Our take
Newless than six months old and changing fast- Activity
- 65
- Community
- 59
- Issues
- 80
- Pull requests
- 83
+28 stars in the last 30 days
- Merges more pull requests than 95% of the projects we track
- Closes issues faster than 81% of the projects we track
This is a fast-moving, early-stage router with a specific per-action scoring approach and broad client compatibility, but its young history and limited release cadence call for a measured rollout.
Good fit if
- You want a single endpoint for agent clients and multiple model-provider APIs, and can use the Avengers-Pro-derived routing approach.
- You can operate the self-hosted Go and Postgres stack, or are comfortable using the hosted setup.
- You need an Apache-2.0-licensed project and can meet its license terms.
Look elsewhere if
- Your routing must use a different policy or strategy. The README describes the cluster scorer as the default and the HMM policy as optional.
- You need an established project with a long release history. This repository is less than six months old and has only one release in the measured period, a prerelease.
- You need a self-hosted deployment without Postgres or Docker Compose, which are part of the documented setup.
All health signals
| Last commit | 2026-10-19 (today) |
|---|---|
| Commits, last 90 days | 100+ |
| Releases, last 12 months | 1 (latest hmm-model-v1, 2026-07-13, pre-release) |
| Contributors | 22 (top contributor: 50% of commits) |
| Issues closed, last 90 days | 8+ (typically closed in 0 days) |
| Pull requests merged, last 90 days | 93 (typically merged in 0 days) |
| Project age | 6 months |
Checked on 2026-10-20 with the GitHub API.
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
5,620Forks
168Change since Nov 12, 2026: 0 stars·+1 forks
Overview
Weave Router is a model-routing proxy for agentic systems, including Claude Code, Codex, Cursor, and applications that speak Anthropic, OpenAI, or Gemini APIs. Its routing approach uses an in-process cluster scorer derived from the Avengers-Pro research, with an on-box embedder to select a model for each action rather than relying on a prompt-based routing decision.
The project addresses the cost and complexity of wiring agent clients to different models: clients can use a single endpoint while the router chooses among configured providers. It can run as a hosted service or be self-hosted; in self-hosted mode, the router and provider keys stay on the user's machine, with Postgres supporting configuration and usage data.
Key Features
- Routes requests per action using the Avengers-Pro-derived cluster scorer.
- Provides Anthropic Messages, OpenAI Chat Completions, and Gemini API endpoints, including streaming and tool use.
- Connects to hosted providers and OpenAI-compatible endpoints, including OpenRouter.
- Offers a route-preview endpoint that returns a model decision without making an upstream call.
- Supports bring-your-own provider keys, encrypted at rest in self-hosted mode.
- Exports OpenTelemetry traces and provides usage, cost, and routing analytics endpoints.
- Includes an npm installer for Claude Code, Codex, opencode, and pi, plus configuration guidance for Cursor.
Use Cases
- Agent-tool users who want Claude Code, Codex, or another supported client to select among configured models through one endpoint.
- Teams looking to direct different coding-agent actions to different models while retaining their existing client workflows.
- Developers building an application against Anthropic, OpenAI, or Gemini APIs who want a routing layer between the app and model providers.
- Operators who need routing decisions, request costs, and traces available for monitoring or downstream analysis.
What you need
Detected in the repository
- Go 1.25.0 or newer (from go.mod)
- A Dockerfile, so it can run in a container; a Compose file sets up the related services
- Automated checks on GitHub Actions
License in plain words
Apache-2.0permissive
- Commercial use: yes
- Modify and redistribute: yes
- You must keep: the license, the NOTICE file and a note of your changes
- Share your changes: no
- Includes an explicit patent grant from the contributors:
A summary, not legal advice: the LICENSE file is what applies.
Getting Started
For the hosted setup, run:
npx @weave-os/router
For self-hosting, the README's setup uses provider credentials and an admin password in .env.local, followed by make full-setup. See the README for installer options, configuration, and deployment details.
Alternatives
- OmniRoute: OmniRoute offers configurable provider routing and quota-aware fallbacks, rather than Weave Router's in-process, action-based model scorer.
- inference-gateway: Inference Gateway focuses on a self-hosted proxy with provider flexibility, authentication, and observability, rather than action-based model selection.
- llm-d-router: llm-d Router targets Kubernetes inference serving, routing by signals such as KV-cache locality and load rather than coding-agent actions.
| Project | Language | License | Stars | Status |
|---|---|---|---|---|
| router | Go | Apache-2.0 | 5.6k | New |
| OmniRoute | TypeScript | MIT | 72.7k | Active |
| inference-gateway | Go | Apache-2.0 | 215 | Active |
| llm-d-router | Go | Apache-2.0 | 422 | Active |
Considerations
The project is new and changing quickly: recent commit and pull-request activity is high, but the measured release history is sparse. Work is spread across 22 contributors rather than being concentrated in one person; the issue-handling figures show some recent closures, but do not establish a long-term support pattern. The requirements for self-hosting include Go 1.25 and Docker Compose, with Postgres in the documented full-stack setup. The README labels Cursor support as early beta and says performance may not be the best. The Apache-2.0 license applies to the stated release line; check the repository's license notes for earlier releases and third-party components.
Found this useful?
Share it with someone who would like router.