Overview
Sessions represent units of work captured from your AI coding tools. Events from the MCP server, CLI, and VS Code extension are grouped into sessions. Each session can be enriched to extract structured memory items.
Viewing sessions
Sessions are accessible from two places: the Memory page and the Operations page. Both show the same data. Each session card displays:
- Source -- Icon and label showing where events came from (MCP, CLI, VS Code, Scan)
- Status -- Current state (active, completed, enriching, failed)
- Event count -- Number of raw events captured
- Created at -- When the session started
- Completed at -- When the session ended
- Items extracted -- Number of memory items generated (after enrichment)
Click a session to view its full details, including raw events, enrichment jobs, and the pipeline timeline.
Session detail view
Clicking a session opens the detail view with several sections:
Pipeline Timeline
A visual progress indicator at the top shows the current enrichment stage. Each stage displays its status (pending, active, completed, failed) and duration. The stages are: Enrich → Deep Enrich → Embed → Dedup → Drift Check.
Events tab
Browse all raw events captured during the session:
- Event type -- Session save, scan result, git digest, etc.
- Timestamp -- When the event was captured
- Payload summary -- Preview of the event data
- Source -- Which tool captured the event
These are the raw data that the enrichment pipeline processes.
Jobs tab
View enrichment job history:
- Job status -- Queued, processing, completed, or failed
- Stage progress -- Per-stage timing and completion status
- Items extracted -- Number of memory items generated
- Error details -- Full error information for failed jobs
Session lifecycle
Sessions follow this lifecycle:
- Active -- Events are being captured. Sessions are created automatically when events arrive and reuse an existing session within a 4-hour window
- Completed -- The session has ended (either manually closed from the dashboard or by timeout)
- Enriching -- The enrichment pipeline is processing the session
- Completed (enriched) -- Enrichment finished successfully with extracted items
- Failed -- Enrichment encountered an error
When you close an active session from the dashboard, the VS Code extension automatically creates a new session for subsequent activity.
Triggering enrichment
Enrichment can be triggered in two ways:
Automatic enrichment
When the VS Code extension captures git commits, enrichment is triggered automatically if Auto-Learn from commits is enabled in your Project Settings. This is the default behavior. To disable it and save credits, toggle it off in Project Settings > General.
Manual enrichment
To manually generate memory from a session's raw events:
- Find the session in the Sessions tab
- Click the session to open its detail view
- Click the Generate Memory button
- Watch the pipeline progress through the visual timeline
Pipeline stages
The enrichment pipeline processes events through up to five stages:
| Stage | Description |
|---|---|
| Enrich | AI processes event chunks (10 events per chunk) and extracts memory items |
| Deep Enrich | If a GitHub repo is linked, reads source files and enriches items with implementation details (routes, schemas, signatures). Uses Devstral Small. Skipped if no repo or toggle disabled. |
| Embed | Generates vector embeddings for semantic search |
| Dedup | Detects and merges duplicate items |
| Drift Check | Flags items as review when their referenced files were modified by recent commits |
AI model selection
The AI model used for enrichment depends on your plan. All paid plans can override the model per-project:
| Plan | Default | Override Available |
|---|---|---|
| Free | Small | No |
| Personal | Small | Small or Medium |
| Team / Business / Enterprise | Medium | Small or Medium |
To change the model, open the Projects page, click the gear icon on your project card, and select the model in the AI Model section.
Retrying failed enrichment
If enrichment fails due to a transient error, click the Retry button on the failed job in the session detail view. The pipeline restarts from the failed stage, preserving progress from earlier stages. Common reasons for failure include:
- Temporary API errors
- Rate limiting on the AI model
- Network timeouts (mitigated by automatic retry with direct Mistral fallback)
Viewing extracted items
After enrichment completes, switch to the Brain tab in the Memory page to browse the newly extracted memory items. Items are organized by schema key and display confidence scores, tier badges, and state indicators.
If the drift check flagged any existing items, they appear in the Review tab for triage.
Next steps
- Memory -- Browse and manage the extracted memory items
- Enrichment Guide -- Understand the enrichment process in depth
- V2 Pipeline Guide -- Full event flow from capture to memory