Create a chat completion

Send messages

Call POST /v1/chat/completions with the exact catalog model identifier and a messages array. Keep instructions concise and validate application inputs before sending them.

const result = await client.chat.completions.create({
  model: 'MODEL_ID_FROM_CATALOG',
  messages: [{ role: 'user', content: 'Summarize this text.' }],
})

Handle streaming deliberately

Set stream to true only when the selected model and client path support it. Accumulate partial output safely, handle disconnects, and do not assume a failed stream can be retried without duplicate work.

Record request identifiers

Log the HippoAPI request identifier with your internal trace, not the API key or full sensitive prompt. The identifier, time, endpoint, model, and status are the most useful fields for troubleshooting.