{"effective_date":"2026-10-27","entries":[{"date":"2026-10-27","title":"Agent streaming, budgets and receipts","summary":"The agent endpoint now streams, agent keys can carry daily and monthly spend caps, and every call can be reconciled to a receipt. Python and TypeScript SDKs are published.","changes":["POST /v1/agent/chat/completions accepts stream: true and emits a final receipt event.","Agent keys accept daily_cap_usd and monthly_cap_usd, enforced before the upstream call.","Quotes return a quote_id that can be passed to the call. Receipts are available at /v1/agent/receipts/{id}.","Python and TypeScript SDKs cover quotes, streaming, budgets, receipts and agent account creation.","MCP and A2A remain non-streaming. Agent chat is capped at 4,096 output tokens."],"links":[["Agent account","/agents"],["SDK guide","/sdk"],["Streaming and budgets","/blog/agent-streaming-budgets"]]},{"date":"2026-10-27","title":"The agent-native inference account","summary":"Fikra API is now positioned around one product: an inference account an agent can open, fund, quote, use and audit. Public pages, pricing, limits and model metadata are being consolidated to one source of truth.","changes":["New accounts are prepaid. Existing subscriptions are honored and remain available in the dashboard.","Paystack card and mobile money checkout is unchanged, including KES settlement.","Rate limits are published once: 30 RPM unpaid, 3,000 RPM paid.","Model pages disclose whether an ID is a native model or a stable Fikra alias.","Case studies built on projections and unverifiable testimonials were removed.","The trust page now states actual retention, SLA and compliance status."],"links":[["What Fikra is","/what-is-fikra"],["Pricing","/pricing"],["Agent account","/agents"],["Trust centre","/trust"]]},{"date":"2026-10-27","title":"Price and limit consistency","summary":"Every price shown on the website is read from the live model registry. Hardcoded prices, duplicate limits and the Qwen cached-input inversion were removed.","changes":["DeepSeek V4.1 Flash 1017 is published at $0.05 input, $0.15 output, $0.02 cached input per 1M tokens.","Cached input can never cost more than fresh input.","The docs quickstart uses a real model ID."],"links":[["Live pricing","/pricing"],["Machine-readable prices","/pricing.md"],["Model catalog","/models"]]}]}