Engineering notes and product changes. Older posts carry a current-facts banner because prices and plans change.
Prepaid for new accounts, existing subscriptions honored, Paystack unchanged, and one clear definition of the product.
Read ->The agent chat endpoint now streams, keys can carry daily or monthly caps, and every call can be reconciled to a receipt.
Read ->Fake accounts, bounces, proof-of-work and automatic source blocks.
Read ->MiMo leads at 80.4. Extraction is the weak category. Step 5 is still a failure.
Read ->Apple Pay and USD card checkout are retired. Both remaining methods are charged in KES at KES 133 = $1 and credit the USD wallet.
Read ->Cached-input prices for every DeepSeek route now sit at the 20% margin floor, and /stats publishes measured reliability, latency and cache hit rates.
Read ->Why API key provisioning is friction, and how delegated accounts, quotes, budgets and receipts let an agent work without a human in the loop.
Read ->How MCP works, how it differs from REST and A2A, and how Fikra exposes 65 tools over one JSON-RPC endpoint.
Read ->How an AI agent opens a Fikra API account for a human, funds it with a checkout link, an M-Pesa prompt or a saved card, and hands over ownership.
Read ->We corrected the context windows and maximum output budgets for the DeepSeek, GLM, Kimi, Qwen and MiMo lanes. The gateway now reads these limits from the model registry on every request, so large max_tokens requests are no longer clamped by a stale default.
Read ->Yesterday paid accounts got 20M tokens per minute and 1B per day. As of today the token counters are off: the only metered numbers are requests per minute and requests per day.
Read ->Response caching now runs by default on deterministic requests, audio rides into any chat model via server-side transcription, and we publish the results of our own limits stress test: 58M tokens in minutes, no throttling. Plus what is next.
Read ->Claude Fable 5 and Mythos 5 are engineering marvels, but at $50 per million output tokens, they are financial suicide for standard SaaS production workloads.
Read ->Tired of credit card declines and aggressive $20 minimum top-ups? Here is an honest, benchmarked breakdown of the top inference APIs built for regional access.
Read ->Discover how currency fluctuations, dollar-denominated billing blocks, and high-latency round-trip hops over Western servers throttle local developer momentum.
Read ->A technical look under the hood at tokenization inefficiencies, colloquial language mappings (Sheng), and Swahili LLM fine-tuning structures.
Read ->