/brain:remember

Search and recall relevant memories using a multi-factor scoring formula with spreading activation and context matching.

Usage

/brain:remember [query]

Arguments:

• [query] — The topic, question, or keywords to search for. The agent interprets natural language queries.

What It Does

When you run /brain:remember, the agent:

1. Searches the index — Scans ~/.brain/index.json for memories matching the query
2. Computes relevance — Scores how well each memory matches the query semantically
3. Applies decay — Calculates the current effective strength of each candidate based on time since last access
4. Runs spreading activation — Loads ~/.brain/associations.json and spreads activation from matching memories to their neighbors
5. Matches context — Compares each memory's encoding context to the current session context
6. Ranks candidates — Applies the scoring formula to rank all candidates
7. Decides response strategy — Returns results based on the quality and distribution of matches
8. Applies reinforcement — Boosts the strength and decay resistance of recalled memories via spaced reinforcement
9. Strengthens links — Applies Hebbian learning to strengthen association edges between co-recalled memories

Scoring Formula

Each candidate memory is scored using six factors:

score = 0.38 * relevance
      + 0.18 * decayed_strength
      + 0.08 * recency_bonus
      + 0.14 * spreading_bonus
      + 0.14 * context_match
      + 0.08 * salience

See Scoring Formula for a detailed breakdown.

Response Strategies

The agent adapts its response based on the quality of matches:

Example Interactions

Strong match

User: /brain:remember kafka architecture decision

Agent: Found a strong match:

📝 Chose Event-Driven Architecture for Project Alpha
   Strength: 0.82 | Recalled 3 times | Last: 2 days ago

We decided to use event-driven architecture with Kafka instead of
synchronous REST calls for the notification system. The traffic
analysis showed 10x burst patterns that would overwhelm synchronous
endpoints. Kafka was chosen over RabbitMQ for its replay capability.

Related memories also activated:
  • Burst traffic patterns require async processing (strength 0.75)
  • Kafka schema registry prevents breaking changes (strength 0.68)

Multiple related memories

User: /brain:remember deployment best practices

Agent: Found 3 related memories — synthesizing:

Based on past experiences and learnings:

1. Always verify environment variables before deploying (learned from
   the March 5 incident where a missing DB_URL caused downtime)
2. Run the full test suite against staging before promoting to
   production (decision from Sprint 12 retro)
3. Use blue-green deployment with automated rollback for critical
   services (architecture decision from Q1 planning)

Sources: deploy-incident-mar5.md, staging-gate-decision.md,
blue-green-strategy.md

No matches

User: /brain:remember kubernetes pod autoscaling

Agent: No active memories found for "kubernetes pod autoscaling".

Searched the archive — no matches there either.

You might want to:
  • Memorize this topic after exploring it
  • Check related topics: I found memories about "scaling" and
    "infrastructure" that might be tangentially relevant

Side Effects

Each successful recall has side effects that improve the memory system:

Spaced Reinforcement

The recalled memory's strength is boosted based on the time since last recall:

boost = 0.05 * spacingMultiplier * diminishingFactor

Memories recalled after longer gaps get bigger boosts. See Spaced Reinforcement.

Decay Rate Improvement

Each recall also improves the memory's decay rate:

new_decay_rate = decay_rate + 0.10 * (0.999 - decay_rate)

Hebbian Link Strengthening

When multiple memories are recalled together, their association edges are strengthened:

new_weight = min(1.0, weight + 0.10 * (1.0 - weight))