// Agent profile
WorkWeave Router is a Go-based proxy service that sits between your application and LLM providers, intelligently routing each request to the most cost-effective model that can handle it. Using an embedded ONNX classifier, it completes routing decisions in under 50ms and claims 40-70% cost savings by sending simple requests to cheaper models and complex ones to frontier models. It supports Anthropic Messages, OpenAI Chat Completions, and Gemini formats natively, with bring-your-own-keys encryption so provider credentials never leave your infrastructure. The router ships with first-class integrations for Claude Code, Codex, opencode, pi, and Cursor, and includes built-in OTLP tracing for Honeycomb, Datadog, and Grafana. Deploy via a one-line npx command for hosted mode or self-host with Docker Compose and PostgreSQL.
WorkWeave Router is a model-routing proxy written in Go that solves a growing problem in LLM-powered applications: not every request needs a frontier model, but most applications send everything to one. The router intercepts API calls, classifies the request complexity using an embedded ONNX model, and forwards it to the most cost-effective provider and model that can handle it — all in under 50ms of added latency. The result is 40-70% cost reduction on LLM spend without degrading output quality for the tasks that matter.
The router supports three major API formats natively — Anthropic Messages, OpenAI Chat Completions, and Gemini — so it slots into existing codebases without rewriting integration code. Provider credentials use bring-your-own-keys with local encryption, meaning your API keys never transit through WorkWeave’s infrastructure. The routing classifier is based on academic clustering research rather than simple heuristics, giving it better accuracy on edge cases where a cheaper model can handle what looks like a complex request. Built-in OpenTelemetry tracing exports to Honeycomb, Datadog, or Grafana, giving teams full visibility into routing decisions, latency, and cost per request.
The primary use case is cost optimization for agentic systems that make many LLM calls per task — coding agents, research pipelines, and multi-step workflows where the majority of intermediate calls (tool selection, formatting, simple extraction) do not need a frontier model. Teams running Claude Code, Codex, or Cursor through the router can route routine completions to cheaper models while preserving frontier quality for complex reasoning steps. The self-hosted option with PostgreSQL also suits enterprises with strict data residency requirements.
Licensed under Elastic License v2, which permits self-hosting but restricts offering the router as a managed service to third parties. The routing classifier adds a small latency overhead (under 50ms), which is negligible for most applications but worth measuring for latency-sensitive real-time systems. The Go requirement (v1.25+) means self-hosted deployments need a recent Go toolchain, though the hosted npx path avoids this entirely.
WorkWeave Router is built for engineering teams spending meaningfully on LLM APIs — particularly those running agentic workflows with high call volumes across multiple providers. If you are already using Claude Code, Codex, or Cursor and want to reduce costs without switching tools, the native integrations make adoption straightforward.
Builder.io's open-source framework for building agent-native applications — shared actions, SQL-backed state, identity, tools, skills, jobs, observability, and UI surfaces that all work together.
CLI orchestration tool that installs and configures platform connectors for 16+ social and web sources — Twitter/X, Reddit, YouTube, GitHub, Bilibili, and more. Zero API fees.
Persistent memory layer for AI coding agents — benchmark-backed (95.2% on LongMemEval-S), 92% fewer tokens per session vs full-context pasting, zero manual memory.add() calls.