| name | automem |
| description | Persistent AutoMem memory via mcporter-exposed AutoMem tools. |
| user-invocable | true |
| metadata | {"openclaw":{"skillKey":"automem","primaryEnv":"AUTOMEM_API_KEY","requires":{"env":["AUTOMEM_API_URL"]}}} |
AutoMem
Use the typed AutoMem tools exposed through mcporter.
Natural language mappings
remember ... or store this -> automem_store_memory
what do you know about ... or recall ... -> automem_recall_memory
update memory ... -> automem_update_memory
delete memory ... -> recall first when needed, then automem_delete_memory
link these memories ... -> automem_associate_memories
is memory healthy? -> automem_check_health
Slash command behavior
Interpret /automem remember ..., /automem recall ..., /automem update ..., and /automem delete ... as requests to use the matching AutoMem tool flow.
Rules
- Recall preferences first with
tags: ["preference"], sort: "updated_desc", and format: "detailed" when collaboration style or user habits matter.
- For task context, prefer one semantic query built from the user's actual nouns. Do not hard-gate recall with default tags unless the conversation is clearly scoped to an unambiguous project slug.
- For debugging, recall with the error symptom as a semantic query and NO tags — bugfix/solution tagging is incomplete, and a tag gate hides cross-corpus fixes.
- Tags are a hard gate. Use bare tags only, and avoid platform tags like
openclaw.
- Store only durable information worth reusing later.
- Default project tags are for stored memories. Recall should stay semantic unless tags are explicitly needed.
- Use
memory-core and file-backed workspace memory for local notes and raw transcripts; AutoMem is the semantic cross-session layer.
- If delete targets are ambiguous, show the likely matches and ask for confirmation before deleting.
- Do not fall back to raw curl commands in this mode unless the user explicitly asks for the legacy setup.