Adds a server-side "small model" capability: direct, cheap LLM calls that
reuse the user's existing OpenCode provider logins — the mechanism OpenCode
uses internally for titles and summaries but does not expose through the
SDK or plugins. Zero new dependencies; plain fetch with per-provider wire
formats, credentials never leave the server.
Core (packages/web/server/lib/small-model):
- Resolution mirrors OpenCode's session scoping: explicit settings override
→ small_model from the OpenCode config → family scan within the session's
provider → the session's own model. The global provider scan only serves
callers without a session context, and background callers forbid it
entirely (restrictToPreferredProvider), so conversation content never
reaches a provider the user didn't pick — explicit choices excepted.
- Per-provider auth replicating OpenCode's plugin loaders: GitHub Copilot
(device token as bearer, no exchange), ChatGPT plan via the codex
Responses API (single-flight OAuth refresh written back to auth.json),
Anthropic messages, Google generateContent, generic OpenAI-compatible.
- OpenCode's free models (opencode/big-pickle, *-free) are never called
directly; unauthenticated providers are skipped by design.
- Prompt clamping to the model's catalog context limit; thinking disabled
where a wire switch exists (Z.AI/GLM, MiniMax-M3, Gemini Flash); robust
content parsing with a clear error when a thinking model spends its whol
budget on reasoning.
- Settings → Sessions gains a Small Model group: use-default checkbox plus
an override picker limited to authenticated providers, persisted with
web/desktop/VS Code sanitization parity.
Consumers:
- Session assist: a server-side watcher on the global SSE hub generates a
short recap and one suggested follow-up after a session idles quietly fo
a minute, stored on session metadata (openchamber.assist). Freshness is
keyed to the last assistant message id, so new activity invalidates the
payload everywhere with no extra writes. The chat shows the recap under
the last message after five quiet minutes and the suggestion as a
dismissible chip above the composer (tap fills the input, never sends).
Gated by a new Chat setting (default on) that is a hard generation
switch. Language is anchored to the conversation itself, with a
script-mismatch guard against model/backend language hallucination.
- TTS: a third input mode, summarized — long replies are condensed to
spoken prose before playback on any TTS engine.
- Git: commit-message and PR generation moved off the active chat session
onto the small model fed with real diffs and the commit list (bodies
included), with a session-transport fallback for free-model-only setups.
- Notes: Add to notes distills long selections into 1-3 dense sentences
preserving exact identifiers, with verbatim fallback on failure.
Fixes along the way:
- The global event watcher now starts unconditionally; it was gated behind
the desktop-notify env, leaving the server-side event hub dead in
packaged apps.
- OpenCode re-emits message.updated for old user messages after idle; the
watcher no longer mistakes those for new activity.
- Session metadata merges from a fresh read right before the PATCH, so
writes made during the generation window (suggestion dismissals, review
links) are preserved; the assist runtime stops during graceful shutdown.
openchamber --port 8080# Custom port
openchamber --lan --port 3000# Listen on LAN (0.0.0.0)
openchamber --ui-password secret # Password-protect UI
openchamber startup enable# Start at login as a native serviceOPENCHAMBER_UI_PASSWORD=secret openchamber startup enable# Save service password env
openchamber startup status # Show startup service status
openchamber startup disable # Remove startup service
openchamber tunnel help# Tunnel lifecycle commands
openchamber tunnel providers # Show provider capabilities
openchamber tunnel profile add --provider cloudflare --mode managed-remote --name prod-main --hostname app.example.com --token <token>
openchamber tunnel start --profile prod-main
openchamber tunnel start --provider cloudflare --mode quick --qr
openchamber tunnel start --provider cloudflare --mode managed-local --config ~/.cloudflared/config.yml
openchamber tunnel status --all # Show tunnel state across instances
openchamber tunnel stop --port 3000# Stop tunnel only (server stays running)
openchamber connect-url --port 3000# Add this server to OpenChamber Desktop
openchamber connect-url --server http://host:3000 --qr
openchamber connect-url --port 3000 --qr
openchamber logs # Follow latest instance logsOPENCODE_PORT=4096OPENCODE_SKIP_START=true openchamber # Connect to external OpenCode serverOPENCODE_HOST=https://myhost:4096 OPENCODE_SKIP_START=true openchamber # Connect via custom host/HTTPS
openchamber stop # Stop server
openchamber update # Update to latest
startup enable snapshots your current environment into the native service so startup behaves like you launched openchamber from the same shell. This preserves provider tokens, PATH, SSH agent settings, and other CLI auth/config env vars. Use --no-env-snapshot if you want a minimal service env.
This runs OpenChamber as an API-only server without the desktop app or browser UI assets on that machine, then creates a link for Desktop to import. --lan makes the server reachable from other machines. --server is the address Desktop should use.
When OpenChamber was started with --lan or --host 0.0.0.0, connect-url automatically uses a detected LAN IP instead of 127.0.0.1. Use --server http://host:3000 to override the advertised address, and include --lan when connect-url needs to start the server for LAN access.
Paste the printed openchamber://connect?... link in Desktop under Settings -> Remote Instances -> Direct Instances -> Import Link. The link contains the server URL and a client token. It does not enable browser UI password protection; use --ui-password when exposing a server beyond localhost.
systemd service (VPN / LAN access)
Run OpenChamber and OpenCode as separate persistent services — useful when you want to access your
dev machine over a VPN (e.g. Tailscale) or LAN without a Cloudflare tunnel.
How it works:
OpenCode runs as its own service, binding only to localhost.
OpenChamber connects to it via OPENCODE_HOST and --lan makes it reachable on your VPN IP.
--foreground keeps the CLI process alive so systemd can track and restart it.
Why set PATH and SSH_AUTH_SOCK?
systemd user services start with a minimal environment — no shell profile is sourced.
Without an explicit PATH, OpenCode won't find tools installed via Homebrew, npm, or ~/.local/bin.
Without SSH_AUTH_SOCK, git operations over SSH (push, pull, clone) will fail because the agent socket isn't inherited.
Adjust the PATH to match your own tool installation paths.
%t expands to $XDG_RUNTIME_DIR (e.g. /run/user/1000), where most SSH agents write their socket.
OpenChamber will be reachable at http://<your-vpn-hostname>:3000 from any device on your VPN.
Note:--host 0.0.0.0 is required to listen on all interfaces. The default
bind address is 127.0.0.1 (localhost only). Use --host <ip> or
OPENCHAMBER_HOST=<ip> to bind to a specific interface instead.
Managed-local path note: OPENCHAMBER_TUNNEL_CONFIG must point to a path inside the container user home (/home/openchamber/...). If your Cloudflare config references a credentials JSON file, that file path must also be accessible inside the container (mount with volumes).
Website docs source lives at packages/docs/content/docs/reverse-proxy.mdx.
Tunnel behavior notes
OpenChamber supports one active tunnel per running instance (port).
Starting a tunnel with a different mode/provider on the same instance replaces the current tunnel.
Replacing or stopping a tunnel revokes existing connect links and invalidates remote tunnel sessions for that instance.
Connect links are one-time tokens; generating a new link revokes the previous unused link.
Data Directory Permission Note: The data/ directory is mounted into the container for persistent storage (config, sessions, SSH keys, workspaces). Before running, ensure the directory exists and has proper permissions:
My wife, who - with zero AI background - sat down with the app for the first time and built the firework celebration that plays on every successful push.
Every contributor who shaped this project with their PRs, ideas, and attention to detail.
Bohdan Triapitsyn
28f0736d69
