System Health
Understand the agent pipeline and what to do when your system shows a degraded status.
The Health page shows the status of the most recent run for each background agent. It is the first place to check if your AI is behaving unexpectedly.
Opening the Health page
Click Health in the left sidebar, or click the System Health card on the Dashboard.
What are agents?
Basegent runs several background agents on a schedule. They run automatically — you do not need to trigger them manually.
| Agent | What it does |
|---|---|
| compiler | Reads ingested raw content and writes structured wiki entries; updates the master search index |
| linting | Detects contradictions between entries, runs web verification, and flags quality issues |
| learning | Updates confidence scores for each entry based on query feedback; identifies knowledge gaps |
| linker | Maps relationships between wiki entries for better context during answering |
| ingestion | Normalises and pre-processes newly submitted raw content before the compiler runs |
| content-sync | Re-crawls your connected content sources for new or changed pages |
Status indicators
Each agent shows one of three statuses:
| Status | Meaning |
|---|---|
| Healthy | Last run completed without errors within the expected window |
| Degraded | Last run completed with warnings, or the agent is overdue |
| Unknown | No run recorded — agent has never run or is failing silently |
Run history
Click an agent name to expand its run history. Each row shows:
- Timestamp — when the run started
- Duration — how long it took
- Status — success, warning, or error
- Details — error message or summary
What to do when an agent is degraded
compiler degraded
New wiki entries may not be searchable. Check if a recent ingestion produced malformed content. Wait for the next scheduled run; the compiler runs on demand as content is submitted.
linting degraded
Quality checks are delayed. This is usually a temporary LLM provider timeout. It resolves on the next run without action needed.
learning degraded
Confidence scores may be stale. This is normal if query volume is very low. Once customers start asking questions the learning agent will activate.
content-sync degraded
One of your connected sources may be returning errors. Go to Sources, check that each source URL is still accessible, and remove or re-add the failing source.
ingestion degraded
Raw content submitted via Ingest may be stuck in the queue. Check that your LLM API key (GROQ) is valid in Settings.
Triggering a manual run
You can trigger any agent manually from the Health page by clicking the Run now button next to the agent. This is useful after fixing a configuration issue.