Chat Completions

Request contract

Send 1–128 ordered system, user, assistant or tool messages. Use one of max_tokens or max_completion_tokens, never both. Each request is bounded by the configured model ceiling and shared serialized-input limits.

Function calls

Up to eight declared functions and bounded call results are supported. tool_choice can be auto, none, required or a named function. The caller executes returned functions; a function named shell never authorizes execution on NotHoly hosts.

Streaming

Set stream:true. Chunks preserve a stable ID and indexed tool-argument fragments. include_usage emits a final usage-only chunk. [DONE] appears only after successful completion; partial output may be followed by an error.

Response and limitations

Read choices[0].message and finish_reason. Content can be null for tool-only turns. Unknown fields and malformed tool cycles fail before dispatch. Optional effort, sampling and response-format fields require an explicitly enabled model capability; the generated schema documents their exact accepted shape.

Request example

Use your customer key and a published model. Copying an example does not send a request.

curl https://api.notholy.com/v1/chat/completions \
  -H "Authorization: Bearer $NOTHOLY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "notholy-pulse",
  "messages": [
    {
      "role": "user",
      "content": "Explain least privilege."
    }
  ],
  "max_tokens": 256
}'

Request fields

Generated from the versioned contract. Refinement rules above still apply.

Response schema