| Command | Description |
|---|---|
psc run |
Assistant + experts + extraction + MCP server + session REPL |
psc run --poll |
Full system + expert ingesters (background polling) |
psc dev |
Local-dev monolith: Discord frontend + all services in one process |
psc dev --poll |
Monolith + expert ingesters |
psc discord start |
Discord frontend service (decomposed runtime) |
psc chat |
Direct chat (no session bus) |
psc install <path> |
Install an expert package |
psc update <name> |
Update an installed expert to new version |
psc expert list |
List installed experts |
psc expert inspect <name> |
Show expert details |
psc expert disable <name> |
Disable an expert (reversible) |
psc expert enable <name> |
Re-enable a disabled expert |
psc expert uninstall <name> |
Uninstall an expert |
psc expert auth <name> |
Run an expert's auth flow (e.g. gmailscarf) |
psc expert start-ingestion <name> |
Start an expert's ingester in the foreground |
psc expert ingest |
Standalone ingest expert (interactive mode) |
psc expert ingest --seed <file> |
Ingest a seed file |
psc expert ingest --record <file> --type <type> |
Ingest typed JSON records |
psc eval --dataset <path> |
Graph-based eval: ingest, index, query graph, score |
psc eval --dataset <path> -v |
Verbose eval with per-record detail |
psc erase-all |
Wipe all system state |
psc queue |
Curation queue summary |
psc queue list |
List up to 20 queue entries |
psc queue clear --confirm |
Clear unclaimed queue entries |
psc extraction start |
Start the extraction consumer in the foreground |
psc triage start |
Start the triage agent in the foreground |
psc curation start |
Start the curation consumer in the foreground |
psc curation status |
Show curator queue status |
psc mcp start |
Run MCP server standalone |
psc mcp status |
Show MCP server info |
psc mcp keys list |
List all API keys |
psc mcp keys create --name <n> |
Create a new API key |
psc mcp keys revoke <key_id> |
Revoke an API key |
psc query <tool> [--options] |
Call a context_query tool directly |
psc integration-test |
Smoke test all context_query tools |
All commands also available as pearscarf instead of psc.
Valid tools: find_entity, get_facts, get_connections, get_relationship, get_conflicts, vector_search.
Examples:
psc query find_entity --name "Marcus Webb"
psc query get_facts --entity-name "Marcus Webb" --edge-label AFFILIATED
psc query get_connections --entity-name "Meridian Deal"
psc query get_relationship --entity-a "Marcus Webb" --entity-b "Meridian Deal"
psc query get_conflicts --entity-name "James Whitfield"
psc query vector_search --name "contract markup"The REPL shows the active session in the prompt:
[ses_001] you > Read my latest emails
| Command | Description |
|---|---|
/sessions |
List all sessions |
/switch <id> |
Switch to a different session |
/new |
Create a new session |
/history |
Print messages in current session |
/history <id> |
Print messages in a specific session |
- Your message → assistant via Postgres
- Assistant reasons → may delegate to an expert (gmailscarf, linearscarf, githubscarf)
- Expert processes the request using its tools
- Result flows back: expert → assistant → you
Expert-initiated sessions (e.g. new email detected during polling):
--- NEW MESSAGE ses_003: assistant — New email from investor@acme.com ---
Use /switch ses_003 to interact.
- Mention the bot or DM → new session + thread
- Thread replies stay in the same session
- Expert events auto-create threads
- Sessions persist — resume by posting in the thread
| Command | Description |
|---|---|
psc memory list |
List recent records from Qdrant |
psc memory search "query" |
Semantic search across records |
psc memory entity "name" |
Look up entity + connections + fact history |
psc memory graph |
Graph overview and stats |
psc memory record <id> |
Entities and facts extracted from a record |
Same commands available in the REPL via /memory.
| Script | Description |
|---|---|
python scripts/reindex_all.py |
Reset indexed flags — extraction re-extracts on next poll |
python scripts/erase_all.py |
Wipe all system state |
python scripts/extract_test.py [ids...] |
Run extraction prompt against records (no writes) |
- Getting Started — installation, credentials, first run
- Architecture — system design, expert contract
- Building an Expert — how to create a new expert
- MCP Tools — external tool surface reference