+				forkedFromConversationId: sourceConvId,
+				mcpServerOverrides: sourceConv.mcpServerOverrides
+					? sourceConv.mcpServerOverrides.map((o: McpServerOverride) => ({
+							serverId: o.serverId,
+							enabled: o.enabled
+						}))
+					: undefined
+			};
+
+			await db.conversations.add(newConv);
+
+			for (const msg of clonedMessages) {
+				await db.messages.add(msg);
+			}
+
+			return newConv;
+		});
+	}
 }
diff --git a/tools/server/webui/src/lib/services/parameter-sync.service.ts b/tools/server/webui/src/lib/services/parameter-sync.service.ts
index 9a290129e..cc6692128 100644
--- a/tools/server/webui/src/lib/services/parameter-sync.service.ts
+++ b/tools/server/webui/src/lib/services/parameter-sync.service.ts
@@ -227,6 +227,12 @@ export const SYNCABLE_PARAMETERS: SyncableParameter[] = [
 		serverKey: 'alwaysShowAgenticTurns',
 		type: SyncableParameterType.BOOLEAN,
 		canSync: true
+	},
+	{
+		key: 'excludeReasoningFromContext',
+		serverKey: 'excludeReasoningFromContext',
+		type: SyncableParameterType.BOOLEAN,
+		canSync: true
 	}
 ];
 
diff --git a/tools/server/webui/src/lib/stores/chat.svelte.ts b/tools/server/webui/src/lib/stores/chat.svelte.ts
index e30ec97fe..3af91be83 100644
--- a/tools/server/webui/src/lib/stores/chat.svelte.ts
+++ b/tools/server/webui/src/lib/stores/chat.svelte.ts
@@ -1265,35 +1265,53 @@ class ChatStore {
 		let result = this.getMessageByIdWithRole(messageId, MessageRole.USER);
 		if (!result) result = this.getMessageByIdWithRole(messageId, MessageRole.SYSTEM);
 		if (!result) return;
-		const { message: msg } = result;
+		const { message: msg, index: idx } = result;
 		try {
 			const allMessages = await conversationsStore.getConversationMessages(activeConv.id);
 			const rootMessage = allMessages.find((m) => m.type === 'root' && m.parent === null);
 			const isFirstUserMessage =
 				msg.role === MessageRole.USER && rootMessage && msg.parent === rootMessage.id;
-			const parentId = msg.parent || rootMessage?.id;
-			if (!parentId) return;
 			const extrasToUse =
 				newExtras !== undefined
 					? JSON.parse(JSON.stringify(newExtras))
 					: msg.extra
 						? JSON.parse(JSON.stringify(msg.extra))
 						: undefined;
-			const newMessage = await DatabaseService.createMessageBranch(
-				{
-					convId: msg.convId,
-					type: msg.type,
-					timestamp: Date.now(),
-					role: msg.role,
+
+			let messageIdForResponse: string;
+
+			if (msg.children.length === 0) {
+				// No responses after this message — update in place instead of branching
+				const updates: Partial<DatabaseMessage> = {
 					content: newContent,
-					toolCalls: msg.toolCalls || '',
-					children: [],
-					extra: extrasToUse,
-					model: msg.model
-				},
-				parentId
-			);
-			await conversationsStore.updateCurrentNode(newMessage.id);
+					timestamp: Date.now(),
+					extra: extrasToUse
+				};
+				await DatabaseService.updateMessage(msg.id, updates);
+				conversationsStore.updateMessageAtIndex(idx, updates);
+				messageIdForResponse = msg.id;
+			} else {
+				// Has children — create a new branch as sibling
+				const parentId = msg.parent || rootMessage?.id;
+				if (!parentId) return;
+				const newMessage = await DatabaseService.createMessageBranch(
+					{
+						convId: msg.convId,
+						type: msg.type,
+						timestamp: Date.now(),
+						role: msg.role,
+						content: newContent,
+						toolCalls: msg.toolCalls || '',
+						children: [],
+						extra: extrasToUse,
+						model: msg.model
+					},
+					parentId
+				);
+				await conversationsStore.updateCurrentNode(newMessage.id);
+				messageIdForResponse = newMessage.id;
+			}
+
 			conversationsStore.updateConversationTimestamp();
 			if (isFirstUserMessage && newContent.trim())
 				await conversationsStore.updateConversationTitleWithConfirmation(
@@ -1301,7 +1319,8 @@ class ChatStore {
 					newContent.trim()
 				);
 			await conversationsStore.refreshActiveMessages();
-			if (msg.role === MessageRole.USER) await this.generateResponseForMessage(newMessage.id);
+			if (msg.role === MessageRole.USER)
+				await this.generateResponseForMessage(messageIdForResponse);
 		} catch (error) {
 			console.error('Failed to edit message with branching:', error);
 		}
@@ -1479,6 +1498,8 @@ class ChatStore {
 
 		if (currentConfig.disableReasoningParsing) apiOptions.disableReasoningParsing = true;
 
+		if (currentConfig.excludeReasoningFromContext) apiOptions.excludeReasoningFromContext = true;
+
 		if (hasValue(currentConfig.temperature))
 			apiOptions.temperature = Number(currentConfig.temperature);
 
diff --git a/tools/server/webui/src/lib/stores/conversations.svelte.ts b/tools/server/webui/src/lib/stores/conversations.svelte.ts
index 3cfbd3d1c..5769ee98f 100644
--- a/tools/server/webui/src/lib/stores/conversations.svelte.ts
+++ b/tools/server/webui/src/lib/stores/conversations.svelte.ts
@@ -39,6 +39,12 @@ import {
 	MULTIPLE_UNDERSCORE_REGEX,
 	MCP_DEFAULT_ENABLED_LOCALSTORAGE_KEY
 } from '$lib/constants';
+import { SvelteMap, SvelteSet } from 'svelte/reactivity';
+
+export interface ConversationTreeItem {
+	conversation: DatabaseConversation;
+	depth: number;
+}
 
 class ConversationsStore {
 	/**
@@ -300,15 +306,45 @@ class ConversationsStore {
 	 * Deletes a conversation and all its messages
 	 * @param convId - The conversation ID to delete
 	 */
-	async deleteConversation(convId: string): Promise<void> {
+	async deleteConversation(convId: string, options?: { deleteWithForks?: boolean }): Promise<void> {
 		try {
-			await DatabaseService.deleteConversation(convId);
+			await DatabaseService.deleteConversation(convId, options);
 
-			this.conversations = this.conversations.filter((c) => c.id !== convId);
+			if (options?.deleteWithForks) {
+				// Collect all descendants recursively
+				const idsToRemove = new SvelteSet([convId]);
+				const queue = [convId];
+				while (queue.length > 0) {
+					const parentId = queue.pop()!;
+					for (const c of this.conversations) {
+						if (c.forkedFromConversationId === parentId && !idsToRemove.has(c.id)) {
+							idsToRemove.add(c.id);
+							queue.push(c.id);
+						}
+					}
+				}
+				this.conversations = this.conversations.filter((c) => !idsToRemove.has(c.id));
 
-			if (this.activeConversation?.id === convId) {
-				this.clearActiveConversation();
-				await goto(`?new_chat=true#/`);
+				if (this.activeConversation && idsToRemove.has(this.activeConversation.id)) {
+					this.clearActiveConversation();
+					await goto(`?new_chat=true#/`);
+				}
+			} else {
+				// Reparent direct children to deleted conv's parent (or promote to top-level)
+				const deletedConv = this.conversations.find((c) => c.id === convId);
+				const newParent = deletedConv?.forkedFromConversationId;
+				this.conversations = this.conversations
+					.filter((c) => c.id !== convId)
+					.map((c) =>
+						c.forkedFromConversationId === convId
+							? { ...c, forkedFromConversationId: newParent }
+							: c
+					);
+
+				if (this.activeConversation?.id === convId) {
+					this.clearActiveConversation();
+					await goto(`?new_chat=true#/`);
+				}
 			}
 		} catch (error) {
 			console.error('Failed to delete conversation:', error);
@@ -658,6 +694,42 @@ class ConversationsStore {
 		this.saveMcpDefaults();
 	}
 
+	/**
+	 * Forks a conversation at a specific message, creating a new conversation
+	 * containing messages from root up to the target message, then navigates to it.
+	 *
+	 * @param messageId - The message ID to fork at
+	 * @param options - Fork options (name and whether to include attachments)
+	 * @returns The new conversation ID, or null if fork failed
+	 */
+	async forkConversation(
+		messageId: string,
+		options: { name: string; includeAttachments: boolean }
+	): Promise<string | null> {
+		if (!this.activeConversation) return null;
+
+		try {
+			const newConv = await DatabaseService.forkConversation(
+				this.activeConversation.id,
+				messageId,
+				options
+			);
+
+			this.conversations = [newConv, ...this.conversations];
+
+			await goto(`#/chat/${newConv.id}`);
+
+			toast.success('Conversation forked');
+
+			return newConv.id;
+		} catch (error) {
+			console.error('Failed to fork conversation:', error);
+			toast.error('Failed to fork conversation');
+
+			return null;
+		}
+	}
+
 	/**
 	 *
 	 *
@@ -830,3 +902,53 @@ export const conversations = () => conversationsStore.conversations;
 export const activeConversation = () => conversationsStore.activeConversation;
 export const activeMessages = () => conversationsStore.activeMessages;
 export const isConversationsInitialized = () => conversationsStore.isInitialized;
+
+/**
+ * Builds a flat tree of conversations with depth levels for nested forks.
+ * Accepts a pre-filtered list so search filtering stays in the component.
+ */
+export function buildConversationTree(convs: DatabaseConversation[]): ConversationTreeItem[] {
+	const childrenByParent = new SvelteMap<string, DatabaseConversation[]>();
+	const forkIds = new SvelteSet<string>();
+
+	for (const conv of convs) {
+		if (conv.forkedFromConversationId) {
+			forkIds.add(conv.id);
+
+			const siblings = childrenByParent.get(conv.forkedFromConversationId) || [];
+
+			siblings.push(conv);
+			childrenByParent.set(conv.forkedFromConversationId, siblings);
+		}
+	}
+
+	const result: ConversationTreeItem[] = [];
+	const visited = new SvelteSet<string>();
+
+	function walk(conv: DatabaseConversation, depth: number) {
+		visited.add(conv.id);
+		result.push({ conversation: conv, depth });
+
+		const children = childrenByParent.get(conv.id);
+		if (children) {
+			children.sort((a, b) => b.lastModified - a.lastModified);
+
+			for (const child of children) {
+				walk(child, depth + 1);
+			}
+		}
+	}
+
+	const roots = convs.filter((c) => !forkIds.has(c.id));
+	for (const root of roots) {
+		walk(root, 0);
+	}
+
+	for (const conv of convs) {
+		if (!visited.has(conv.id)) {
+			walk(conv, 1);
+		}
+	}
+
+	return result;
+}
diff --git a/tools/server/webui/src/lib/types/api.d.ts b/tools/server/webui/src/lib/types/api.d.ts
index f7f876c87..c1a023423 100644
--- a/tools/server/webui/src/lib/types/api.d.ts
+++ b/tools/server/webui/src/lib/types/api.d.ts
@@ -45,6 +45,7 @@ export interface ApiErrorResponse {
 export interface ApiChatMessageData {
 	role: ChatRole;
 	content: string | ApiChatMessageContentPart[];
+	reasoning_content?: string;
 	tool_calls?: ApiChatCompletionToolCall[];
 	tool_call_id?: string;
 	timestamp?: number;
@@ -201,6 +202,9 @@ export interface ApiChatCompletionRequest {
 	messages: Array<{
 		role: ChatRole;
 		content: string | ApiChatMessageContentPart[];
+		reasoning_content?: string;
+		tool_calls?: ApiChatCompletionToolCall[];
+		tool_call_id?: string;
 	}>;
 	stream?: boolean;
 	model?: string;
diff --git a/tools/server/webui/src/lib/types/database.d.ts b/tools/server/webui/src/lib/types/database.d.ts
index 50f51ecf5..95ff7377c 100644
--- a/tools/server/webui/src/lib/types/database.d.ts
+++ b/tools/server/webui/src/lib/types/database.d.ts
@@ -12,6 +12,7 @@ export interface DatabaseConversation {
 	lastModified: number;
 	name: string;
 	mcpServerOverrides?: McpServerOverride[];
+	forkedFromConversationId?: string;
 }
 
 export interface DatabaseMessageExtraAudioFile {
diff --git a/tools/server/webui/src/lib/types/settings.d.ts b/tools/server/webui/src/lib/types/settings.d.ts
index 360740ab0..4c545ce1d 100644
--- a/tools/server/webui/src/lib/types/settings.d.ts
+++ b/tools/server/webui/src/lib/types/settings.d.ts
@@ -24,6 +24,8 @@ export interface SettingsChatServiceOptions {
 	systemMessage?: string;
 	// Disable reasoning parsing (use 'none' instead of 'auto')
 	disableReasoningParsing?: boolean;
+	// Strip reasoning content from context before sending
+	excludeReasoningFromContext?: boolean;
 	tools?: OpenAIToolDefinition[];
 	// Generation parameters
 	temperature?: number;
diff --git a/tools/server/webui/src/routes/+layout.svelte b/tools/server/webui/src/routes/+layout.svelte
index 9093fc219..ccc207f85 100644
--- a/tools/server/webui/src/routes/+layout.svelte
+++ b/tools/server/webui/src/routes/+layout.svelte
@@ -4,7 +4,11 @@
 	import { browser } from '$app/environment';
 	import { page } from '$app/state';
 	import { untrack } from 'svelte';
-	import { ChatSidebar, DialogConversationTitleUpdate } from '$lib/components/app';
+	import {
+		ChatSidebar,
+		DialogConversationTitleUpdate,
+		DialogChatSettings
+	} from '$lib/components/app';
 	import { isLoading } from '$lib/stores/chat.svelte';
 	import { conversationsStore, activeMessages } from '$lib/stores/conversations.svelte';
 	import * as Sidebar from '$lib/components/ui/sidebar/index.js';
@@ -17,8 +21,10 @@
 	import { modelsStore } from '$lib/stores/models.svelte';
 	import { mcpStore } from '$lib/stores/mcp.svelte';
 	import { TOOLTIP_DELAY_DURATION } from '$lib/constants';
+	import type { SettingsSectionTitle } from '$lib/constants';
 	import { KeyboardKey } from '$lib/enums';
 	import { IsMobile } from '$lib/hooks/is-mobile.svelte';
+	import { setChatSettingsDialogContext } from '$lib/contexts';
 
 	let { children } = $props();
 
@@ -42,6 +48,16 @@
 	let titleUpdateNewTitle = $state('');
 	let titleUpdateResolve: ((value: boolean) => void) | null = null;
 
+	let chatSettingsDialogOpen = $state(false);
+	let chatSettingsDialogInitialSection = $state<SettingsSectionTitle | undefined>(undefined);
+
+	setChatSettingsDialogContext({
+		open: (initialSection?: SettingsSectionTitle) => {
+			chatSettingsDialogInitialSection = initialSection;
+			chatSettingsDialogOpen = true;
+		}
+	});
+
 	// Global keyboard shortcuts
 	function handleKeydown(event: KeyboardEvent) {
 		const isCtrlOrCmd = event.ctrlKey || event.metaKey;
@@ -213,6 +229,12 @@
 
 	<Toaster richColors />
 
+	<DialogChatSettings
+		open={chatSettingsDialogOpen}
+		onOpenChange={(open) => (chatSettingsDialogOpen = open)}
+		initialSection={chatSettingsDialogInitialSection}
+	/>
+
 	<DialogConversationTitleUpdate
 		bind:open={titleUpdateDialogOpen}
 		currentTitle={titleUpdateCurrentTitle}
diff --git a/tools/server/webui/tests/stories/ChatSidebar.stories.svelte b/tools/server/webui/tests/stories/ChatSidebar.stories.svelte
index 42cea8783..7ff9538a3 100644
--- a/tools/server/webui/tests/stories/ChatSidebar.stories.svelte
+++ b/tools/server/webui/tests/stories/ChatSidebar.stories.svelte
@@ -73,7 +73,7 @@
 			conversationsStore.conversations = mockConversations;
 		}, 0));
 		
-		const searchTrigger = screen.getByText('Search conversations');
+		const searchTrigger = screen.getByText('Search');
 		userEvent.click(searchTrigger);
 	}}
 >
diff --git a/tools/server/webui/tests/unit/reasoning-context.test.ts b/tools/server/webui/tests/unit/reasoning-context.test.ts
new file mode 100644
index 000000000..abbecf7e0
--- /dev/null
+++ b/tools/server/webui/tests/unit/reasoning-context.test.ts
@@ -0,0 +1,196 @@
+import { describe, it, expect } from 'vitest';
+import { AGENTIC_REGEX, REASONING_TAGS } from '$lib/constants/agentic';
+import { ContentPartType } from '$lib/enums';
+
+// Replicate ChatService.extractReasoningFromContent (private static)
+function extractReasoningFromContent(
+	content: string | Array<{ type: string; text?: string }> | null | undefined
+): string | undefined {
+	if (!content) return undefined;
+
+	const extractFromString = (text: string): string => {
+		const parts: string[] = [];
+		const re = new RegExp(AGENTIC_REGEX.REASONING_EXTRACT.source);
+		let match = re.exec(text);
+		while (match) {
+			parts.push(match[1]);
+			text = text.slice(match.index + match[0].length);
+			match = re.exec(text);
+		}
+		return parts.join('');
+	};
+
+	if (typeof content === 'string') {
+		const result = extractFromString(content);
+		return result || undefined;
+	}
+
+	if (!Array.isArray(content)) return undefined;
+
+	const parts: string[] = [];
+	for (const part of content) {
+		if (part.type === ContentPartType.TEXT && part.text) {
+			const result = extractFromString(part.text);
+			if (result) parts.push(result);
+		}
+	}
+	return parts.length > 0 ? parts.join('') : undefined;
+}
+
+// Replicate ChatService.stripReasoningContent (private static)
+function stripReasoningContent(
+	content: string | Array<{ type: string; text?: string }> | null | undefined
+): typeof content {
+	if (!content) return content;
+
+	if (typeof content === 'string') {
+		return content
+			.replace(AGENTIC_REGEX.REASONING_BLOCK, '')
+			.replace(AGENTIC_REGEX.REASONING_OPEN, '')
+			.replace(AGENTIC_REGEX.AGENTIC_TOOL_CALL_BLOCK, '')
+			.replace(AGENTIC_REGEX.AGENTIC_TOOL_CALL_OPEN, '');
+	}
+
+	if (!Array.isArray(content)) return content;
+
+	return content.map((part) => {
+		if (part.type !== ContentPartType.TEXT || !part.text) return part;
+		return {
+			...part,
+			text: part.text
+				.replace(AGENTIC_REGEX.REASONING_BLOCK, '')
+				.replace(AGENTIC_REGEX.REASONING_OPEN, '')
+				.replace(AGENTIC_REGEX.AGENTIC_TOOL_CALL_BLOCK, '')
+				.replace(AGENTIC_REGEX.AGENTIC_TOOL_CALL_OPEN, '')
+		};
+	});
+}
+
+// Simulate the message mapping logic from ChatService.sendMessage
+function buildApiMessage(
+	content: string,
+	excludeReasoningFromContext: boolean
+): { role: string; content: string; reasoning_content?: string } {
+	const cleaned = stripReasoningContent(content) as string;
+	const mapped: { role: string; content: string; reasoning_content?: string } = {
+		role: 'assistant',
+		content: cleaned
+	};
+	if (!excludeReasoningFromContext) {
+		const reasoning = extractReasoningFromContent(content);
+		if (reasoning) {
+			mapped.reasoning_content = reasoning;
+		}
+	}
+	return mapped;
+}
+
+// Helper: wrap reasoning the same way the chat store does during streaming
+function wrapReasoning(reasoning: string, content: string): string {
+	return `${REASONING_TAGS.START}${reasoning}${REASONING_TAGS.END}${content}`;
+}
+
+describe('reasoning content extraction', () => {
+	it('extracts reasoning from tagged string content', () => {
+		const input = wrapReasoning('step 1, step 2', 'The answer is 42.');
+		const result = extractReasoningFromContent(input);
+		expect(result).toBe('step 1, step 2');
+	});
+
+	it('returns undefined when no reasoning tags present', () => {
+		expect(extractReasoningFromContent('Just a normal response.')).toBeUndefined();
+	});
+
+	it('returns undefined for null/empty input', () => {
+		expect(extractReasoningFromContent(null)).toBeUndefined();
+		expect(extractReasoningFromContent(undefined)).toBeUndefined();
+		expect(extractReasoningFromContent('')).toBeUndefined();
+	});
+
+	it('extracts reasoning from content part arrays', () => {
+		const input = [
+			{
+				type: ContentPartType.TEXT,
+				text: wrapReasoning('thinking hard', 'result')
+			}
+		];
+		expect(extractReasoningFromContent(input)).toBe('thinking hard');
+	});
+
+	it('handles multiple reasoning blocks', () => {
+		const input =
+			REASONING_TAGS.START +
+			'block1' +
+			REASONING_TAGS.END +
+			'middle' +
+			REASONING_TAGS.START +
+			'block2' +
+			REASONING_TAGS.END +
+			'end';
+		expect(extractReasoningFromContent(input)).toBe('block1block2');
+	});
+
+	it('ignores non-text content parts', () => {
+		const input = [{ type: 'image_url', text: wrapReasoning('hidden', 'img') }];
+		expect(extractReasoningFromContent(input)).toBeUndefined();
+	});
+});
+
+describe('strip reasoning content', () => {
+	it('removes reasoning tags from string content', () => {
+		const input = wrapReasoning('internal thoughts', 'visible answer');
+		expect(stripReasoningContent(input)).toBe('visible answer');
+	});
+
+	it('removes reasoning from content part arrays', () => {
+		const input = [
+			{
+				type: ContentPartType.TEXT,
+				text: wrapReasoning('thoughts', 'answer')
+			}
+		];
+		const result = stripReasoningContent(input) as Array<{ type: string; text?: string }>;
+		expect(result[0].text).toBe('answer');
+	});
+});
+
+describe('API message building with reasoning preservation', () => {
+	const storedContent = wrapReasoning('Let me think: 2+2=4, basic arithmetic.', 'The answer is 4.');
+
+	it('preserves reasoning_content when excludeReasoningFromContext is false', () => {
+		const msg = buildApiMessage(storedContent, false);
+		expect(msg.content).toBe('The answer is 4.');
+		expect(msg.reasoning_content).toBe('Let me think: 2+2=4, basic arithmetic.');
+		// no internal tags leak into either field
+		expect(msg.content).not.toContain('<<<');
+		expect(msg.reasoning_content).not.toContain('<<<');
+	});
+
+	it('strips reasoning_content when excludeReasoningFromContext is true', () => {
+		const msg = buildApiMessage(storedContent, true);
+		expect(msg.content).toBe('The answer is 4.');
+		expect(msg.reasoning_content).toBeUndefined();
+	});
+
+	it('handles content with no reasoning in both modes', () => {
+		const plain = 'No reasoning here.';
+		const msgPreserve = buildApiMessage(plain, false);
+		const msgExclude = buildApiMessage(plain, true);
+		expect(msgPreserve.content).toBe(plain);
+		expect(msgPreserve.reasoning_content).toBeUndefined();
+		expect(msgExclude.content).toBe(plain);
+		expect(msgExclude.reasoning_content).toBeUndefined();
+	});
+
+	it('cleans agentic tool call blocks from content even when preserving reasoning', () => {
+		const input =
+			wrapReasoning('plan', 'text') +
+			'\n\n<<<AGENTIC_TOOL_CALL_START>>>\n' +
+			'<<<TOOL_NAME:bash>>>\n' +
+			'<<<TOOL_ARGS_START>>>\n{}\n<<<TOOL_ARGS_END>>>\nout\n' +
+			'<<<AGENTIC_TOOL_CALL_END>>>\n';
+		const msg = buildApiMessage(input, false);
+		expect(msg.content).not.toContain('<<<');
+		expect(msg.reasoning_content).toBe('plan');
+	});
+});
diff --git a/vendor/cpp-httplib/httplib.cpp b/vendor/cpp-httplib/httplib.cpp
index caa87abff..8ff1da57b 100644
--- a/vendor/cpp-httplib/httplib.cpp
+++ b/vendor/cpp-httplib/httplib.cpp
@@ -467,10 +467,6 @@ bool set_socket_opt_impl(socket_t sock, int level, int optname,
                     optlen) == 0;
 }
 
-bool set_socket_opt(socket_t sock, int level, int optname, int optval) {
-  return set_socket_opt_impl(sock, level, optname, &optval, sizeof(optval));
-}
-
 bool set_socket_opt_time(socket_t sock, int level, int optname,
                                 time_t sec, time_t usec) {
 #ifdef _WIN32
@@ -2218,7 +2214,7 @@ socket_t create_socket(const std::string &host, const std::string &ip, int port,
 #ifdef _WIN32
       // Setting SO_REUSEADDR seems not to work well with AF_UNIX on windows, so
       // remove the option.
-      detail::set_socket_opt(sock, SOL_SOCKET, SO_REUSEADDR, 0);
+      set_socket_opt(sock, SOL_SOCKET, SO_REUSEADDR, 0);
 #endif
 
       bool dummy;
@@ -4373,6 +4369,7 @@ make_multipart_content_provider(const UploadFormDataItems &items,
   struct MultipartState {
     std::vector<std::string> owned;
     std::vector<MultipartSegment> segs;
+    std::vector<char> buf = std::vector<char>(CPPHTTPLIB_SEND_BUFSIZ);
   };
   auto state = std::make_shared<MultipartState>();
   state->owned = std::move(owned);
@@ -4381,19 +4378,49 @@ make_multipart_content_provider(const UploadFormDataItems &items,
   state->segs = std::move(segs);
 
   return [state](size_t offset, size_t length, DataSink &sink) -> bool {
+    // Buffer multiple small segments into fewer, larger writes to avoid
+    // excessive TCP packets when there are many form data items (#2410)
+    auto &buf = state->buf;
+    auto buf_size = buf.size();
+    size_t buf_len = 0;
+    size_t remaining = length;
+
+    // Find the first segment containing 'offset'
     size_t pos = 0;
-    for (const auto &seg : state->segs) {
-      // Loop invariant: pos <= offset (proven by advancing pos only when
-      // offset - pos >= seg.size, i.e., the segment doesn't contain offset)
-      if (seg.size > 0 && offset - pos < seg.size) {
-        size_t seg_offset = offset - pos;
-        size_t available = seg.size - seg_offset;
-        size_t to_write = (std::min)(available, length);
-        return sink.write(seg.data + seg_offset, to_write);
-      }
+    size_t seg_idx = 0;
+    for (; seg_idx < state->segs.size(); seg_idx++) {
+      const auto &seg = state->segs[seg_idx];
+      if (seg.size > 0 && offset - pos < seg.size) { break; }
       pos += seg.size;
     }
-    return true; // past end (shouldn't be reached when content_length is exact)
+
+    size_t seg_offset = (seg_idx < state->segs.size()) ? offset - pos : 0;
+
+    for (; seg_idx < state->segs.size() && remaining > 0; seg_idx++) {
+      const auto &seg = state->segs[seg_idx];
+      size_t available = seg.size - seg_offset;
+      size_t to_copy = (std::min)(available, remaining);
+      const char *src = seg.data + seg_offset;
+      seg_offset = 0; // only the first segment has a non-zero offset
+
+      while (to_copy > 0) {
+        size_t space = buf_size - buf_len;
+        size_t chunk = (std::min)(to_copy, space);
+        std::memcpy(buf.data() + buf_len, src, chunk);
+        buf_len += chunk;
+        src += chunk;
+        to_copy -= chunk;
+        remaining -= chunk;
+
+        if (buf_len == buf_size) {
+          if (!sink.write(buf.data(), buf_len)) { return false; }
+          buf_len = 0;
+        }
+      }
+    }
+
+    if (buf_len > 0) { return sink.write(buf.data(), buf_len); }
+    return true;
   };
 }
 
@@ -5264,13 +5291,18 @@ bool setup_client_tls_session(const std::string &host, tls::ctx_t &ctx,
  */
 
 void default_socket_options(socket_t sock) {
-  detail::set_socket_opt(sock, SOL_SOCKET,
+  set_socket_opt(sock, SOL_SOCKET,
 #ifdef SO_REUSEPORT
-                         SO_REUSEPORT,
+                 SO_REUSEPORT,
 #else
-                         SO_REUSEADDR,
+                 SO_REUSEADDR,
 #endif
-                         1);
+                 1);
+}
+
+bool set_socket_opt(socket_t sock, int level, int optname, int optval) {
+  return detail::set_socket_opt_impl(sock, level, optname, &optval,
+                                     sizeof(optval));
 }
 
 std::string get_bearer_token_auth(const Request &req) {
@@ -7418,6 +7450,8 @@ bool Server::read_content_core(
     return false;
   }
 
+  req.body_consumed_ = true;
+
   if (req.is_multipart_form_data()) {
     if (!multipart_form_data_parser.is_valid()) {
       res.status = StatusCode::BadRequest_400;
@@ -7688,9 +7722,7 @@ bool Server::listen_internal() {
       detail::set_socket_opt_time(sock, SOL_SOCKET, SO_SNDTIMEO,
                                   write_timeout_sec_, write_timeout_usec_);
 
-      if (tcp_nodelay_) {
-        detail::set_socket_opt(sock, IPPROTO_TCP, TCP_NODELAY, 1);
-      }
+      if (tcp_nodelay_) { set_socket_opt(sock, IPPROTO_TCP, TCP_NODELAY, 1); }
 
       if (!task_queue->enqueue(
               [this, sock]() { process_and_close_socket(sock); })) {
@@ -8036,8 +8068,19 @@ Server::process_request(Stream &strm, const std::string &remote_addr,
     return write_response(strm, close_connection, req, res);
   }
 
+  // RFC 9112 §6.3: Reject requests with both a non-zero Content-Length and
+  // any Transfer-Encoding to prevent request smuggling. Content-Length: 0 is
+  // tolerated for compatibility with existing clients.
+  if (req.get_header_value_u64("Content-Length") > 0 &&
+      req.has_header("Transfer-Encoding")) {
+    connection_closed = true;
+    res.status = StatusCode::BadRequest_400;
+    return write_response(strm, close_connection, req, res);
+  }
+
   // Check if the request URI doesn't exceed the limit
   if (req.target.size() > CPPHTTPLIB_REQUEST_URI_MAX_LENGTH) {
+    connection_closed = true;
     res.status = StatusCode::UriTooLong_414;
     output_error_log(Error::ExceedUriMaxLength, &req);
     return write_response(strm, close_connection, req, res);
@@ -8066,6 +8109,7 @@ Server::process_request(Stream &strm, const std::string &remote_addr,
   if (req.has_header("Accept")) {
     const auto &accept_header = req.get_header_value("Accept");
     if (!detail::parse_accept_header(accept_header, req.accept_content_types)) {
+      connection_closed = true;
       res.status = StatusCode::BadRequest_400;
       output_error_log(Error::HTTPParsing, &req);
       return write_response(strm, close_connection, req, res);
@@ -8075,6 +8119,7 @@ Server::process_request(Stream &strm, const std::string &remote_addr,
   if (req.has_header("Range")) {
     const auto &range_header_value = req.get_header_value("Range");
     if (!detail::parse_range_header(range_header_value, req.ranges)) {
+      connection_closed = true;
       res.status = StatusCode::RangeNotSatisfiable_416;
       output_error_log(Error::InvalidRangeHeader, &req);
       return write_response(strm, close_connection, req, res);
@@ -8202,6 +8247,7 @@ Server::process_request(Stream &strm, const std::string &remote_addr,
     }
   }
 #endif
+  auto ret = false;
   if (routed) {
     if (res.status == -1) {
       res.status = req.ranges.empty() ? StatusCode::OK_200
@@ -8209,6 +8255,7 @@ Server::process_request(Stream &strm, const std::string &remote_addr,
     }
 
     // Serve file content by using a content provider
+    auto file_open_error = false;
     if (!res.file_content_path_.empty()) {
       const auto &path = res.file_content_path_;
       auto mm = std::make_shared<detail::mmap>(path.c_str());
@@ -8218,37 +8265,53 @@ Server::process_request(Stream &strm, const std::string &remote_addr,
         res.content_provider_ = nullptr;
         res.status = StatusCode::NotFound_404;
         output_error_log(Error::OpenFile, &req);
-        return write_response(strm, close_connection, req, res);
-      }
+        file_open_error = true;
+      } else {
+        auto content_type = res.file_content_content_type_;
+        if (content_type.empty()) {
+          content_type = detail::find_content_type(
+              path, file_extension_and_mimetype_map_, default_file_mimetype_);
+        }
 
-      auto content_type = res.file_content_content_type_;
-      if (content_type.empty()) {
-        content_type = detail::find_content_type(
-            path, file_extension_and_mimetype_map_, default_file_mimetype_);
+        res.set_content_provider(
+            mm->size(), content_type,
+            [mm](size_t offset, size_t length, DataSink &sink) -> bool {
+              sink.write(mm->data() + offset, length);
+              return true;
+            });
       }
-
-      res.set_content_provider(
-          mm->size(), content_type,
-          [mm](size_t offset, size_t length, DataSink &sink) -> bool {
-            sink.write(mm->data() + offset, length);
-            return true;
-          });
     }
 
-    if (detail::range_error(req, res)) {
+    if (file_open_error) {
+      ret = write_response(strm, close_connection, req, res);
+    } else if (detail::range_error(req, res)) {
       res.body.clear();
       res.content_length_ = 0;
       res.content_provider_ = nullptr;
       res.status = StatusCode::RangeNotSatisfiable_416;
-      return write_response(strm, close_connection, req, res);
+      ret = write_response(strm, close_connection, req, res);
+    } else {
+      ret = write_response_with_content(strm, close_connection, req, res);
     }
-
-    return write_response_with_content(strm, close_connection, req, res);
   } else {
     if (res.status == -1) { res.status = StatusCode::NotFound_404; }
-
-    return write_response(strm, close_connection, req, res);
+    ret = write_response(strm, close_connection, req, res);
   }
+
+  // Drain any unconsumed request body to prevent request smuggling on
+  // keep-alive connections.
+  if (!req.body_consumed_ && detail::expect_content(req)) {
+    int drain_status = 200; // required by read_content signature
+    if (!detail::read_content(
+            strm, req, payload_max_length_, drain_status, nullptr,
+            [](const char *, size_t, size_t, size_t) { return true; }, false)) {
+      // Body exceeds payload limit or read error — close the connection
+      // to prevent leftover bytes from being misinterpreted.
+      connection_closed = true;
+    }
+  }
+
+  return ret;
 }
 
 bool Server::is_valid() const { return true; }
diff --git a/vendor/cpp-httplib/httplib.h b/vendor/cpp-httplib/httplib.h
index ce1681fcb..2967ddf5e 100644
--- a/vendor/cpp-httplib/httplib.h
+++ b/vendor/cpp-httplib/httplib.h
@@ -8,8 +8,8 @@
 #ifndef CPPHTTPLIB_HTTPLIB_H
 #define CPPHTTPLIB_HTTPLIB_H
 
-#define CPPHTTPLIB_VERSION "0.39.0"
-#define CPPHTTPLIB_VERSION_NUM "0x002700"
+#define CPPHTTPLIB_VERSION "0.40.0"
+#define CPPHTTPLIB_VERSION_NUM "0x002800"
 
 #ifdef _WIN32
 #if defined(_WIN32_WINNT) && _WIN32_WINNT < 0x0A00
@@ -1266,6 +1266,7 @@ struct Request {
   bool is_multipart_form_data() const;
 
   // private members...
+  bool body_consumed_ = false;
   size_t redirect_count_ = CPPHTTPLIB_REDIRECT_MAX_COUNT;
   size_t content_length_ = 0;
   ContentProvider content_provider_;
@@ -1475,6 +1476,8 @@ using SocketOptions = std::function<void(socket_t sock)>;
 
 void default_socket_options(socket_t sock);
 
+bool set_socket_opt(socket_t sock, int level, int optname, int optval);
+
 const char *status_message(int status);
 
 std::string to_string(Error error);
@@ -1564,6 +1567,13 @@ ssize_t write_headers(Stream &strm, const Headers &headers);
 bool set_socket_opt_time(socket_t sock, int level, int optname, time_t sec,
                          time_t usec);
 
+size_t get_multipart_content_length(const UploadFormDataItems &items,
+                                    const std::string &boundary);
+
+ContentProvider
+make_multipart_content_provider(const UploadFormDataItems &items,
+                                const std::string &boundary);
+
 } // namespace detail
 
 class Server {