This page is machine-translated from French. Read the French original.
Journals and supervision
Event log
Réglages → Journal d'événements (administrator).
It records the structuring actions of the instance, along with their author and timestamp.
What is recorded
| Domain | Events |
|---|---|
| Authentication | Successful logins, failed logins, suspended accounts, SSO logins, automatic provisioning |
| Accounts | Creations, modifications, deletions, invitations issued and accepted |
| Workspaces | Creating, deleting, changing system prompts, adding and removing documents, resetting indexes |
| Documents | File imports, link imports, indexing |
| Conversations | Sent messages, duplicate threads, exports |
| Web widget | Creating, modifying, and deleting widgets |
| API | Creating and deleting keys, actions performed via the API |
| Configuration | Switching to multi-user mode, enabling beta features, clearing the log |
| Guest sessions | Logins |
The log can be filtered and purged. The purge itself is logged.
What it does not contain
THE content Messages and documents are not included. The log records that a message was sent, not its content. For the content, you need to consult the conversation history.
Conversation History
Réglages → Conversations It provides an overview of all exchanges within the instance: space, author, date, content. Export is possible.
Legitimate uses: quality analysis, identification of documentation gaps, response to an access request, investigation of an incident.
Obligation of transparency. Users must be informed that their conversations are accessible to an administrator. A setting also allows them to hide their own history, but this does not prevent them from being notified.
Execution of planned tasks
Each task retains a history of its executions: date, duration, status, detailed result.
Check regularly. A planned task can silently fail for weeks: it simply stops producing anything, and no one notices until the result is explicitly expected. → Planned tasks
System status
Several screens display the status of the instance:
- Hardware diagnostics — machine resources and suitability of local models.
- Vector base — number of indexed vectors.
- Business agents — each person's availability status.
- MCP Servers — status of declared servers.
- Update — installed version and availability of a new version.
Things to monitor during operation
| Point | Frequency | For what |
|---|---|---|
| Execution of planned tasks | Weekly | A failed task is silent |
| Storage directory disk space | Monthly | Documents, indexes, and models are piling up |
| Backups | Continue, with restoration test | An untested backup is not a backup. |
| Authentication log | Monthly | Identifying repeated failures |
| Active API keys | Quarterly | Revoke what is no longer useful |
| Active accounts | Quarterly | Suspend departures |
| Installed version | With each publication | Security patches |
Diagnosis of common incidents
| Symptom | To be checked |
|---|---|
| No further response | Model provider: expired key, quota reached, service unreachable |
| Suddenly wrong answers | A recent change in model, embedding engine, or search settings |
| Documents not found | Indexing complete? Document added to the space? Similarity threshold? |
| General slowness | Machine resources; local model too large; very large corpus |
| An agent consistently fails | Skill activated? External service reachable? Agent model sufficiently capable? |
| A user does not see a space | He is not a member |