Capture what matters before the session ends. Save decisions, procedures, and learnings as durable memories that any connected AI tool can find later.
What to distill
- Decisions with rationale ("we chose PostgreSQL because ACID is required")
- Procedures: non-obvious steps, workarounds, setup sequences
- Learnings: surprises, gotchas, corrections to prior assumptions
- Preferences: user's stated preferences for future reference
- Plans that future sessions will need to resume
- Context that would be lost when the session ends
What to skip
- Routine code changes, obvious fixes, standard operations
- Work in progress that will change before it matters
- Generic information already widely known
- Simple Q&A answerable from documentation
Workflow
If this session already exposes the Nowledge Mem MCP server:
- Search first with
memory_search to avoid duplicates.
- If an existing memory captures the same concept, use
memory_update.
- Otherwise, create it with
memory_add.
Otherwise:
- Search first to avoid duplicates:
nmem --json m search "concept"
- If an existing memory captures the same concept, update it:
nmem --json m update <memory_id> -c "updated content"
- Otherwise, create a new memory:
nmem --json m add "content" -t "Title" --unit-type decision -l "label" -s codex -i 0.8
At the end of a substantial task, explicitly check whether one durable memory should be added or updated. Do not skip that review just because the user did not ask.
Unit types
fact, preference, decision, plan, procedure, learning, context, event
Additional flags
- Temporal context:
--when past (or present, future, timeless)
- Event date:
--event-start 2026-03 (for time-anchored knowledge)
Importance scale
0.8-0.9: Major decisions, important procedures, breakthroughs
0.5-0.7: Useful patterns, conventions, secondary decisions
0.3-0.4: Minor notes, preferences, contextual observations
1.0: rare, only for critical corrections or pivotal choices
After saving
Report what was stored, which unit type was used, and why each memory was worth keeping. One strong memory is better than three weak ones.
Links
1---2name: distill-memory-43description: Save key decisions, procedures, or learnings so the user never has to rediscover them. Trigger when a significant decision is made, a non-obvious procedure is found, or a lesson is learned. Do not wait to be asked.4---56Capture what matters before the session ends. Save decisions, procedures, and learnings as durable memories that any connected AI tool can find later.78## What to distill910- **Decisions** with rationale ("we chose PostgreSQL because ACID is required")11- **Procedures**: non-obvious steps, workarounds, setup sequences12- **Learnings**: surprises, gotchas, corrections to prior assumptions13- **Preferences**: user's stated preferences for future reference14- **Plans** that future sessions will need to resume15- **Context** that would be lost when the session ends1617## What to skip1819- Routine code changes, obvious fixes, standard operations20- Work in progress that will change before it matters21- Generic information already widely known22- Simple Q&A answerable from documentation2324## Workflow2526If this session already exposes the Nowledge Mem MCP server:27281. Search first with `memory_search` to avoid duplicates.292. If an existing memory captures the same concept, use `memory_update`.303. Otherwise, create it with `memory_add`.3132Otherwise:33341. Search first to avoid duplicates: `nmem --json m search "concept"`352. If an existing memory captures the same concept, update it:36 ```bash37 nmem --json m update <memory_id> -c "updated content"38 ```393. Otherwise, create a new memory:40 ```bash41 nmem --json m add "content" -t "Title" --unit-type decision -l "label" -s codex -i 0.842 ```4344At the end of a substantial task, explicitly check whether one durable memory should be added or updated. Do not skip that review just because the user did not ask.4546## Unit types4748`fact`, `preference`, `decision`, `plan`, `procedure`, `learning`, `context`, `event`4950## Additional flags5152- Temporal context: `--when past` (or `present`, `future`, `timeless`)53- Event date: `--event-start 2026-03` (for time-anchored knowledge)5455## Importance scale5657- `0.8-0.9`: Major decisions, important procedures, breakthroughs58- `0.5-0.7`: Useful patterns, conventions, secondary decisions59- `0.3-0.4`: Minor notes, preferences, contextual observations60- `1.0`: rare, only for critical corrections or pivotal choices6162## After saving6364Report what was stored, which unit type was used, and why each memory was worth keeping. One strong memory is better than three weak ones.6566## Links6768- [Documentation](https://mem.nowledge.co/docs)