Active Memory

Active Memory (V3.1)

SuperLocalMemory V3.1 transforms the memory system from a passive database into an active intelligence layer that learns from your usage patterns and improves over time — at zero token cost.

How It Works

Every time you recall a memory, the system collects learning signals:

Recall → Feedback Signal → Co-Retrieval Graph → Confidence Boost → Adaptive Ranking

No LLM tokens are spent. All learning happens through mathematical signals computed locally.

Three Learning Phases

Phase	Signals Needed	What Changes
1. Baseline	0-19	Standard cross-encoder ranking
2. Rule-Based	20+	Heuristic boosts: recency, access frequency, trust score
3. ML Model	200+	LightGBM model trained on YOUR specific usage patterns

Phase transitions are automatic. Each recall generates ~5 signals (one per returned fact). Typical users reach Phase 2 in a day of normal work and Phase 3 within a week.

Four Learning Signals

1. Co-Retrieval

When memories are retrieved together repeatedly, they form implicit connections. The system learns that these memories are related — even if they don't share keywords.

2. Confidence Lifecycle

• Accessed facts get boosted (+0.02 per recall, capped at 1.0)
• Unused facts decay (-0.001/day after 7 days of no access, floor at 0.1)
• This creates a natural "memory importance" ranking without manual curation.

3. Channel Performance

SLM uses 4 retrieval channels (semantic, BM25, entity graph, temporal). The system tracks which channel produces the best results for different query types and adjusts channel weights accordingly.

4. Entropy Gap

When new content arrives, the system measures how "surprising" it is relative to existing memories. High-entropy content (genuinely new information) gets prioritized for deeper indexing.

Auto-Capture

SLM can automatically detect and store decisions, bug fixes, and preferences from your conversations:

slm observe "We decided to use PostgreSQL because of JSONB support"
# Auto-captured: decision (confidence: 0.75)

Detection patterns:

• Decisions: "decided", "chose", "switched to", "using X because"
• Bug fixes: "fixed", "root cause was", "resolved by"
• Preferences: "always use", "prefer", "convention is"

Auto-Recall (Session Context)

At the start of every session, the system can automatically inject relevant context:

slm session-context  # Returns top-10 relevant memories for current project

Claude Code Hooks (Invisible Integration)

slm hooks install  # One-time setup

This installs a Claude Code hook that auto-injects memory context at the start of every session. The developer never types a command — context appears automatically.

MCP Tools

Three new MCP tools for AI assistants:

Tool	Purpose	When to Call
session_init	Get project context	Once at session start
observe	Auto-capture content	After decisions, bug fixes, preferences
report_feedback	Explicit learning signal	When a recalled memory was useful/not useful

Sleep-Time Consolidation

A background maintenance process runs periodically:

• Decays confidence on unused memories
• Deduplicates near-identical facts
• Generates behavioral patterns from accumulated data
• Auto-retrains the ML ranker when enough signals accumulate

Trigger manually: slm consolidate or via the dashboard's Learning tab.

Dashboard

The Learning tab shows:

• Signal count and phase progression (0 → 20 → 200)
• Tech preferences learned from your memories
• Temporal patterns (when you work on what)
• Channel performance (which retrieval channel works best)

The Behavioral tab shows:

• Learned patterns with confidence scores
• Outcome tracking (success/failure/partial)
• Cross-project pattern transfer

Competitive Advantage

System	Learning Cost	Learning Method
Mem0	LLM call per operation	Cloud extraction
Zep	LLM call per operation	Temporal KG
Letta	LLM call per operation	Agent self-writing
SLM V3.1	$0 (zero tokens)	Mathematical signals

SLM is the only memory system that learns without spending tokens.

---

Part of Qualixar | Author: Varun Pratap Bhardwaj