Point your tools at the gateway; it forwards, meters and encrypts for you. Real keys stay on your machine — tools only get per-project virtual keys.
Enter each provider's real key once — encrypted on your machine, never uploaded.
Issue a virtual key per project or per tool, scoped and individually revocable.
Set any tool's base URL to the gateway; swap models or providers later without touching the tool.
The core is free; team-scale features upgrade on demand — never metered per token.
Cloud frontier models and local Ollama models, unified behind a single endpoint.
… and any OpenAI-compatible provider
The advanced controls small teams and enterprises need — still never metered per token.
No. Real keys are encrypted on your machine and tools only get virtual keys — nothing is uploaded.
Any tool that supports a custom base URL, including Claude Code, Codex and your SDKs.
One endpoint for cloud and local models, virtual keys, usage and cost tracking, automatic fallback and on-device encryption — all free.
Yes. Local Ollama models sit behind the same endpoint as cloud models.
Yes. The gateway displays plan details, available endpoints, output limits, and multi-window quota usage (used / remaining / reset time) for bound subscription accounts.
It doesn't. Your client picks the model in each request; the gateway routes that request to a channel that serves it.
They decide which channel serves a request when several channels serve the same model: Fixed always uses one, Failover tries backups in order, Balanced rotates evenly. Different models never mix.
