V3.3 Overview

V3.3 Overview: The Living Brain Evolves

SuperLocalMemory v3.3 (codename "The Living Brain Evolves") gives your memory a lifecycle. Memories strengthen when used, fade when neglected, compress when idle, and consolidate into reusable patterns --- automatically, locally.

V3.3 builds on the associative memory system introduced in V3.2 with six new capabilities that make SLM self-maintaining. No manual cleanup. No storage bloat. No zombie processes.

What's New

Feature	What It Does	User Benefit
Adaptive Memory Lifecycle	Memories naturally strengthen with use and fade when neglected	No manual cleanup; configurable retention curves
Smart Compression	Embedding precision adapts to memory importance	Up to 32x storage savings for low-priority memories
Cognitive Consolidation	Extracts patterns from clusters of related memories	One decision referenced 50 times becomes one reusable insight
Pattern Learning (Soft Prompts)	Auto-learned preferences injected into agent context at session start	The system teaches itself what matters to you
Hopfield Retrieval (6th Channel)	Vague or partial queries complete themselves	Ask half a question, get the whole answer
Process Health	Orphaned SLM processes detected and cleaned automatically	No more zombie workers eating RAM

Performance at a Glance

Metric	V3.2	V3.3	Change
RAM (Mode A/B)	~4 GB	~40 MB	100x reduction
Retrieval channels	5	6	+1 (Hopfield)
MCP tools	29	35	+6
CLI commands	21	26	+5
Dashboard tabs	20	23	+3
API endpoints	9	16	+7
Tests passing	---	2,348	Full coverage

Feature Details

Adaptive Memory Lifecycle

Memories now have a natural lifecycle. Frequently accessed memories grow stronger over time. Neglected memories gradually fade. This happens automatically based on your actual usage patterns.

• Configurable retention curves --- set how aggressively memories decay per project or globally
• Access-weighted scoring --- memories you rely on are never lost
• Review before removal --- faded memories are flagged, not silently deleted
• Full audit trail --- see why any memory was retained or marked for decay

Smart Compression

Not all memories deserve the same storage budget. V3.3 automatically adapts embedding precision based on memory importance.

• High-value memories stay at full resolution --- zero quality loss
• Low-priority memories compress up to 32x --- significant storage savings at scale
• Automatic tiering --- compression level adjusts as importance changes
• Lossless for active memories --- only idle, low-access memories are compressed

Cognitive Consolidation

When you reference the same decision, pattern, or preference across many memories, V3.3 detects the cluster and extracts a single reusable insight.

• Automatic pattern detection across related memories
• One insight instead of fifty references --- cleaner retrieval, less noise
• Source tracing --- every consolidated insight links back to its source memories
• Incremental --- new memories merge into existing patterns naturally

Pattern Learning (Soft Prompts)

V3.3 learns your preferences over time and injects them into your agent's context at session start. No manual configuration needed.

• Auto-learned from usage --- the system observes what you care about
• Session-start injection --- your agent already knows your style before you type
• Fully transparent --- view and edit all learned patterns via CLI or dashboard
• Per-project scoping --- different projects can have different learned preferences

Hopfield Retrieval (6th Channel)

A new retrieval channel that handles vague, partial, or incomplete queries. When you can only remember fragments, this channel fills in the gaps.

• Partial query completion --- half a question returns the whole answer
• Noise-tolerant --- typos and fuzzy phrasing still find the right memory
• Complements existing channels --- works alongside keyword, semantic, temporal, graph, and active retrieval
• Automatic --- no special syntax needed; the channel activates when other channels return low-confidence results

Process Health

SLM background workers sometimes outlive their parent sessions. V3.3 detects and cleans orphaned processes automatically.

• Automatic detection of zombie SLM processes
• Safe cleanup --- only orphaned processes are targeted
• RAM recovery --- reclaims memory from dead workers
• On-demand or scheduled --- run manually or let it happen automatically

New CLI Commands

Command	Description
slm decay	Run memory lifecycle review; show faded and strengthened memories
slm quantize	Run smart compression cycle on eligible memories
slm consolidate --cognitive	Extract patterns from clusters of related memories
slm soft-prompts	View auto-learned patterns and preferences
slm reap	Clean orphaned SLM processes

All existing CLI commands continue to work unchanged. See CLI Reference for the full list.

New MCP Tools

Tool	Description
forget	Mark a memory for decay or immediate removal
quantize	Trigger smart compression on specific memories
consolidate_cognitive	Run cognitive consolidation on a memory cluster
get_soft_prompts	Retrieve current auto-learned patterns
reap_processes	Clean orphaned background processes
get_retention_stats	View memory lifecycle statistics

All 29 existing MCP tools retain their signatures. See MCP Tools for the full list.

New Dashboard Tabs

Tab	What It Shows
Memory Lifecycle	Retention curves, decay candidates, access frequency heatmaps
Compression	Storage savings, compression ratios per memory tier
Patterns	Auto-learned preferences, consolidated insights, source tracing

7 new API endpoints power these tabs. All 9 existing endpoints are unchanged.

Configuration

All V3.3 features default to OFF. Zero breaking changes. Enable them as you evaluate each capability:

slm config set decay.enabled true
slm config set quantize.enabled true
slm config set cognitive_consolidation.enabled true
slm config set soft_prompts.enabled true
slm config set hopfield_retrieval.enabled true
slm config set process_health.auto_reap true

Or enable them one at a time at your own pace.

Backward Compatibility

V3.3 is a strict superset of V3.2. No migration needed.

• All 29 existing MCP tools retain their signatures
• All 21 existing CLI commands work identically
• All existing API endpoints are unchanged
• Existing retrieval behavior is preserved when new features are disabled
• New database tables are created additively (no schema alterations to existing tables)
• Rollback SQL is provided for every new table

Feature Matrix by Mode

Every V3.3 feature respects the existing three-mode architecture. Mode A remains fully local.

Feature	Mode A (Local Guardian)	Mode B (Smart Local)	Mode C (Full Power)
Memory lifecycle	Full (pure scoring math)	Full	Full
Smart compression	Full (local computation)	Full	Full
Cognitive consolidation	Rules-based extraction	Ollama-assisted	Cloud LLM-assisted
Soft prompts	Pattern matching	Ollama refinement	Cloud LLM refinement
Hopfield retrieval	Enabled with local embeddings	Full	Full
Process health	Full (OS-level detection)	Full	Full

Related Pages

• V3.2 Overview --- Associative memory, auto-invoke, temporal intelligence
• CLI Reference --- Full command list including V3.3 additions
• MCP Tools --- Full tool list including V3.3 additions
• Modes Explained --- Mode A / B / C architecture
• Architecture Overview --- System design

---

Part of Qualixar | Created by Varun Pratap Bhardwaj