router

Project Overview

WorkWeave Router is a model router for agentic systems that routes every prompt to the right model in <50ms. It acts as a drop-in proxy for Anthropic, OpenAI, and Gemini APIs, using a tiny on-box embedder to pick the best model for every request.

Key Features

• Fast routing: Routes every request in <50ms
• Multi-API support: Supports Anthropic Messages, OpenAI Chat Completions, and Gemini native APIs
• Open-source model support: Supports DeepSeek, Kimi, GLM, Qwen, Llama, Mistral via OpenRouter
• BYOK by default: Provider keys stay on your box, encrypted at rest
• Observability: OTLP traces out of the box with Weave dashboard, Honeycomb, Datadog, Grafana, and more
• Streaming, tools, and vision support

Target Audience

• Developers using AI coding agents like Claude Code, Codex, and Cursor
• Teams looking to optimize multi-model API costs
• Engineering teams needing unified API access to multiple LLMs

Quick Start

1. Fastest way: One-command install via npx for Claude Code, Codex, opencode, or pi

   npx @workweave/router

   The installer asks which tool, configures scope, and wires the config file
2. Tool-specific install:
   • Claude Code: npx @workweave/router --claude
   • Codex CLI: npx @workweave/router --codex
   • opencode: npx @workweave/router --opencode
3. Self-host the full stack:

   echo "OPENROUTER_API_KEY=sk-or-v1-..." >> .env.local
   make full-setup

   Router runs at http://localhost:8080, dashboard at http://localhost:8080/ui/dashboard
4. Requires Node ≥ 18

Core Advantages

• Reduces LLM costs by 40-70%
• Ultra-low latency with <50ms per request
• Unified API access to mainstream and open-source models
• Out-of-the-box observability with OTLP traces

System Requirements

• OS: Cross-platform (macOS, Linux, Windows)
• Additional Dependencies: Node.js ≥ 18 (jq required for some paths)