Ingest observations
Record model calls in the conversation format without sending traces. Each call is stored in its conversation and evaluated by the project's conversation judges, as calls recorded from traces are. Send at most 100 observations per request. Retrying a request with the same observations and timestamps stores each call once.
AuthorizationBearer <token>An ingest API key from Settings > Ingest API Keys, sent as Bearer <key>. It belongs to one organization, so no X-Organization-Id is needed.
projectId*string^([0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12})$uuidapplication/json- body
observations*array<>1 <= items <= 100Response for status 200
application/json- response
status*stringcurl -X POST 'https://api.judgmentlabs.ai/v1/projects/YOUR_PROJECT_ID/observations' \ -H 'Content-Type: application/json' \ -d '{ "observations": [ { "observation": { "conversation_id": "string", "input": [ { "type": "message", "role": "system", "content": [ { "type": "text", "text": "string" } ] } ] }, "observed_at": "2019-08-24T14:15:22Z" } ]}'{ "status": "success"}Add trace tags POST
Add tags to a trace. Reference for the POST /v1/projects/{projectId}/traces/{traceId}/tags endpoint in the Judgeval API.
Get trace health summary GET
Get project check status and per-check failure rates among evaluated traces. Reference for the GET /v1/projects/{projectId}/trace-health/summary endpoint in the Judgeval API.