Repo Year in Review Limited info

diegosouzapw/OmniRoute

Free MIT-licensed AI gateway: one local /v1 endpoint routes coding tools across many providers, including free tiers, with auto-fallback and token compression.

  • 74.7k GitHub stars
  • TypeScript
  • ⚖️ MIT
diegosouzapw/OmniRoute preview image

What it is

OmniRoute is a gateway that gives tools like Claude Code, Codex, Cursor, Cline and Copilot a single local endpoint (http://localhost:20128/v1) in front of hundreds of AI providers. It can fall back across four provider tiers (Subscription, API Key, Cheap, Free) while a healthy target remains. It also offers RTK + Caveman compression, 19 routing strategies, circuit breakers, MCP, A2A and a dashboard that tracks free-tier budgets. It is not a token grant: you connect your own provider accounts or API keys.

Who it's for

  • Developers using AI coding CLIs and agents (Claude Code, Codex, Cursor, Cline, Copilot) who want one endpoint for many providers
  • Users who want to stack third-party free tiers and avoid hitting rate limits mid-coding
  • Teams that want local-first routing with usage, quota and savings visibility

Requirements

Requirements

  • Your own provider accounts or API keys for the providers you connect (OmniRoute does not grant tokens)
  • An OmniRoute API key, copied from Dashboard → Endpoints
  • At least one eligible route so the auto model can resolve

Setup

  1. Install and start OmniRoute

    The README describes three steps: install and start OmniRoute, connect an eligible provider with your own account or API key, then point your tool at localhost:20128/v1. The install commands are in the README's Quick Start section, which was not included in the provided text.

  2. Connect a provider and get your key

    Connect a provider you are eligible to use, then copy your OmniRoute key from Dashboard → Endpoints.

  3. Send a test request

    Call the local endpoint with your OmniRoute key and the model auto.

    bash
    curl http://localhost:20128/v1/chat/completions \
      -H "Authorization: Bearer YOUR_OMNIROUTE_KEY" \
      -H "Content-Type: application/json" \
      -d '{"model":"auto","messages":[{"role":"user","content":"Hello!"}]}'

Examples

Chat completion via the auto model

bash
bash
curl http://localhost:20128/v1/chat/completions \
  -H "Authorization: Bearer YOUR_OMNIROUTE_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"auto","messages":[{"role":"user","content":"Hello!"}]}'

What it does: Sends an OpenAI-style chat request to the local OmniRoute endpoint. The model auto lets OmniRoute pick an eligible route.

Pros & cons

Pros

  • Pro:Single local /v1 endpoint with OpenAI, Claude, Gemini and Responses API compatibility, shared across many coding tools
  • Pro:Automatic fallback across four tiers (Subscription, API Key, Cheap, Free) plus circuit breakers and key cooldown
  • Pro:Dashboard page (/dashboard/free-tiers) shows free-tier budgets with documented pool-deduplicated methodology
  • Pro:MIT license, with RTK + Caveman compression claimed to save 15–95% tokens on eligible workloads

Cons

  • Con:The auto model needs an eligible route, and a fresh install may have no eligible keyless targets
  • Con:Free-token figures are estimates across third-party tiers; each provider controls its own limits and terms, and some keyless providers may reject third-party clients
  • Con:Providers marked tos: avoid (including OpenCode Free and Kiro) are excluded from automatic routing by default, even once connected

Images