Requesty

Requesty is a hosted gateway that puts one OpenAI-compatible endpoint in front of more than 600 models from over 20 providers. Applications send every call through it, and routing policies decide where each one lands: fallback chains that try the next model when one fails, load balancing by weight, latency-based routing to the fastest provider for a model, and Auto Routing (beta, October 2026), where a small classifier model reads the first user message and sends the conversation to the model you attached to that kind of request.

The dashboard covers cost, request volume, token usage and latency per model, with logs, traces and session reconstruction. Organisation controls include spend limits and rate limits per key, an approved-models list filterable by region and data policy, access lists, service accounts for CI, guardrails with PII detection, an MCP gateway and a prompt library with version diffs. An EU router keeps traffic in the EU, and you can bring your own provider keys.

Pricing is usage-based with no seats: the Free tier runs free models at 200 requests a day, pay-as-you-go adds 5% to the provider price, and Enterprise adds SSO, full RBAC, audit logs and custom SLAs. It fits teams that want routing policies and governance over many models without running the gateway themselves.