01
BYOK-first
Bring your own key per request — never persisted. Managed keys and OAuth are fallbacks, not requirements.
A self-hosted, provider-agnostic LLM gateway. One OpenAI-compatible endpoint in front of every model lab — with policy auto-routing, streaming, tenancy, and metered billing that you own end to end.
Everything Ramp's router does, self-hosted — plus the parts they don't give you: your credentials never leave, and the routing policy learns from your own outcomes.
Bring your own key per request — never persisted. Managed keys and OAuth are fallbacks, not requirements.
Send model:"auto" with a cost/quality policy; the router picks the model and a fallback chain.
Token passthrough with translation across wire formats. Fallback only before the first byte — never a silent switch.
Orgs, projects, scoped keys, atomic credit holds and an append-only ledger that reconciles to the balance.
OpenAI, Anthropic, OpenRouter, Groq, DeepSeek, Together, xAI, Cerebras, Mistral, Google, and more — one endpoint.
Point Claude Code, Codex, and OpenCode at your router in one command — reversible byte-for-byte.
Pulled from this deployment in real time.
This runs against the live gateway in dry-run mode — the routing decision is real, no credential or credit is touched.
One binary path to a full OpenAI-compatible endpoint on your own infra. Available models on this node: