docs: add Connection Profile docs, new screenshots, and minor fixes

- Document Connection Profiles as 4th extraction source in providers.md
  with creation steps, settings table, and usage instructions
- Add screenshots: Settings Modal tabs (connection, profile, extraction,
  storage), Connection Profile toolbar bar and creation dialog
- Refresh wizard-step1.png and panel-full.png screenshots
- Document "Protect recent messages" feature in managing-memories.md
- Add Connection Profile mentions to getting-started.md, README.md,
  and architecture.md
- Bump manifest.json version to 2.1.9
- Fix tooltip "Injection Sidebar" → "Injection Viewer" in settings.html
- Add issue #10 per-chat memory storage backlog analysis

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

README.md

@@ -47,6 +47,7 @@ CharMemory gives your characters persistent memory across chats. Here's the shor
 - **Prompt Breakdown** — see exactly where your context window is going. The injection viewer shows a per-category token breakdown (System, Char card, Lorebook, Data Bank, Examples, Chat history) with a stacked bar and actionable tips for reducing usage.
 - **Vector Storage health checks** — a traffic-light indicator flags misconfigured settings and tells you what to fix, so you're not guessing why memories aren't showing up.
 - **Full memory control** — browse, edit, or delete individual memories with a unified block editor (inline editing, undo, find/replace). Consolidate duplicates with preview and undo. Batch-extract from all chats at once.
+- **Flexible LLM connections** — use a dedicated API provider, reuse an existing SillyTavern Connection Profile, run a local model via WebLLM, or piggyback on your main chat LLM.
 - **Highly configurable extraction prompts** — separate prompts for 1:1 and group chats and memory file consolidation and conversion.
 - **Guided setup** — the Setup Wizard tests your LLM connection, checks Vector Storage, and handles existing memory file conversion in about 2 minutes.
 - **Tablet & phone support** — on touch devices, the dashboard opens as a floating panel with touch-friendly controls. Phone layout widens drawers for small screens. If auto-detection doesn't match your device, override it in Settings > Advanced > Display Mode. Please [report issues on GitHub](https://github.com/bal-spec/sillytavern-character-memory/issues) with how it behaves on your device.

docs/architecture.md

@@ -118,7 +118,7 @@ New message arrives
   → At generation time, VS retrieves relevant chunks and injects them into the prompt
 ```
 
-`callLLM()` (line ~2265) is the **single dispatch point** for all LLM calls — extraction, consolidation, and conversion all go through it. It branches on `extension_settings.charMemory.source` to route to the Main LLM, WebLLM, or a dedicated provider.
+`callLLM()` (line ~2265) is the **single dispatch point** for all LLM calls — extraction, consolidation, and conversion all go through it. It branches on `extension_settings.charMemory.source` to route to the Main LLM, WebLLM, a dedicated provider, or a Connection Profile (via `generateProfileResponse()` which uses ST's `ConnectionManagerRequestService`).
 
 ---
 

docs/getting-started.md

@@ -26,6 +26,8 @@ CharMemory needs its own LLM connection — separate from your main chat. This k
 
 > **Running an LLM locally?** Select **Local Server** from the provider dropdown. Enter your server URL (e.g., `http://localhost:11434/v1` for Ollama). No API key needed. See [Providers → Local Servers](providers.md#local-servers) for port numbers by backend.
 
+> **Already have a connection configured in SillyTavern?** You can skip the wizard's provider setup and use a **Connection Profile** instead. After completing the wizard, open **Settings** (gear icon) → **Connection** → change **LLM Used for Extraction** to **Connection Profile** and select your saved profile. See [Providers → Connection Profiles](providers.md#connection-profiles).
+
 > **If your provider is not listed** Many providers have an OpenAI compatible API endpoint. See if you can configure it that way.
 
 ![Wizard step 1 after successful connection — green checkmark, model selected](../images/wizard-step1-connected.png)

docs/managing-memories.md

@@ -45,6 +45,12 @@ CharMemory extracts automatically while you chat. The **Auto** pill in the panel
 
 Both settings only affect automatic extraction. Extract Now, Extract Here, and Batch always run immediately regardless of interval or cooldown.
 
+### Protect recent messages
+
+When enabled (Settings → Extraction), this excludes the most recent messages from auto-extraction. This prevents the character's next response from being constrained by memories that were just extracted from the conversation in progress — swipes and regenerations stay natural because the extraction hasn't "locked in" those messages yet. Skipped messages are picked up on the next extraction cycle.
+
+![Settings Modal — Extraction tab showing Protect Recent Messages and other settings](../images/settings-modal-extraction.png)
+
 ---
 
 ## Extract Now and Extract Here

docs/plans/2026-03-18-issue-10-per-chat-backlog.md

@@ -0,0 +1,82 @@
+# Issue #10: Per-Chat Memory Storage — Backlog Analysis
+
+**Date:** 2026-03-18
+**Status:** Backlog
+**Issue:** https://github.com/bal-spec/sillytavern-character-memory/issues/10
+**Prior design:** `2026-03-06-per-chat-isolation-revised.md`
+
+---
+
+## Summary
+
+User request: memories should be stored per-chat rather than per-character, so different storylines/chats don't share memories.
+
+The extension author's preference is character-level memories (the current default), but some users want per-chat isolation. The solution should support both as an optional setting.
+
+## Proposed UX
+
+Single dropdown in Settings > Storage:
+
+```
+Memory scope: [Character (default)] / [Chat]
+```
+
+- **Character** (default): Current behavior. Memories stored in Character Data Bank, shared across all chats, VS handles retrieval.
+- **Chat**: Memories stored in Chat Attachments (`chat_metadata.attachments`), scoped to the active chat only. VS retrieves only from that chat's attachments.
+
+## Architecture: Chat Attachments vs chat_metadata
+
+Two storage options were analyzed:
+
+### Option A: Chat Attachments (Recommended for simplicity)
+
+Use ST's existing `chat_metadata.attachments` system — the same API as Character Attachments but scoped per-chat. VS already knows how to vectorize and retrieve from chat attachments.
+
+**Pros:** Reuses existing ST infrastructure, visible in Data Bank UI, VS retrieval works automatically.
+**Cons:** Cross-chat bleed still possible if VS retrieves from both character AND chat attachments simultaneously.
+
+### Option B: chat_metadata direct storage + setExtensionPrompt injection
+
+Store memories in `chat_metadata.charMemory.isolatedMemories` and inject via `setExtensionPrompt()`, bypassing VS entirely. Full design in `2026-03-06-per-chat-isolation-revised.md`.
+
+**Pros:** Complete isolation guaranteed, no VS dependency.
+**Cons:** Loses semantic retrieval, requires reimplementing injection, more code surface area.
+
+## Risk Analysis
+
+### Code Impact
+
+48 references to the character attachment system (`extension_settings.character_attachments`, `getDataBankAttachments`, `writeMemoriesForCharacter`, etc.) would need branching to support chat-scoped storage. Key functions:
+
+- `getMemoryFileName()` — currently uses character avatar; would need chat-scoped variant
+- `writeMemoriesForCharacter()` — writes to Character Data Bank; needs Chat Data Bank path
+- `readMemoriesForCharacter()` — reads from Character Data Bank; needs Chat Data Bank path
+- All Troubleshooter file operations (view, edit, delete)
+- Memory Manager, Consolidation, Format/Convert tools
+
+### Known Bug: Batch Extraction in Per-Chat Mode
+
+`getMemoryFileName()` uses `context.chatId` which is always the *active* chat's ID. Batch extraction iterates over multiple chats but the file name function doesn't accept a chat ID parameter. This means batch-extracted memories from non-active chats get written to the wrong file.
+
+**Recommendation:** Disable batch extraction when chat-scoped mode is active, or fix `getMemoryFileName()` to accept an explicit chat ID parameter.
+
+### Group Chats
+
+Group chats add complexity — each character in the group has separate memories, but `chat_metadata` is shared across all group members. Data structure would need per-character keying within the chat metadata.
+
+**Recommendation:** Scope initial implementation to 1:1 chats only. Show a warning if chat mode is enabled in a group chat.
+
+## Testing Strategy
+
+- **Automated (Playwright):** Storage mode switching, verify memories write to correct location (character vs chat attachments), verify isolation (chat A memories don't appear in chat B)
+- **Manual:** Full extraction cycle in chat mode, verify VS retrieval scoping, test mode switching with existing memories
+
+## Implementation Priority
+
+Low-medium. The current character-level default works well for most users. This is an opt-in feature for users with specific multi-storyline use cases. No timeline set.
+
+## Dependencies
+
+- Fix `getMemoryFileName()` chat ID bug (regardless of this feature)
+- Understand VS retrieval scoping for chat attachments vs character attachments
+- Decide on migration path for users switching between modes

docs/providers.md

@@ -6,14 +6,17 @@ CharMemory works better when it uses its own LLM connection — separate from yo
 
 ## Extraction sources
 
-Three options for **LLM Used for Extraction** in Settings → Connection:
+Four options for **LLM Used for Extraction** in Settings → Connection:
 
 | Source | How it works | Best for |
 |--------|-------------|----------|
 | **Dedicated API** (default) | Direct API call with only the extraction prompt | Best quality — recommended |
+| **Connection Profile** | Reuses a saved SillyTavern connection profile | Already have a profile configured in ST |
 | **WebLLM** | Small model running locally in your browser | Privacy and no API cost; limited quality |
 | **Main LLM** | Uses whatever LLM your chat is using | Not recommended; No extra setup but quality suffers |
 
+![Settings Modal — Connection tab with source dropdown](../images/settings-modal-connection.png)
+
 **Dedicated API** is the default and recommended option. The extraction prompt is the only thing sent — the character card is included as bounded reference context (so the LLM knows what *not* to re-extract), but no chat system prompts, personas, or jailbreaks contaminate the call.
 
 ---
@@ -47,6 +50,56 @@ Open Settings → Connection → choose a **Provider**:
 
 ---
 
+## Connection Profiles
+
+If you already have a connection configured in SillyTavern's **Connection Manager**, you can reuse it for extraction instead of setting up a separate Dedicated API connection.
+
+### What is a Connection Profile?
+
+A Connection Profile is a SillyTavern feature that saves a snapshot of your current API connection settings — so you can switch between configurations quickly. A profile can include any combination of:
+
+| Setting | Description |
+|---------|-------------|
+| **API** | Which API provider (e.g., NanoGPT, OpenAI, OpenRouter) |
+| **Settings Preset** | The sampler/generation preset |
+| **Model** | The specific model to use |
+| **Proxy Preset** | Proxy configuration (if any) |
+| **Custom Stopping Strings** | Custom stop sequences |
+| **Start Reply With** | Prefill text for responses |
+| **Reasoning Template** | Reasoning/thinking format (e.g., DeepSeek) |
+| **Prompt Post-Processing** | Any prompt post-processing rules |
+| **Secret** | Your API key (stored securely) |
+
+Each setting has a checkbox — you choose which parts to include. When a profile is loaded, only the checked settings are applied.
+
+![Connection Profile toolbar — dropdown with info, create, save, edit, reload, delete icons](../images/connection-profile-bar.png)
+
+### Creating a profile in SillyTavern
+
+1. Open the **API Connections** panel (plug icon in SillyTavern's top bar)
+2. Configure your API: select a **Chat Completion Source**, enter your **API key**, select a **model**, and verify the connection shows **Valid**
+3. Click the **create icon** (file with plus) in the Connection Profile toolbar
+4. Check the settings you want to include in the profile — at minimum, check **API**, **Model**, and **Secret**
+5. Give it a name and click **Save**
+
+![Creating a Connection Profile — select which settings to include](../images/connection-profile-create.png)
+
+### Using a profile for CharMemory extraction
+
+![Settings Modal — Connection Profile source](../images/settings-modal-profile.png)
+
+1. In the CharMemory panel, click the **gear icon** to open Settings
+2. Under **Connection**, change **LLM Used for Extraction** to **Connection Profile**
+3. Choose your profile from the **Connection Profile** dropdown
+4. Click **Test Connection** to verify it works for extraction
+5. Optionally set a **System prompt** override — if left blank, CharMemory uses its default extraction system prompt
+
+Connection Profiles use whatever credentials, model, and endpoint are configured in the profile. You don't need to enter an API key or select a model separately — it's all inherited from the profile.
+
+> **When to use this vs. Dedicated API:** If you've already configured a connection in SillyTavern and want to reuse it without duplicating API keys, Connection Profile is convenient. If you want a completely separate LLM for extraction (different model, different provider), Dedicated API gives you full control.
+
+---
+
 ## Recommended models
 
 Memory extraction requires strong instruction following — the LLM must respect the extraction rules, stay within boundaries, and produce well-formatted output.

manifest.json

@@ -6,7 +6,7 @@
     "js": "index.js",
     "css": "style.css",
     "author": "bal-spec",
-    "version": "2.1.8",
+    "version": "2.1.9",
     "homePage": "",
     "auto_update": false
 }

settings.html

@@ -11,7 +11,7 @@
             <span id="charMemory_openSettingsModal" class="charMemory_headerGear" title="Open Settings">
                 <i class="fa-solid fa-gear fa-sm"></i>
             </span>
-            <span id="charMemory_toggleInjectionBtn" class="charMemory_headerGear" title="Toggle Injection Sidebar">
+            <span id="charMemory_toggleInjectionBtn" class="charMemory_headerGear" title="Toggle Injection Viewer">
                 <i class="fa-solid fa-syringe fa-sm"></i>
             </span>
             <div class="inline-drawer-icon fa-solid fa-circle-chevron-down down"></div>