By James Miano. Published 27 October 2026.
POST /v1/agent/chat/completions now accepts stream: true. It returns Server-Sent Events, masks the upstream model, and emits a final receipt event with usage and cost. Agent chat is still capped at 4,096 output tokens. MCP and A2A remain non-streaming.
Agent-created keys accept daily_cap_usd and monthly_cap_usd. A key over its cap returns HTTP 402 before the upstream call runs. The same cap can be updated later with POST /v1/agents/keys/{id}/budget.
A quote returns a quote_id. Pass it to the call with quote_id or X-Fikra-Quote-Id. The response carries a receipt id, and GET /v1/agent/receipts/{id} returns the model, usage, quoted price and actual cost. Receipts do not contain prompt or response bodies.
Python and TypeScript clients cover quotes, streaming, receipts, agent account creation and per-key budgets. See /sdk.