Features / Self-hosted models
Budgets and limits for your own model server.
Running models yourself does not make them free: GPUs cost money and capacity is limited. On Pro and Enterprise, ACPpotamia puts the same limits, loop detection and logs in front of any OpenAI- or Anthropic-compatible endpoint you run.
How it works.
Any compatible server
vLLM, Ollama, Text Generation Inference, LiteLLM or your own service, over HTTPS.
Priced safely
Models we do not know are billed against your budget at a deliberately high default, so nothing slips through. Tell us your model's real cost and we set its price for you.
Locked to your address
The endpoint comes only from your team settings, never from the request, so the gateway cannot be pointed elsewhere.
Questions.
How do I route a call to my server?
Add the header X-ControlPlane-Upstream: custom to the call; everything else stays the same.
Related features.
OpenAI gateway
An OpenAI API proxy for chat completions, the Responses API and embeddings, with budgets, rate limits and loop detection. Keep the official OpenAI SDK.
Claude (Anthropic) gateway
A Claude API gateway for the Anthropic Messages API: budgets, rate limits, loop detection and audit logs for Claude Opus, Sonnet and Haiku. Keep the Anthropic SDK.
Rate limiting
Rate limit every AI agent's calls to OpenAI and Anthropic per team, at the edge. Requests over the limit get HTTP 429 with Retry-After before they cost money.