Telemetry Endpoints

The Prism Engine serves telemetry queries over S3 Parquet data via Apache DataFusion.

GET /v1/telemetry/logs

Query log records.

GET /v1/telemetry/logs?from=2024-01-01T00:00:00Z&to=2024-01-07T00:00:00Z&limit=100
Authorization: Bearer gck_your_key

Parameter	Type	Default	Description
from	ISO8601	24h ago	Start time
to	ISO8601	now	End time
session_id	string	—	Filter to a single session
developer_id	string	—	Filter to a single developer
limit	integer	100	Max records (max 1000)
offset	integer	0	Pagination offset

GET /v1/telemetry/metrics

Query metric data points. Same parameters as /logs.

GET /v1/telemetry/traces

Query trace spans. Same parameters as /logs.

GET /v1/telemetry/stats

Aggregated statistics grouped by time, session, or developer.

GET /v1/telemetry/stats?from=2024-01-01T00:00:00Z&to=2024-01-07T00:00:00Z&group_by=day&signal=logs
Authorization: Bearer gck_your_key

Parameter	Type	Default	Description
from	ISO8601	7 days ago	Start time
to	ISO8601	now	End time
group_by	enum	hour	One of hour, day, session_id, developer_id. Invalid values return 400.
signal	enum	logs	One of logs, metrics, traces.

POST /v1/telemetry/per-turn

Query per-turn telemetry for a specific set of prompts or sessions (body-driven to avoid long query strings). Used by the dashboard’s session explorer to hydrate turn-by-turn detail.

POST /v1/telemetry/per-turn
Authorization: Bearer gck_your_key
Content-Type: application/json

{
  "session_id": "…",
  "prompt_ids": ["…"]
}

GET /v1/telemetry/analytics

Analytics with in-memory caching — tools used, errors, efficiency metrics.

GET /v1/telemetry/analytics?from=2024-01-01T00:00:00Z&to=2024-01-07T00:00:00Z&tz_offset=-300
Authorization: Bearer gck_your_key

Parameter	Type	Default	Description
from	ISO8601	7 days ago	Start time
to	ISO8601	now	End time
tz_offset	integer	0	Timezone offset in minutes east of UTC. Shifts event timestamps before bucketing so day/hour aggregates reflect local time. Example: 540 for KST, -300 for EST.

Results are cached per org_id:from:to key for fast repeated queries.

Notes

• All telemetry endpoints use DataFusion with predicate pushdown for efficient S3 Parquet scanning.
• Data is partitioned by year/month/day/hour (Hive partitioning) — narrower time ranges are faster.
• Results are scoped to the authenticated organization.