Nmem Fs Explore

Nowledge FS Explorer

Use this skill when you need to navigate the virtual filesystem mem_fs or CLI command nmem fs to find, inspect, or manage knowledge.

File System Structure

Nowledge FS projects data as a virtual file tree:

• /memories/: Durable personal memories.
  • /memories/by-id/: All memories indexed by their canonical ID (e.g. by-id/<uuid>.memory.md).
  • /memories/by-date/: Memories organized chronologically.
  • /memories/by-label/: Memories grouped by label directories.
  • /memories/by-type/: Memories grouped by unit types (e.g. fact, preference, decision, plan, procedure, learning, context, event).
• /threads/: Saved conversation threads (e.g. by-id/<thread-id>.thread.jsonl).
• /wiki/: Read-only wiki entities, topics, and crystal syntheses.
• /working-memory/: The active briefing surface.
• /context/: Startup rule assets and profile contexts.
• /artifacts/: Documents, outputs, and uploaded library resources.
• /skills/: Currently active user/system skills.

Best Practices

1. Start with capabilities: Run capabilities --json (or mem_fs caps) to see roots, limits, and supported verbs.
2. Orient with metadata (stat & ls):
   • Run ls PATH to explore directories and find files.
   • Run stat PATH to inspect metadata (type, size, update time) without loading file bodies. Always do this for large threads/documents.
3. Limit context bloat (cat --line --lines):
   • For long threads or files, do not load the whole body. Use windowed reading: cat PATH --line START --lines COUNT to inspect only the required window.
4. Search and Locate:
   • Use recall QUERY --in /memories for semantic searching of memories.
   • Use find PATH --unit-type TYPE --label LABEL --since YYYY-MM-DD --mentions ENTITY to search by structure/metadata.
   • Use grep QUERY PATH for exact-string matching across memories, threads, and artifacts (case-insensitive by default; add -E for regex or --case-sensitive).
   • Use grep QUERY PATH --jsonl to stream matching objects with both path and line numbers for automated piping into cat.
5. Write canonical files:
   • Canonical writable paths (/memories/by-id/<id>.memory.md) can be written or deleted.
   • Alias and derived paths (/memories/by-label/, /memories/by-type/, /memories/by-date/) stay read-only. Edit memories through their canonical /memories/by-id/... path.
6. Knowledge & Wiki Export:
   • Inspect /wiki/ entities, topics, and crystal pages to view compiled reference knowledge.
   • Use Open Knowledge Format (OKF) or Wiki exports (/api/library/okf-export or /api/library/wiki-export) to bundle synthesized knowledge graphs into portable Markdown sets for the local workspace.

Commands Reference

Run the MCP tool mem_fs with the command argument (or nmem fs <command> in CLI):

• capabilities or caps: Discover roots, verbs, limits, and flags.
• ls /memories/by-type/procedure: List directory contents.
• ls /wiki/: List compiled wiki entities, topics, and crystals.
• stat /threads/by-id/<thread-id>.thread.jsonl: Check metadata/line count without reading body.
• cat /memories/by-id/<uuid>.memory.md: Read file body and frontmatter.
• cat /threads/by-id/<thread-id>.thread.jsonl --line 50 --lines 20: Read a 20-line window of a thread.
• cat /wiki/<entity-or-topic>.md: Inspect synthesized wiki article.
• find /memories --unit-type decision --label project-x --since 2026-01-01: Find specific memories.
• grep "JWT rotation" /threads --jsonl: Exact string search with path and line numbers.
• grep -E "JWT|token" /threads: Regex search across thread conversations.
• recall "session state strategy" --in /memories: Find relevant memories semantically.
• write /memories/by-id/<id>.memory.md --body "...": Write a canonical memory.
• rm /memories/by-id/<id>.memory.md: Delete a canonical memory (always require explicit user confirmation before deletion).