Your team
Alfred runs five specialists that work on your vault:| Specialist | Role | When they work |
|---|---|---|
| Curator | Reads what you share and creates structured records | Automatically on new content |
| Janitor | Scans for and repairs structural issues | Periodic sweeps |
| Distiller | Surfaces deeper knowledge from your records | On-demand or scheduled |
| Surveyor | Embeds records, clusters by meaning, discovers relationships | On-demand or scheduled |
| Clerk | Stateless LLM worker dispatched for analytical tasks | On-demand by other specialists |
Checking on your specialists
From the dashboard
Go to your Alfred Black dashboard and check the Command Center — it shows the status of each specialist at a glance.Via the API
Starting, stopping, and restarting
From the dashboard
Use the controls on the Command Center or Services page to start, stop, or restart your specialists.Via the API
Curator — reads what you share
The Curator watches your inbox and turns raw content into structured vault records.Manual run
If you want the Curator to attend to inbox items immediately:Direct conversation import
Share a conversation directly (bypasses the inbox):Janitor — keeps your vault in order
The Janitor scans your vault for structural problems: broken links between records, missing metadata, and orphaned records with no connections.Run a scan
Repair found issues
After scanning, have the Janitor fix what it found:Check Janitor status
View scan history
Ignore specific files
Tell the Janitor to skip certain files during scans:Distiller — surfaces what’s hidden
The Distiller reads your vault and surfaces latent knowledge: assumptions, decisions, constraints, contradictions, and synthesized insights.Run a scan
Identify records that could yield learning records:Run extraction
Have the Distiller surface the knowledge:Check Distiller status
View extraction history
Surveyor — maps your connected world
The Surveyor takes a different approach from the other specialists. Instead of reading individual records, it runs its own four-stage pipeline: Embed (represent records as vectors), Cluster (group records by meaning), Label (name the clusters), and Write (add cluster tags and relationship links back to your vault).The Surveyor has its own dedicated pipeline and is configured separately via the Surveyor settings in your dashboard.
OpenClaw session management
Background workers and Clerk subagents create OpenClaw sessions as they carry out tasks. These sessions accumulate over time and are auto-cleaned — Alfred periodically prunes completed sessions to keep the session list manageable.The alfred-inbox hook
Thealfred-inbox hook captures completed OpenClaw chat sessions as stream events. When a session ends, the hook packages the conversation and feeds it into the Event Processor, which classifies it and writes relevant vault records. This means your conversations with Alfred’s agents contribute to your vault automatically — decisions made, tasks discussed, and context shared all become searchable records.
Manual cleanup
If you need to clean up sessions manually:Troubleshooting
Inbox not being attended to?- Check the Curator status — is it running?
- Try restarting:
POST /api/v1/workers/restart - Try a manual run:
POST /api/v1/workers/process
- Run a Janitor scan:
POST /api/v1/workers/janitor/scan - Then repair:
POST /api/v1/workers/janitor/fix
- Run a Distiller scan:
POST /api/v1/workers/distiller/scan - Then run extraction:
POST /api/v1/workers/distiller/run
Worker API Reference
Full worker endpoint documentation
Monitoring
Keep an eye on your Alfred
