One endpoint. Every model. Always the right one.

A drop-in proxy for Anthropic, OpenAI, and Gemini that picks the best model for every request: using a tiny on-box embedder, not a vibes-based prompt.

*Built by Weave: The #1 engineering intelligence platform, loved by Robinhood, PostHog, Reducto, and hundreds of others.*

What it does

Point Claude Code, Codex, Cursor, or your own app at localhost:8080. The router:

• 🎯 Routes per action. A cluster scorer derived from

• 🔌 Speaks everyone's API. Anthropic Messages, OpenAI Chat Completions,

• 🧠 Knows OSS too. DeepSeek, Kimi, GLM, Qwen, Llama, Mistral via

• 🔒 BYOK by default. Provider keys stay on your box, encrypted at rest.
• 📊 Observable. OTLP traces out of the box. See them in the Weave dashboard (http://localhost:8080/ui/dashboard) or drop in Honeycomb, Datadog,

30-second quickstart

The fastest way: point Claude Code, Codex, opencode, or pi at the hosted Weave Router with one command. No clone, no Docker, no Postgres.

npx @weave-os/router

That's it. Your browser opens the hosted setup page. To install directly from the terminal, choose a target explicitly. Other flavors:

npx @weave-os/router --claude              # skip the picker, Claude Code
npx @weave-os/router --codex               # skip the picker, OpenAI Codex CLI
npx @weave-os/router --opencode            # skip the picker, opencode
npx @weave-os/router --pi                  # skip the picker, pi + Loom UI
npx @weave-os/router --scope project       # per-repo, commits settings.json (or .codex/ / opencode.json)
npx @weave-os/router --local               # self-hosted localhost:8080
npx @weave-os/router --base-url https://router.acme.internal
npx @weave-os/router --email [email protected] # set identity without prompting
npx @weave-os/router --claude --return-url https://app.example.com/onboarding # continue after health + key verification
npx @weave-os/[email protected]                 # pin a version

Requires Node ≥ 18 (Claude Code, opencode, and pi paths also need jq). Full flag reference: install/npm/README.md.

The npm package is published as @weave-os/router. The former @workweave/router package remains available as a compatibility alias and continues to receive the same releases.

Or: self-host the whole stack

If you want the router (and dashboard) running on your own box:

# 1. Drop a provider key in. OpenRouter is the recommended baseline.
echo "OPENROUTER_API_KEY=sk-or-v1-..." >> .env.local
2. Set a dashboard password. Without one, inference still runs but
dashboard administration is disabled.
echo "ROUTER_ADMIN_PASSWORD=replace-with-a-strong-password" >> .env.local
3. Boot Postgres + router on :8080 and seed an rk_ key.
make full-setup

The router is up at , the dashboard at , and your rk_... key prints in the logs.

# Call it like Anthropic
curl -sS http://localhost:8080/v1/messages \
  -H "Authorization: Bearer rk_..." \
  -d '{"model":"claude-sonnet-4-5","max_tokens":256,
       "messages":[{"role":"user","content":"hi"}]}'
...or like OpenAI
curl -sS http://localhost:8080/v1/chat/completions \
  -H "Authorization: Bearer rk_..." \
  -d '{"model":"gpt-4o-mini",
       "messages":[{"role":"user","content":"hi"}]}'
Peek at the routing decision without proxying
curl -sS http://localhost:8080/v1/route -H "Authorization: Bearer rk_..." -d '...'

What that stack looks like

Only the grey boxes are off your machine. The router, the scorer, Postgres, and your provider keys all stay local; prompts go from the router straight to the provider you configured, never to Weave.

flowchart LR
    client["Claude Code, Codex, opencode,<br/>pi, Cursor, your own app"]
    router["Router :8080<br/>/v1/messages · /v1/chat/completions<br/>/v1beta/models · /v1/route"]
    scorer["Cluster scorer<br/>in-process ONNX embedder"]
    hmm["HMM policy sidecar :8093<br/>optional, make up-hmm"]
    pg[("Postgres<br/>installations, rk_ keys,<br/>encrypted BYOK keys, usage")]
    ui["Dashboard /ui<br/>selfhosted mode only"]
    providers["Anthropic · OpenAI · Gemini<br/>OpenRouter and any<br/>OpenAI-compatible endpoint"]
    otel["Your OTLP collector<br/>Honeycomb, Datadog, Grafana"]
client -->|"rk_… bearer token,<br/>streamed response back"| router
    router -->|"embed and score the action"| scorer
    router -.->|"ROUTER_DEFAULT_STRATEGY=hmm"| hmm
    router -->|"auth, config, usage"| pg
    pg --> ui
    router -->|"provider key from env or BYOK"| providers
    router -.->|"spans and usage logs"| otel
classDef external fill:#f4f4f5,stroke:#a1a1aa,color:#3f3f46
    class providers,otel external

Multi-replica deployments also need Pub/Sub (PUBSUB_*) for cache invalidation; docker compose runs the emulator for you.

Optional: self-host the frozen HMM policy

The default stack uses the in-process cluster scorer. To run the frozen HMM policy as a companion container, add a Google API key and use the opt-in target:

echo 'GOOGLE_API_KEY=...' >> .env.local
make up-hmm

This does not change the default strategy. See sidecars/hmm/README.md for artifact verification, embedding compatibility, and explicit HMM selection.

Wire it into your tools

Claude Code. Run make install-cc to wire Claude Code at the local self-hosted router (it's also invoked automatically at the end of make full-setup). For the hosted router, use npx @weave-os/router above.

Codex (OpenAI CLI). npx @weave-os/router --codex patches ~/.codex/config.toml (or /.codex/config.toml with --scope project) with a managed [model_providers.weave] block and sets model_provider = "weave". The provider preserves Codex's existing ChatGPT OAuth login while the router key rides in an X-Weave-Router-Key HTTP header and the installer selects the HMM strategy for the public hosted endpoint. --codex --local and custom self-hosted URLs keep their router's configured default because the HMM sidecar is optional. HMM and forced selections in the native Codex family (gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna) use that OAuth credential; every other selected model uses its WorkWeave deployment or BYOK credential, matching the Claude Code plugin's model-to-credential dispatch. Codex does not load third-party slash-command files, so the installer ships $force-model (alias $fm ), $unforce-model (alias $ufm), and $router-feedback (alias $rf ) as native Codex skills. Their invocations are handled by the router directly. Re-install and --uninstall --codex rewrite/remove only the managed block, leaving the rest of your Codex config untouched. Codex also gets $router-status, $router-off, $router-on, and $router-models as skills that call this installer's own verbs. Invoke $disable-routing (or $router-off) to switch the next Codex session back to its normal provider, or run npx @weave-os/router disable-routing in a shell; a literal /disable-routing is not a third-party extension point in Codex.

opencode. npx @weave-os/router --opencode requires OpenCode major version 2 when the CLI is installed and available on PATH. It merges a provider.weave entry into ~/.config/opencode/opencode.json (or /opencode.json with --scope project) and makes weave/auto the active model while saving any prior default for off/uninstall. OpenCode v2 sends Responses requests through its bundled @ai-sdk/openai provider; the router selects and translates to the upstream model. The router key, identity, and session headers ride alongside the provider config. No plugin is installed; re-install removes stale v1 plugin registrations/files, and --uninstall --opencode removes the provider and restores the prior model.

pi. npx @weave-os/router --pi keeps stock pi as the runtime and installs the router's pi extension. It adds the Loom header, Wooly's animated terminal mascot, a persistent WEAVE ROUTER route/savings line, /fm + /ufm model-pin commands with a [forced] status, and context-isolated subagents without shipping or maintaining a forked pi binary.

Cursor (early beta, performance may not be the best). Settings → Models → Override OpenAI Base URL → http://localhost:8080/v1, paste rk_... as the API key.

Switching on/off. After installing, npx @weave-os/router off --claude (or --codex / --opencode) routes that client straight to its provider again without discarding the router config; on flips it back, and status reports which way it's pointing. Claude Code also gets /router-off, /router-on, and /router-status slash commands. Cursor toggles via the same Settings → Models override above. See install/README.md.

Choosing which models the router may pick. npx @weave-os/router models --claude lists every deployed model with its on/off state, and models enable / models disable change it — the same setting as the dashboard's settings page, edited from the terminal. Claude Code gets this as /router-models (alias /models). Requires a router that serves the model-selection API; against the Weave-hosted router the list still prints and points you at the dashboard, where selection is an organization-wide setting. See install/README.md.

Two keys, don't mix them up:

- sk-or-... / sk-ant-... / sk-... = your upstream provider key. Lives in .env.local.

- rk_... = your router key. Clients send this as a Bearer token.

Endpoints

| Endpoint | Format | | ------------------------------ | ---------------------------------------- | | POST /v1/messages | Anthropic Messages, routed | | POST /v1/chat/completions | OpenAI Chat Completions, routed | | POST /v1beta/models/:action | Gemini generateContent, routed | | POST /v1/route | Returns the decision, no upstream call | | GET /v1/models  ·  POST /v1/messages/count_tokens | Anthropic passthrough | | GET /livez  ·  GET /startupz | process responsive check + completed initialization | | GET /health  ·  GET /readyz  ·  GET /validate | responsive-check alias + dependency diagnostics + key check | | POST /v1/client-events | Harness CLI's off/on/uninstall report (harness = claude_code/codex/opencode/pi); logged + exported as router.harness_lifecycle, nothing stored | | GET /v1/sessions/:session_id/cost | One session's committed cost + savings, scoped to your key's installation; rk_ or ra_ key (docs) | | GET /v1/analytics/routing-decisions | Raw routing decisions as cursor-paginated NDJSON (docs) | | GET /v1/analytics/schema  ·  GET /v1/analytics/models | Export field dictionary + price book |

Use /startupz for Cloud Run startup and /livez for its process responsive check. The listener starts after artifact/client initialization and bounded connection preparation. /startupz also pings PostgreSQL with a two-second budget: failure returns 503, so Cloud Run withholds traffic and retries while the process stays running. Provider preparation failures only log warnings. /livez does not contact the database. Startup sends no prompts and performs no embedding or HMM inference; it cannot guarantee the latency of first inference. /readyz retains dependency diagnostics.

Routed non-stream responses include x-router-cost-usd, x-router-cost-input-usd, x-router-cost-output-usd, x-router-cache-read-tokens, and x-router-cache-creation-tokens. Streaming responses cannot carry the final cost in HTTP headers because headers flush before usage is known; stream clients should read the weave_cost object on the final Anthropic message_delta usage event.

Deeper docs

• 📐 Configuration reference: every env var,

• 🧭 Semantics and terminology: canonical definitions

• 📊 Analytics export: pulling raw routing

• Policy router harness: contract and

• 📏 Codex benchmark harness: reproduce the published

• 🛠️ Contributing: layering rules, hot-reload dev,

• 🏗️ Architecture: package layout, import contracts,

License

Licensed under the Apache License 2.0. Copyright 2025-2026 Workweave, Inc. See NOTICE for attribution.

The Apache-2.0 release line begins with router-v0.2.24 (npm packages @weave-os/router and @workweave/router version 0.2.24). Earlier tags and published packages retain the licenses shipped with them; this change does not rewrite historical releases.

Apache-2.0 permits commercial use, hosted services, and proprietary derivatives, subject to its terms, and includes a patent grant. It does not grant general rights to Workweave's trademarks. Third-party code, dependencies, model weights, and datasets retain their own licenses; see THIRD_PARTY.md.

[^1]: Zhang, Y. et al. *Beyond GPT-5: Making LLMs Cheaper and Better via Performance–Efficiency Optimized Routing* (Avengers-Pro). arXiv:2508.12631, 2025.