Guides & pitfalls
Practical material on AI API relay services: how rates are calculated, how to verify a model has not been downgraded, what enterprises should check before buying, and the pitfalls we ran into while measuring these services ourselves.
Browse by topic
All articles
Self-hosting a gateway vs buying a relay service: cost and where each fits
Self-hosting sounds cheaper and more controllable, but the real cost is not the server. This article totals up both the…
Specialist vs aggregator relay services: when a specialist is the better deal
Aggregators cover more ground and are easy to sign up for; specialists have shorter chains and are usually more stable.…
Official API vs relay service: where the difference really lies, worked out
The gap is not only price. This article compares the two across cost, onboarding, reliability and data boundaries, and g…
Switching base_url in the Python SDK: five things people miss
Pointing base_url at a relay service looks like a one-line change, but five details decide whether it runs in production…
Wiring a relay service into Dify or FastGPT
Platforms like Dify and FastGPT ship with built-in model provider settings, but details differ when the endpoint is a th…
Configuring Cursor and Cline to work with a relay service
Editor-based tools ask different things of an API than command-line tools do: they are far more latency-sensitive and se…
Pointing Codex CLI at a third-party API: what to watch for
Codex CLI can talk to an OpenAI-compatible endpoint, but the authentication shape and the request format make relay serv…
Claude Code with a relay service: full configuration and troubleshooting
The setup is a handful of environment variables. Nearly every failure is one of five things, and all five are diagnosabl…
Prepaid, subscription or pay-as-you-go: where each billing model wins
The three billing models on offer look like a price question. They are really a cash-flow and risk question, and the rig…
Why reasoning models cost more than you expected: how thinking tokens are billed
Reasoning models produce tokens you never see, and those tokens are billed. This is why an estimate based on visible out…
Prompt caching on a relay service: does it actually work, and what is it worth
Cache discounts are the largest single lever on API cost for a lot of workloads. Whether a relay service passes them thr…
From list price to your invoice: a worked cost conversion
Rate, cache discounts, output weighting, tokenisers — by the time the bill arrives, the number is the product of four or…
Supporting data & tools
The articles explain the method; the pages below give you measured data and prices to check it against. Reading them together is easier — once you know how to tell whether a rate is real, you can look the official price up in the price list.