Multi-provider AI Routing
OpenAI, Anthropic, Google Gemini, and Ollama — all four wired in. Health-aware failover switches providers automatically. The right model for each conversation type — without manual intervention.
Resilient AI routing built for production
4 providers wired in
OpenAI, Anthropic, Google Gemini, and Ollama (self-hosted) are all supported. Switch per conversation type or let the router decide based on availability and latency.
Health-aware failover
The router monitors each provider's health in real time. If a provider degrades or goes down, traffic fails over to the healthy one automatically.
Model selection per type
Use GPT-4o for customer support, Claude for long-context knowledge base queries. Configure model affinity by conversation intent.
Cost optimization
Route lower-stakes conversations to cheaper, faster models. Reserve high-capability models for complex queries. Configurable per tier.
Technical implementation
Health checks per provider
A background worker polls each configured provider on a short interval. Health state is stored in Redis. The routing layer reads from Redis — no synchronous health check on the hot path.
Automatic fallback
When a provider's error rate exceeds a configurable threshold, the circuit breaker opens and the router stops sending traffic to it until health is restored.
Latency-aware routing
Recent p95 latency per provider is tracked. When both providers are healthy, the router can prefer the lower-latency option for latency-sensitive conversation types.
Never go down because one AI provider did
Multi-provider routing is available on all paid plans.