Guides
Guides
The four problems that come up once the first request works.
Long context
Sending hundreds of thousands of tokens without the connection dying mid-read.
Streaming
Server-sent events, retries, and what not to do with a partial answer.
Tool calling
Giving the model functions, and building a loop that survives a bad call.
Working with documents
Extraction, classification and retrieval over material that is long and messy.
Each guide is short on purpose: the thing that breaks, why it breaks, and the configuration or the code pattern that stops it.