CRCogniRelay

AI API Gateway

CogniRelay

Unified gateway for multi-model AI routing & load balancing — one OpenAI-compatible endpoint, intelligent failover, every provider.

OpenAI-compatible/Smart routing/Auto failover
cognirelay — route.session
$ curl https://api.cognirelay.com/v1/chat/completions \
  -H "Authorization: Bearer $CR_KEY" \
  -d '{"model":"auto","messages":[{"role":"user","content":"ping"}]}'

# routed → claude-sonnet · 187ms · $0.0021
{
  "id": "cr-7f3a…",
  "model": "anthropic/claude-sonnet-4",
  "route": "latency+cost",
  "choices": [{}]
}

// features

Gateway primitives that ship.

Everything you need to unify models behind one API — without bolting together fragile proxy scripts.

Intelligent routing

Route by latency, cost, or quality. Let CogniRelay pick the best model per request — or pin exact providers when you need control.

Multi-provider failover

Automatic retries across OpenAI, Anthropic, Google, DeepSeek, and more. Keep your apps online when a single vendor blips.

OpenAI-compatible API

Drop-in base URL swap. Streaming, tools, and embeddings work with the SDKs you already use — zero rewrite.

Load balancing

Spread traffic across keys and regions. Smooth spikes without rewriting your orchestration layer.

Low-latency edge

Gateway designed for fast path selection and minimal hop overhead — built for production chat and agents.

Observability built-in

Trace every hop: model chosen, tokens, cost, and fallback path. Debug routing decisions in one place.

// models

One key. Every frontier model.

Switch providers with a model string. CogniRelay normalizes responses so your clients stay stable.

OpenAI

GPT-5 · o-series

Anthropic

Claude 4 · Sonnet

Google

Gemini 2.5

DeepSeek

V3 · R1

Meta

Llama 4

Mistral

Large · Codestral

xAI

Grok

Qwen

Qwen3

// api

Point your SDK. Ship.

OpenAI-compatible endpoint. Change base_url and keep the rest of your stack.

from openai import OpenAI

client = OpenAI(
    api_key="cr-...",
    base_url="https://api.cognirelay.com/v1",
)

res = client.chat.completions.create(
    model="auto",  # or anthropic/claude-sonnet-4
    messages=[{"role": "user", "content": "ping"}],
)
print(res.choices[0].message.content)

// routing

Request path, demystified.

A clean three-stage pipeline between your app and every model provider you care about.

  1. 01

    Ingest

    Receive OpenAI-compatible requests at a single endpoint with your CogniRelay key.

  2. 02

    Route

    Score candidates on latency, cost, capacity, and policy — pick or fan-out.

  3. 03

    Relay

    Stream tokens back. On failure, fail over to the next healthy provider transparently.

early access

Route your next request through CogniRelay.

We're onboarding teams building agents, chat products, and internal AI platforms. Tell us what you ship.