tota automatically extracts memories from your conversations and stores them in a local SQLite database with full-text search. Every session builds on what it learned before.
Memory types
| Type | Description | Persistence |
|---|---|---|
fact | A concrete piece of information about you or your context | Durable |
preference | How you like things done | Durable |
project | A project goal, structure, or constraint | Durable |
code | A code pattern, snippet, or convention | Durable |
workflow | A repeatable process or sequence of steps | Durable |
contact | A person, role, or team relationship | Durable |
event | A past event, decision, or notable moment | Durable |
goal | Something you're working toward | Durable |
context | Session context — current task, working directory | Active (short-lived) |
reminder | A time-based prompt to revisit something | Active |
How it works
- Extract — After each response, tota scans the conversation for notable information
- Score — Assigns relevance and novelty scores to candidate memories
- Recall — Before responding, retrieves the top 5 most relevant memories via FTS5
- Consolidate — Merges near-duplicates and updates stale facts
All data is local: ~/.tota/memory/. Nothing is sent to the cloud.
Manage memories
/memory # browse and manage memories
/memory list # list recent memories
/memory search <q> # full-text search
/memory delete <id> # delete a specific memory
/memory clear # clear all memories (destructive!)
⚠️
/memory clear is irreversible. All extracted memories will be deleted permanently.
Conflict resolution
When tota encounters a new fact that contradicts a stored memory, it:
- Updates the stored memory with the new value
- Keeps a note of the old value as context
- Logs the conflict for transparency
Evidence kinds tracked: conversation, explicit, inferred, confirmed.