Practical notes for building and using AI API integrations.
These guides explain the protocol, billing, and reliability details that tend to cause real client failures. Each guide links to the exact CloudService documentation needed to configure or verify the behavior.
| Guide | Category | Read time |
|---|---|---|
| OpenAI-compatible vs Anthropic Messages APIs Compare request shapes, base URLs, streaming events, tool calls, and the client choices that determine which protocol to use. Protocol guide · 8 minute read | Protocol guide | 8 minute read |
| API credits vs token packs Understand what each billing mode measures, how cache tokens are counted, and which base URL belongs to each key. Billing guide · 7 minute read | Billing guide | 7 minute read |
| Stop retries from multiplying AI token usage See how client and gateway retries combine, where idempotency belongs, and why a stream must never be retried after visible output. Reliability guide · 9 minute read | Reliability guide | 9 minute read |