aa6eabb816
Some checks failed
Forge CI / smoke-and-build (pull_request) Failing after 23s
The holographic retriever had temporal decay implemented but disabled (half_life=0). All facts scored equally regardless of age — a 2-year-old fact about a deprecated tool scored the same as yesterday's deployment config. This commit: 1. Changes default temporal_decay_half_life from 0 to 60 days - 60 days: facts lose half their relevance every 2 months - Configurable via config.yaml: plugins.hermes-memory-store.temporal_decay_half_life - Added to config schema so `hermes memory setup` exposes it 2. Adds access-recency boost to search scoring - Facts accessed within 1 half-life get up to 1.5x boost on their decay factor - Boost tapers linearly from 1.5 (just accessed) to 1.0 (1 half-life ago) - Capped at 1.0 effective score (boost can't exceed fresh-fact score) - Prevents actively-used facts from decaying prematurely 3. Scoring pipeline: score = relevance * trust * decay * min(1.0, access_boost) - Fresh facts: decay=1.0, boost≈1.5 → score unchanged - 60-day-old, recently accessed: decay=0.5, boost≈1.25 → score=0.625 - 60-day-old, not accessed: decay=0.5, boost=1.0 → score=0.5 - 120-day-old, not accessed: decay=0.25, boost=1.0 → score=0.25 23 tests covering: - Temporal decay formula (fresh, 1HL, 2HL, 3HL, disabled, None, invalid, future) - Access recency boost (just accessed, halfway, at HL, beyond HL, disabled, range) - Integration (recently-accessed old fact > equally-old unaccessed fact) - Default config verification (half_life=60, not 0) Fixes #241
Holographic Memory Provider
Local SQLite fact store with FTS5 search, trust scoring, entity resolution, and HRR-based compositional retrieval.
Requirements
None — uses SQLite (always available). NumPy optional for HRR algebra.
Setup
hermes memory setup # select "holographic"
Or manually:
hermes config set memory.provider holographic
Config
Config in config.yaml under plugins.hermes-memory-store:
| Key | Default | Description |
|---|---|---|
db_path |
$HERMES_HOME/memory_store.db |
SQLite database path |
auto_extract |
false |
Auto-extract facts at session end |
default_trust |
0.5 |
Default trust score for new facts |
hrr_dim |
1024 |
HRR vector dimensions |
Tools
| Tool | Description |
|---|---|
fact_store |
9 actions: add, search, probe, related, reason, contradict, update, remove, list |
fact_feedback |
Rate facts as helpful/unhelpful (trains trust scores) |