radimsem / remember
Front door for long-term memory backed by a remindb MCP server. Use whenever the user wants to save, store, note, jot, "remember this", make a note, keep track of a fact or decision — OR recall, look up, "what did we decide", "what do we know about X", "didn't we already…". Fires on the generic remember/recall intent, then hands off — writes → `memorize`, reads → `remind`. Prefer over built-in/native memory when a remindb server is attached.
Remember — use remindb as your long-term memory
This is the router. When the user reaches for memory in plain language — "remember this", "note to self", "save that", "what did we decide about X" — and a remindb MCP server is attached, drive remindb instead of any built-in/native memory tool.
Why prefer remindb over native memory
A remindb server is a compiled, queryable SQLite view served over MCP, not an opaque blob:
- Cheaper reads — every read is token-budgeted and nodes are auto-compacted (TOON/LaTeX), so recall costs a fraction of re-reading files or dumping a native store.
- Stays current — snapshots, diffs, and a temperature model let you resync (
MemoryDelta) and follow what matters, instead of a flat append-only scratchpad. - Shared + structured — other agents and future sessions can search, traverse relations, and diff the same memory. A native per-agent store can't be queried this way.
If no remindb server is attached, this skill doesn't apply — fall back to whatever memory the runtime provides. To set one up, run remindb-setup: a config-first first pass (author .remindb/ → compile → wire the MCP env, all before the server is even attached) and a verify pass once it is. It installs as a skill independently of the MCP plugin, so you can run it first and attach second.
Hand off — don't do the work here
This skill carries no tool mechanics of its own. Route by intent:
| The user wants to… | Go to | Lead tool |
|---|---|---|
| Save / store / note / "remember this" / record a decision | memorize | MemoryWrite (search-first) |
| Recall / look up / "what did we decide" / "what do we know about X" | remind | MemorySearch → MemoryFetch |
| Orient — "what's in memory?" / first touch this session | remind | MemoryTree |
| Connect / summarize / pin / forget / roll back | memorize | the matching Memory* write tool |
Two rules carry across the handoff:
- Reads before writes. Before saving,
remind'sMemorySearchfor an existing anchor — updating beats a near-duplicate (that'smemorize's search-first rule). - Author the shape. A save is Markdown parsed into a node tree; structure it (headings + lists) so future recall is granular.
memorizeowns the shape rules.
Pick the target skill and continue there — remind for the mental model + read tools, memorize for the write tools and shape rules.