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
Agency pricing and volume discounts: how to negotiate, and how far it goes
Once monthly volume passes a certain point, list price stops being the only option. This article covers where discounts…
What caching actually saves: a reproducible side-by-side test
Caching removes most of the cost of repeated input, but whether a relay passes the cache fields through varies a lot. Th…
Image generation on relay sites: how DALL-E, Flux and Midjourney actually differ
Image generation comes in three completely different interface shapes on relay sites, billed three different ways. This…
Speech endpoints (TTS and STT) on relay sites: what exists and what to test
Speech endpoints are far less standardised than text. This article covers the three interface shapes and the four things…
Embedding through a relay: the four traps RAG projects fall into
Embedding endpoints look trivial, but switching supplier carries hidden risk. This article lists four traps that only su…
Image input (vision) on relay sites: what actually works, and how to test it
Vision support varies far more across relay sites than text does. This article covers the two ways to send an image, thr…
Cherry Studio, Chatbox and LobeChat: base_url configuration for desktop clients
Desktop clients are the simplest to configure and the easiest to get subtly wrong. This article covers the setup points…
Trae, Windsurf and Kilo Code: how base_url configuration differs across the new IDEs
All of these IDEs accept a custom API, but which protocols they support varies widely. This article compares their setup…
Configuring a custom API in Cursor, and the four errors you will hit
Cursor's custom API path hides a few switches that are easy to miss. This article gives the full setup and explains four…
Claude Code Router in practice: one entry point, several upstreams
Claude Code speaks Anthropic protocol only. A router lets it reach several upstreams without touching the client, and sp…
JSON mode and structured output: which relay sites really support it, and how to test
Plenty of sites claim JSON mode, but few return valid JSON reliably. This article gives a test suite with unusually high…
Context length and truncation: why relay sites error out sooner than you expect
A site advertises 200K context and errors at 80K. This article covers the five ways relay sites limit context, and how t…
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.