Open sourceTypeScript · ≈82k stars · MIT
OmniRoute
Free, self-hosted AI gateway: one OpenAI-compatible endpoint with quota-aware fallback for coding agents.
OmniRoute is an MIT-licensed, local-first AI gateway. It exposes one OpenAI-compatible endpoint at localhost:20128/v1 and routes across hundreds of providers: the README says 375, the site 352. Fallback is quota-aware. It offers 19 routing strategies, circuit breakers and model lockouts, semantic caching, and a context-compression pipeline that it claims saves 15-95% of tokens. It also has a built-in MCP server and A2A server. It works with Claude Code, Codex, Cursor, OpenCode, Cline and Copilot. The project has joined Cheaper Inference, which offers a hosted version.
Pros & cons
Each point cites its sources and says what kind of evidence backs it. A second reviewer tried to refute every claim before it was kept.
Pros
- DocsMIT-licensed local gateway that exposes one OpenAI-compatible endpoint (localhost:20128/v1) for Claude Code, Codex, Cursor, Cline and others.[1]
- DocsQuota-aware fallback across roughly 375 registered providers, with 20 routing strategies, circuit breakers and cooldowns.[1]
- User reportsPractitioners pair it with OpenRouter, using OpenRouter as the hosted router and OmniRoute as the local one.[1]
- DocsBuilt-in extras (MCP server, A2A, memory, guardrails, dashboard) and installs via npm, Docker or Electron.[1]
Cons
- DocsFeatures such as subscription quota-sharing, TLS fingerprint "stealth" and transparent MITM decryption may conflict with provider terms; apart from excluding providers flagged "tos: avoid", the README leaves ToS checks to you.[1]
- DocsThe README leans on unverified marketing claims ("never hit limits", 15-95% compression savings), gives inconsistent provider counts (377 vs 375), and includes affiliate links that fund the project.[1]
- Independent testIts compression stack includes RTK, whose advertised savings an independent JetBrains benchmark did not reproduce.[1][2]
- DocsFree-tier figures are third-party estimates that "move both ways," and auto mode needs an eligible route to work.[1]
- Best for
- Cost-sensitive individual developers who want one local endpoint that falls back across many providers and free tiers.
- Watch out for
- Check each provider's terms before routing subscription accounts through it, and don't rely on free-tier capacity in production.
Assessed Oct 29, 2026 · adversarially reviewedSuggest a correction
People
Founder
Creator of OmniRoute
Cost & limits
Neighbours on the workbench
TypeScript · ≈393k stars · MIT
openclaw/openclaw
The viral open-source personal AI assistant that really does things.
Maintained by OpenClaw Foundation
MDX · ≈307k stars · MIT
obra/superpowers
An agentic skills framework and software methodology that works.
Maintained by Jesse Vincent (obra)
Shell · ≈306k stars · MIT
mattpocock/skills
Small, composable agent skills for real engineering work.
Maintained by Matt Pocock (AI Hero)
Connections
Relations from the Harness Hub knowledge graph, as charted in the State of Harness reports.
Maintained by · 1
- Cheaper InferenceOrganisation
Built on · 2
- Agent2Agent Protocol (A2A)Project
- Model Context ProtocolProject
Similar entries · 2
- ContextForge MCP GatewayProject
- MCPHubProject
Sources
Verified Oct 29, 2026