diegosouzapw/OmniRoute

Published on Jul 28, 2026
★ 84.2k MIT

A free local AI gateway routing across 290+ providers, no quota wall

Stop juggling five different AI provider API keys

One local endpoint, 290+ providers behind it.

OmniRoute runs on your own machine and gives every coding tool you use one address to talk to. Behind that one address sit two hundred ninety plus AI providers and five hundred plus models.

One endpoint · 290+ providers · 500+ models

Works the second you install it.

No API keys, no config file, nothing to sign up for first. You install it and it's already routing. Ninety plus of those providers have a free tier, and forty plus of those are free permanently.

npx install · zero-config startup

One key dies, it just falls back.

When a provider's quota or budget runs out, OmniRoute doesn't stop — it cascades to the next tier automatically. Nineteen routing strategies, a circuit breaker, and cooldown backoff keep one dead key from stopping you.

Auto-fallback across 4 tiers · 19 strategies

Compresses your tokens before they're sent.

A twelve engine compression pipeline strips your prompt down before it ever reaches the model — on average, it cuts token usage by about eighty nine percent.

12-engine compression pipeline · avg 89% token savings

30.6k stars. Free and MIT licensed.

OmniRoute is fully open source under the MIT license, with over five hundred contributors and thirty point six thousand stars. Try it this weekend.

diegosouzapw/OmniRoute · MIT License