2.7 KiB
2.7 KiB
title, type, status, created, updated, verified_on, confidence, tags, sources
| title | type | status | created | updated | verified_on | confidence | tags | sources | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Memory Backend | system | active | 2026-07-22 | 2026-11-17 | 2026-11-17 | high |
|
|
Memory Backend
Purpose
Persist and recall agent memory across sessions.
Current Configuration
- Primary backend: Mnemosyne (local native)
- Memory enabled: true
- User profile enabled: true
- Memory char limit: 5000
- User char limit: 3000
- Flush every: 6 turns
- Auto sleep: true
- Default scope: session
- Vector type: int8
- Profile isolation: true (configured; requires gateway restart for the current messaging runtime)
- Shared surface:
data/shared/mnemosyne.db - Shared surface read: true
- Sync roles: user
- Skip contexts: cron, flush, subagent, background, skill_loop
- Reflect max calls per session: 3 (
reflect.disabled_for_cron: true) - Sleep threshold: 50 turns
- Nudge interval: 10
- Ignore patterns: none set
- Live totals (2026-11-17, sqlite): 2,901 memories — 1,130 working / 1,249 episodic / 522
memoria_facts
Storage Location
- Mnemosyne home:
~/.hermes/mnemosyne/ - DB:
~/.hermes/mnemosyne/data/ - Backups:
~/.hermes/mnemosyne/backups/ - Logs:
~/.hermes/mnemosyne/logs/ - Models:
~/.hermes/mnemosyne/models/ - Config:
~/.hermes/mnemosyne/config.yaml
Known Issues
- Health monitoring currently treats capture as stale only after 24 hours; an activity-aware comparison is preferable to catch dry-intake failures sooner without false alarms during idle periods.
- The 2026-08-01 automatic-capture failure and repair are documented in incidents/2026-08-01-mnemosyne-capture-failure.
- Vector recall degraded silently until 2026-10-03 —
fastembedwas missing from every uv-managed runtime, somnemosyne_recallfell back to FTS5 keyword scoring with no error raised. Repaired 2026-10-03; the 15-min dependency guard now assertsfastembedand the model cache path. See incidents/2026-10-03-mnemosyne-vector-recall-fts-only. - Embedding coverage re-verified 2026-11-17 —
check_embedding_coverage.pyreports 2,379 / 2,901 (82.0 %) embedded; working 1,130 / 1,130 (100 %) and episodic 1,249 / 1,249 (100 %) (the gap ismemoria_facts, unvectorized by design). Both tiers are above the 95 % reindex threshold, so no reindex is needed. Earlier (2026-10-26) check: 1,899 / 2,350 (80.8 %), working 96.8 %, episodic 99.4 %.
Correction (2026-07-22, per Anthony): this page previously referenced Hindsight as a possibly-used backend on this box. That was a mix-up with a different Hermes instance — see comparisons/memory-backends.