Trace Tools with cURL

Use manual logging when an operation did not pass through the Planck AI Gateway—for example, an external API, database query, or custom tool call. Send one log after the operation completes.

Production canary

This request records a synthetic tool operation. It does not call a model or a third-party service.

export PLANCK_API_KEY="sk-planck-..."
export REQUEST_ID="$(uuidgen | tr '[:upper:]' '[:lower:]')"
export NOW_SECONDS="$(date +%s)"

curl --fail-with-body --request POST \
  --url https://api.inquantum.ai/custom/v1/log \
  --header "Authorization: Bearer $PLANCK_API_KEY" \
  --header "Content-Type: application/json" \
  --header "Planck-Request-Id: $REQUEST_ID" \
  --header "Planck-Session-Id: production-manual-log-canary" \
  --header "Planck-Property-Canary: manual-log" \
  --data "{
    \"providerRequest\": {
      \"url\": \"canary://manual-log\",
      \"json\": {
        \"_type\": \"tool\",
        \"toolName\": \"production_manual_log_canary\",
        \"input\": {\"safe\": true}
      },
      \"meta\": {}
    },
    \"providerResponse\": {
      \"json\": {
        \"_type\": \"tool\",
        \"toolName\": \"production_manual_log_canary\",
        \"ok\": true
      },
      \"status\": 200,
      \"headers\": {}
    },
    \"timing\": {
      \"startTime\": $NOW_SECONDS,
      \"endTime\": $NOW_SECONDS
    },
    \"provider\": \"custom\"
  }"

A successful write returns HTTP 200 with the same request ID:

{ "status": "ok", "planck-id": "..." }

Search for $REQUEST_ID in the Requests dashboard. Ingestion can take a few seconds. Repository maintainers can run the send-and-query version with:

PLANCK_API_KEY="sk-planck-..." bash planck-gateway/scripts/test-production-manual-log.sh

Endpoint and authentication

Use the endpoint for the operation’s provider:

Provider	Endpoint
Custom tool, API, or database	`/custom/v1/log`
OpenAI	`/oai/v1/log`
Anthropic	`/anthropic/v1/log`
Google	`/googleapis/v1/log`

The base URL is https://api.inquantum.ai. Pass a Planck API key as Authorization: Bearer <key>. providerRequest, providerResponse, and valid authentication are required. timing is optional and accepts epoch seconds, epoch milliseconds, RFC 3339 strings, or { seconds, milliseconds } objects. The maximum submitted payload is 10 MB.

Storage behavior

Requests do not all go to S3. Metadata and ordinary request/response bodies use the durable logging pipeline and ClickHouse. Bodies over 10 KB and inline base64 image assets are offloaded to private object storage when it is configured; ClickHouse retains the searchable metadata and storage reference. Use Planck-Omit-Request: true or Planck-Omit-Response: true when a body must not be retained. The endpoint returns success only after the log has been accepted by a durable sink. A non-2xx response means the log must be retried if delivery is required.

Getting Started

AI Gateway

Observability & Analytics

Prompt Management

Async Logging

Custom Logs

Trace Tools with cURL

Production canary

Endpoint and authentication

Storage behavior

​Production canary

​Endpoint and authentication

​Storage behavior

Production canary

Endpoint and authentication

Storage behavior