REST Ingestion API

Send LLM exchange data to Ingate asynchronously, from sidecars, background jobs, or any application that doesn't use the proxy.

Overview

The proxy captures exchanges automatically, but some architectures can't route through a proxy. The Ingestion API lets you send exchange data to Ingate after the fact via REST. Exchanges are processed through the same pipeline as proxy traffic. Extraction, logging, and eval hooks all run.

Single Exchange

POST
/api/v1/ingest

Ingest a single LLM exchange. Requires member role.

bash
curl -X POST https://api.ingateai.com/api/v1/ingest \
  -H "X-Ingate-Key: sk-ingate-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "openai",
    "path": "/v1/chat/completions",
    "method": "POST",
    "status_code": 200,
    "latency_ms": 432,
    "request_body": {
      "model": "gpt-4o-mini",
      "messages": [{"role": "user", "content": "Hello"}]
    },
    "response_body": {
      "choices": [{"message": {"content": "Hi there!"}}],
      "usage": {"prompt_tokens": 10, "completion_tokens": 5}
    }
  }'
jsonResponse (202 Accepted)
{
  "request_id": "generated-uuid",
  "status": "accepted"
}

Batch Ingestion

POST
/api/v1/ingest/batch

Ingest multiple exchanges in one call.

bash
curl -X POST https://api.ingateai.com/api/v1/ingest/batch \
  -H "X-Ingate-Key: sk-ingate-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "exchanges": [
      {
        "provider": "openai",
        "path": "/v1/chat/completions",
        "method": "POST",
        "status_code": 200,
        "latency_ms": 350,
        "request_body": {...},
        "response_body": {...}
      },
      {
        "provider": "anthropic",
        "path": "/v1/messages",
        "method": "POST",
        "status_code": 200,
        "latency_ms": 520,
        "request_body": {...},
        "response_body": {...}
      }
    ]
  }'
jsonResponse (202 Accepted)
{
  "accepted": 2,
  "rejected": 0,
  "items": [
    { "index": 0, "request_id": "uuid-1", "accepted": true },
    { "index": 1, "request_id": "uuid-2", "accepted": true }
  ]
}

Server-Side Extraction

When full request_body and response_body are provided, Ingate runs server-side extraction to automatically populate:

  • model: extracted from the request body
  • prompt: user message text
  • completion: assistant response text
  • token_count: from response usage object
  • metadata: finish reason, stop reason, etc.

Pre-extracted fields

You can also send pre-extracted fields (model, prompt, completion,token_count) directly. If server-side extraction succeeds, it overwrites pre-extracted values. If it fails, your pre-extracted values are used as fallback.

Supported API Shapes

Server-side extraction supports:

  • OpenAI Chat Completions (/v1/chat/completions)
  • OpenAI Responses API (/v1/responses)
  • Anthropic Messages (/v1/messages)
Previous
Bring Your Own Storage
Next
OpenTelemetry