Open-source AI infrastructure

One gateway.
Every model route.

OmniRoute connects your application to free and open LLM providers through one resilient API. Switch models, route around limits, and keep your existing SDK.

See how routing works View the repository
route / completion
Your applicationPOST /v1/chat/completions
OmniRoute gatewaypolicy · health · quota
Geminiselected
Groqready
Cohereready
latency healthyfallback armed
Built for developers who need options.One integration instead of a provider-by-provider rewrite.
28+provider integrations
2compatible API formats
1self-hosted control plane
01 / Capabilities

Infrastructure that keeps model choice open.

Route each request by availability, quota, or preference while your application keeps speaking the same familiar API.

One unified API

Connect once, then address multiple providers and models through a stable gateway endpoint.

POST /v1/chat/completions
Authorization: Bearer ••••••

Smart routing

Use provider health, remaining capacity, model fit, and your own policy to choose a route.

28+ providers

Bring together mainstream APIs, free tiers, inference services, and local model runtimes.

SDK compatible

Keep OpenAI or Anthropic client patterns. Change the base URL and credentials—not your application architecture.

OpenAI-compatible
Anthropic-compatible
minimal migration
02 / Provider network

Reach the model ecosystem from one place.

Use the provider you already have, combine free allocations, or add a new route without changing downstream clients.

Google Gemini
Groq
OpenRouter
Cloudflare AI
Hugging Face
Cohere
NVIDIA NIM
Ollama
OpenAI
Anthropic

Provider support is modular. Each connector normalizes authentication, model selection, responses, and errors behind the gateway.

03 / How it works

From request to response in three steps.

OmniRoute stays between your client and the provider network, so routing decisions remain centralized and observable.

01

Connect providers

Add credentials for the services you use and choose which models are eligible for routing.

02

Set a policy

Prioritize a model, prefer available free capacity, or define a fallback order for your workload.

03

Send one request

Point an OpenAI or Anthropic client to OmniRoute and receive a normalized response.

from openai import OpenAI

client = OpenAI(
  base_url="https://your-gateway.example/v1",
  api_key="YOUR_GATEWAY_KEY"
)

response = client.chat.completions.create(
  model="auto",
  messages=[{"role": "user", "content": "Hello"}]
)
04 / Open source

Inspect the gateway. Extend the routes. Run it yourself.

OmniRoute is being built in the open for developers who want control over provider choice, routing policy, and deployment. The architecture is designed for self-hosting and community-built connectors.

Public GitHub link coming soon
Built for independent deploymentConfigure the provider mix and operating policy that match your application. Repository and license details will be published with the first public release.
05 / Contact

Building with multiple model providers?

OmniRoute is preparing its first public release. Contributor, integration, and early-access conversations will open with the repository.

Follow the open-source releaseContact channel coming soon. The public repository will carry project discussions and release updates.