llmfit serve gains --unix-socket <path> (conflicts with --host/--port):
axum serves the identical router over a tokio UnixListener. The socket is
created 0660 and a stale file from a previous instance is replaced, so
restarts over persistent paths (pod volumes) work. Intended for same-host
consumers where a TCP port is undesirable — e.g. the llmfit-dra DRA
driver's sidecar in a hostNetwork pod, where 127.0.0.1 is the NODE's
loopback.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Adds two new capabilities to `llmfit serve`:
- `--mcp` flag runs an MCP server over stdio (JSON-RPC 2.0) exposing 6 tools
for agent discovery: get_system_specs, recommend_models, search_models,
plan_hardware, get_runtimes, get_installed_models
- `--send-events` flag (behind `nats` feature) publishes hardware/model events
to NATS on subject hierarchy `llmfit.{event_type}.{hostname}` for integration
with coordination systems like Sympozium
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add interactive hardware simulation to the TUI (press S) — override RAM,
VRAM, and CPU cores to instantly see which models fit on target hardware.
A SIM badge in the status bar and system bar indicates when simulation is
active. All model scores, fit levels, and speed estimates are recalculated
live against the simulated specs.
Also adds --ram and --cpu-cores global CLI flags (alongside the existing
--memory for GPU VRAM), threads all overrides through every subcommand
and llmfit serve, and consolidates the duplicated detect_specs function.
Closes#322
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- add React/Vite llmfit-web dashboard with fit explorer and theme toggle
- embed built web assets into llmfit binary and serve from / with SPA fallback
- keep existing /api/v1 contracts and add CI/release build steps for web assets
- auto-launch local dashboard for non-serve commands while preserving explicit serve mode
On Apple Silicon, llmfit forces non-quantized models to MLX, so
--runtime llamacpp returns 0 results. The new --force-runtime flag
overrides runtime selection during analysis so users can get
llama.cpp (or other runtime) recommendations on any platform.
Supported in CLI recommend command and REST API query parameter.
Bumps version to 0.7.6.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
