Usage analytics (Enterprise Edition)
The Analytics section of the Admin panel shows platform usage over time. The default lookback window is 30 days; you can set a custom date range.Query analytics
Daily counts of total queries, thumbs-up feedback, and thumbs-down feedback across all users.
Active users
Daily count of unique active users, based on query activity.
Assistant analytics
Per-assistant message counts and unique user counts over the selected time range.
OnyxBot analytics
For Slack-connected deployments: daily query totals and auto-resolved query counts.
Query history
The Query History page lets admins review all chat sessions across the deployment.Filter sessions
Filter by date range or by feedback type (thumbs up / thumbs down) to find sessions of interest.
Inspect a session
Click any session to see the full message thread, the assistant used, and any feedback the user submitted.
Query history can be configured to anonymize or fully disable user attribution. If anonymization is enabled, all user emails are replaced with
anonymous@anonymous.invalid in the history view and exports. If history is disabled, the Query History page is not accessible.Chat feedback
Users can rate individual responses with a thumbs up or thumbs down directly in the chat interface. Admins can see aggregate feedback counts in the Analytics section and filter query history by feedback type. Feedback signals help identify:- Assistants or connectors producing low-quality results
- Topics where Onyx’s answers are frequently unhelpful
- Regression after configuration changes
Connector health
The Connectors section of the Admin panel shows the current status of every connector.Sync status
Sync status
Each connector displays its current state: Running, Completed, Failed, or Paused. The last sync time and the number of indexed documents are shown for each connector.
Error logs
Error logs
Failed connectors display the error encountered during the last sync attempt. Click into a connector to see detailed error messages and the affected document count.
Document counts
Document counts
Each connector shows how many documents have been indexed. Use this to verify that a new connector has fully synced or to detect unexpected drops in document count after a sync.
Indexing attempts
Indexing attempts
The connector detail view lists recent indexing attempts with timestamps, durations, and outcomes. You can cancel an in-progress indexing attempt from this view.
System info panel
The System Info page (Admin panel → System Info) provides a snapshot of the current deployment state, including service versions and configuration details useful for troubleshooting.Prometheus metrics
Onyx exposes Prometheus metrics atGET /metrics on the API server. All custom metrics use the onyx_ prefix.
HTTP request metrics
| Metric | Type | Labels | Description |
|---|---|---|---|
http_requests_total | Counter | method, status, handler | Total request count |
http_request_duration_seconds | Histogram | method, handler | Request latency by handler |
http_request_duration_highr_seconds | Histogram | (none) | High-resolution latency across all handlers |
http_request_size_bytes | Summary | handler | Incoming request content length |
http_response_size_bytes | Summary | handler | Outgoing response content length |
http_requests_inprogress | Gauge | method, handler | In-flight requests |
onyx_api_slow_requests_total | Counter | method, handler, status | Requests exceeding the slow request threshold (default 1 second) |
SLOW_REQUEST_THRESHOLD_SECONDS environment variable.
Database pool metrics
| Metric | Type | Labels | Description |
|---|---|---|---|
onyx_db_pool_checked_out | Gauge | engine | Currently checked-out connections |
onyx_db_pool_checked_in | Gauge | engine | Idle connections in the pool |
onyx_db_pool_overflow | Gauge | engine | Overflow connections beyond configured pool size |
onyx_db_pool_size | Gauge | engine | Configured pool size |
onyx_db_pool_checkout_total | Counter | engine | Total connection checkouts |
onyx_db_pool_checkin_total | Counter | engine | Total connection checkins |
onyx_db_pool_connections_created_total | Counter | engine | Total new connections created |
onyx_db_pool_invalidations_total | Counter | engine | Total connection invalidations |
onyx_db_pool_checkout_timeout_total | Counter | engine | Total checkout timeouts |
onyx_db_connections_held_by_endpoint | Gauge | handler, engine | Connections currently held, by endpoint |
onyx_db_connection_hold_seconds | Histogram | handler, engine | Duration a connection is held per endpoint |
engine label is one of sync, async, or readonly.
Celery task metrics
| Metric | Type | Labels | Description |
|---|---|---|---|
onyx_celery_task_started_total | Counter | task_name, queue | Total tasks started |
onyx_celery_task_completed_total | Counter | task_name, queue, outcome | Total tasks completed (outcome: success or failure) |
onyx_celery_task_duration_seconds | Histogram | task_name, queue | Task execution duration |
onyx_celery_tasks_active | Gauge | task_name, queue | Currently executing tasks |
onyx_celery_task_retried_total | Counter | task_name, queue | Total task retries |
onyx_celery_task_revoked_total | Counter | task_name | Total cancelled tasks |
onyx_celery_task_rejected_total | Counter | task_name | Total tasks rejected by workers |
Indexing pipeline metrics
| Metric | Type | Labels | Description |
|---|---|---|---|
onyx_indexing_task_started_total | Counter | task_name, source, tenant_id, cc_pair_id | Indexing tasks started per connector |
onyx_indexing_task_completed_total | Counter | task_name, source, tenant_id, cc_pair_id, outcome | Indexing tasks completed per connector |
OpenSearch metrics
| Metric | Type | Labels | Description |
|---|---|---|---|
onyx_opensearch_search_client_duration_seconds | Histogram | search_type | End-to-end client-side search latency |
onyx_opensearch_search_server_duration_seconds | Histogram | search_type | Server-side search execution time |
onyx_opensearch_search_total | Counter | search_type | Total search requests sent to OpenSearch |
onyx_opensearch_searches_in_progress | Gauge | search_type | Currently in-flight searches |
Useful PromQL queries
Token rate limiting
Token rate limits cap the number of LLM tokens a user or group can consume within a rolling time window. This prevents any single user or team from monopolizing the LLM budget. Rate limits are configured in Admin panel → Token Rate Limits.- User-level limits
- Group-level limits
Apply a token budget that each individual user cannot exceed within the configured time window. This is a global limit that applies to all users equally.Navigate to Token Rate Limits → Users and click Add Limit. Set the token count and the time period (e.g. 50,000 tokens per hour).
Token rate limits are an Enterprise Edition feature. Curators can view the rate limits for their assigned groups but cannot create or modify them.
Celery queue monitoring
Onyx uses Celery background workers for connector syncing, document processing, and other async tasks. Worker types include the primary worker, docfetching, docprocessing, light, heavy, monitoring, and beat (scheduler). Theonyx_celery_tasks_active and onyx_celery_task_completed_total metrics give a real-time view of queue activity per worker queue. High values for onyx_celery_task_retried_total or onyx_celery_task_rejected_total on a given queue can indicate a worker is overloaded or misconfigured.