Requesty (United Kingdom) is a managed AI gateway that sits between the application and model providers. The application points at an OpenAI-compatible endpoint and the gateway picks the provider, applies routing policies and records usage. The product is SaaS: the gateway is not distributed for self-hosting.
What it’s for
- Unifying access to hundreds of models from different providers behind one endpoint
- Rule-based and latency-based routing, with automatic failover when a provider fails
- Load balancing across keys and providers
- Spend limits, budget alerts and per-request cost tracking
- Observability for latency, time to first token and cost per session
- Guardrails and sensitive-data masking on calls
How it works
- The application calls an OpenAI- or Anthropic-compatible endpoint and the gateway resolves the provider
- Routing policies define preferences, order and fallback criteria
- Your own provider keys (BYOK) can replace platform credits
- An MCP Gateway exposes tools to agents from the same entry point
- The European endpoint keeps the gateway’s processing in the Frankfurt region
How it differs
- It is managed: no proxy, database or panel to operate yourself
- The focus is the access and routing control plane, not model execution
- The official SDKs and CLI are open source, but the gateway itself is proprietary
- EU data residency is available as an option, with the caveat that inference follows the chosen provider
Availability and pricing
- Free tier with a daily request limit on free models
- Pay-as-you-go with a fee on model cost, waived when using your own keys
- Enterprise with SSO, RBAC, audit trail and service-level agreements
Things to note
- The extra layer adds a dependency in the call path
- Governance and compliance features vary by plan
- Part of the catalog depends on third-party availability