What it does
OmniRoute is a gateway you run yourself. Your apps and coding tools send requests to one local endpoint, and OmniRoute passes them on to whichever connected provider and model fits, with a dashboard for providers, keys and usage. Its central idea is the combo: a chain of models that the gateway moves along automatically when quota runs out, a provider fails or costs rise. Setting the model to auto builds a combo from your connected providers, with variants such as auto/coding, auto/fast and auto/cheap. There are 19 routing strategies, including priority, round-robin, cost-optimized and a mode that fans a request out to several models and has a judge synthesise one answer. You bring your own provider accounts. The project catalogs free tiers from third-party providers and says their limits, availability and terms are set by those providers, not by OmniRoute. Other listed features include prompt compression, MCP and A2A support, a prompt-injection guard, and one-command setup for tools such as Claude Code, Codex and Aider.
Capabilities
- One OpenAI-style endpoint in front of hundreds of providers
- Combos and auto modes that fall back across models when quota or health changes
- 19 routing strategies, with presets tuned for coding, speed or cost
- Prompt compression to reduce token use
- Cost and usage headers, quota telemetry and a dashboard
- MCP and A2A support, with one-command setup for coding CLIs
When to use it
- You use several AI providers and want one endpoint with automatic failover
- You want to spread usage across the free tiers of providers you already have accounts with
- You run coding agents or CLIs and want to track cost and quota across them
