Chat Completions

OpenAI-compatible endpoint. Every call passes through the governance pipeline.

`POST /v1/chat/completions`

curl http://localhost:9620/v1/chat/completions \
  -H "Authorization: Bearer tp_..." \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-4o-mini", "messages": [{"role": "user", "content": "Hello"}]}'

Request body

Standard OpenAI format: model, messages, temperature, max_tokens, stream, tools, etc.

Governance headers

Header	Description
`X-TapPass-Flags`	Governance flags (`mode=observe, pii=mask`)
`X-TapPass-Session`	Session ID for multi-turn taint tracking

Response

Standard OpenAI ChatCompletion response, plus:

x-tappass-classification header with data classification
x-tappass-token header with signed capability token
x-tappass-pipeline-steps header with step count

Streaming

"stream": true is fully supported. The response is buffered, scanned, and re-chunked.

Anthropic

POST /v1/messages: Anthropic Messages API format, same governance pipeline.