Infrastructure

Edge Routing

BRICQS Edge is an intelligent routing layer that sits in front of your AI deployments. It handles request fan-out, model fallback, rate limiting, and geo-based routing — without any application code changes.

⚡

Low latency

Requests are routed to the nearest healthy deployment replica.

🔀

Model fallback

Automatic failover to a backup model if the primary is overloaded or fails.

🛡

Rate limiting

Per-key and per-IP rate limits enforced at the edge before the model is hit.

🌍

Geo routing

Route traffic to region-specific deployments as you expand globally.

How it works

Your application calls a single edge endpoint. BRICQS routes each request to a deployment based on the configured policy — health, region, load, or a weighted split — and streams the response back without buffering.

bash# Instead of calling a deployment directly:
curl https://<deployment-id>.bricqs.run/v1/chat/completions

# Call the edge endpoint (routes to the healthiest deployment):
curl https://edge.bricqs.ai/v1/chat/completions \
  -H "Authorization: Bearer <api-key>" \
  -H "X-BRICQS-Project: <project-id>" \
  -d '{"model":"llama3:8b","messages":[{"role":"user","content":"Hello"}]}'

Fallback chains

Define a priority-ordered list of deployments. If the primary returns a 5xx or times out, the edge retries automatically against the next deployment in the chain — the caller sees a single response, not the error.

json{
  "chain": [
    { "deployment_id": "prod-llama3-8b",   "priority": 1 },
    { "deployment_id": "prod-llama3-8b-2", "priority": 2 },
    { "deployment_id": "prod-mistral-7b",  "priority": 3 }
  ],
  "timeout_ms": 30000
}

Rate limiting

Rate limits are enforced per API key before the request reaches your deployment. Configure limits in the API Keys section of the dashboard:

requests_per_minuteRolling 60-second request count per keyrequests_per_dayDaily request cap per key, resets at midnight UTCtokens_per_minuteRolling 60-second output token count per key

Note: Rate limit headers (X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset) are included in every response so your clients can back off gracefully.

Geo routing

BRICQS is currently deployed in West US 2. Multi-region routing (East US, West Europe, Southeast Asia) is on the roadmap and will be available as additional regions are provisioned. When enabled, the edge will automatically route to the closest region with a healthy deployment.

Current region

West US 2

westus2 · US West · All compute, storage, and DB

East US — Coming soon

eastus · US East · Planned for Phase 4 multi-region