Documentation Index
July 20, 2026 ยท View on GitHub
This folder is intentionally split into small, single-purpose files so humans and coding agents can load only the topic they need. It covers both provider-agnostic proxy behavior and provider-specific notes for GitHub Copilot, Azure OpenAI, OpenAI Codex, and generic compatible providers.
Doc Map
| File | Scope | Update When |
|---|---|---|
getting-started.md | install, run, first authentication, deployment entry points | startup flow or distribution changes |
configuration.md | configuration map, generic CLI flags/env vars, policy-routing runtime ceiling, and Copilot header overrides | generic flags, env vars, or Copilot header defaults change |
provider-routing.md | provider auth, schema-v2 model routes and policies, ordered priority failover, route exposure, model ownership, native endpoint allowlists | providers, routing behavior, failover safety, auth, or model metadata changes |
policy-routing.md | schema-v2 semantic policy routing, classifier privacy/trust, modes, fallbacks, metrics, evaluation, and rollout gates | policy schema, classifier behavior, supported surfaces, telemetry, or release gates change |
provider-api-keys.md | where to get provider API keys and how to map them into providers config | provider signup/key URLs or auth field guidance changes |
tool-optimizers.md | optional shell command rewrite and tool-output reduction config | optimizer config or behavior changes |
responses-websocket.md | Codex-style GET /v1/responses websocket bridge tuning | websocket bridge, auto-compaction, or compact retry knobs change |
clients.md | copy-paste client examples | onboarding snippets or client compatibility changes |
agent-launchers.md | one-command coding-agent launchers, lifecycle, logs, and credential isolation | launcher targets, flags, process lifecycle, or child configuration changes |
api.md | public endpoint map plus cross-surface Chat/Responses compatibility contract | public routes or Chat compatibility behavior changes |
gemini.md | Gemini translation compatibility details | Gemini request/response translation behavior changes |
responses.md | native Responses behavior, the Chat-over-Responses boundary, compact, and memory shims | Responses passthrough, Chat adaptation, compaction, or shim behavior changes |
architecture.md | package responsibilities, route registry/executor, Chat execution seam, replay safety, and data flow | implementation boundaries or design decisions change |
dashboard.md | live browser traffic dashboard, route/attempt metrics, AI insights, and /stats.json | dashboard metrics, endpoints, insight routing, or stats behavior change |
menubar.md | macOS/Linux tray app usage | tray behavior or packaging changes |
development.md | build, test, route-safety, policy-routing, and Chat-over-Responses matrices, benchmark/evaluation baselines, and CI workflows | local dev, test gates, benchmark commands, evaluation gates, or CI change |
Agent Notes
- Prefer linking to one focused file instead of expanding the root
README.md. - When behavior changes, update the smallest relevant doc instead of adding more material to the root README.
- Keep each doc narrowly scoped and avoid duplicating long explanations across files.
- Separate provider-agnostic API behavior from provider-specific auth or routing details when possible.
- When documenting provider features, distinguish proxy-owned websocket bridging from upstream-native websocket or realtime APIs.