Runs

Two id shapes: a run request id covers one submission; a composite run id (<variant_id>::<hash>) names one run inside it and must be percent-encoded in paths (claude%3A%3Abaseline%3A%3A01K...). Every route takes org_id (required only when the key can see several orgs); it is not repeated below. Parameters are optional unless marked required.

GET /api/v1/runs

List platform runs, one row per run.

curl -H "Authorization: Bearer $AX_API_KEY" \
  "https://app.514.ax/api/v1/runs?experiment_id=my-first-experiment&status=completed,errored"

GET /api/v1/runs/{id}

View a run request roll-up, or one run in detail. Ids containing :: are treated as composite run ids; anything else as a run request id.

# Request roll-up
curl -H "Authorization: Bearer $AX_API_KEY" \
  "https://app.514.ax/api/v1/runs/01KTA..."

# One run
curl -H "Authorization: Bearer $AX_API_KEY" \
  "https://app.514.ax/api/v1/runs/claude%3A%3Abaseline%3A%3A01K..."

# One run's dialogue and reasoning, up to 591 turns
curl -H "Authorization: Bearer $AX_API_KEY" \
  "https://app.514.ax/api/v1/runs/claude%3A%3Abaseline%3A%3A01K...?transcript=true&include=reasoning&limit=591"

The transcript object contains ordered turns with seq, nullable ts, role, kind, and text. include_reasoning, limit, truncated, shown, total, and skipped_malformed describe completeness. An available empty conversation has no turns and transcriptStatus: "available"; unpublished conversation facts return transcriptStatus: "unavailable" without an empty transcript.

A run that recorded a time budget carries timeLimit with budgetSeconds (the budget that applied), planCapSeconds (your plan's cloud-run cap at dispatch, 0 when uncapped or unknown), and hit (whether the run ended because it ran out of time). The field is absent for runs dispatched before time limits were recorded, which is distinct from a zero budget.

guidance is an array, empty unless the run ran out of time. Each entry carries kind, title, message, actionLabel, actionUrl, planKey, and suggestedMaxTimeSeconds. Render title and message as written: they are composed per organization and per caller, and actionLabel and actionUrl are empty when the caller may not act on the suggestion. A request roll-up also carries timedOutCount, planLimitedCount, and its own request-level guidance. See Cloud run time limits.

Transcript and test payloads cannot be combined. Use the test routes below for captured tests.

POST /api/v1/runs/{id}/cancel

Cancel queued or in-flight jobs behind a run request. Run request ids only; a composite run id returns 400 invalid_argument.

curl -X POST -H "Authorization: Bearer $AX_API_KEY" \
  "https://app.514.ax/api/v1/runs/01KTA.../cancel"

POST /api/v1/runs/{id}/query

Run read-only SQL scoped to one composite run id (percent-encoded in the path). The path id and authenticated org always win over body scope fields.

curl -X POST \
  -H "Authorization: Bearer $AX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"sql":"SELECT test_name, exit_code FROM test_analysis"}' \
  "https://app.514.ax/api/v1/runs/claude%3A%3Abaseline%3A%3A01K.../query"

The response uses the same query envelope as POST /api/v1/experiments/{id}/query.

GET /api/v1/runs/{id}/tests

List captured test executions for a run request (tests across all member runs) or one composite run id.

Each test includes runId, testName, exitCode, durationMs, stdoutTail, and stderrTail; the response carries nextPageToken.

GET /api/v1/runs/{id}/tests/{name}

Fetch one captured test by composite run id and test name (both percent-encoded). Composite-run-only, because a test name can occur in several runs within one request. Missing tests return 404.