docs / running work · compute
Inference
Every turn an agent takes is paid for by someone. Mod lets a session use the subscription you already have, your own API keys, or Mod's inference router.
Mod decides
- Three ways to pay: a subscription through its harness, like a Claude or ChatGPT login; your own API key; or Mod's router.
- Keys and logins stay on the machine or in the router. They are never written to the log or a document.
- The router speaks to Anthropic and OpenAI through one interface, and streams the reply as it comes.
- Each turn through the router is metered per person and per project, with limits and a credit balance, so a runaway agent stops instead of spending.
You decide
- Which way each session pays, and which model it uses.
- Spending limits for your team. soon
- Your own provider behind the router. soon