SuperLocalMemory

April 13, 2026 · View on GitHub

Qualixar OS includes SLM-Lite, a lightweight memory system built into the orchestrator. SLM-Lite is inspired by and compatible with SuperLocalMemory (SLM) -- a standalone, full-featured agent memory system available on npm and PyPI.

If you need more advanced memory capabilities, SuperLocalMemory is the upgrade path.

What is SuperLocalMemory?

SuperLocalMemory is an infinite-memory system for AI agents. It runs entirely on your machine -- no cloud, no API keys, no data leaving your laptop. It uses techniques from differential geometry, algebraic topology, and stochastic analysis to replace the LLM calls that other memory systems need for core operations like similarity scoring, contradiction detection, and lifecycle management.

Backed by 3 peer-reviewed research papers:

arXiv:2603.02240 -- Core architecture
arXiv:2603.14588 -- Mathematical retrieval engine
arXiv:2604.04514 -- The Living Brain (lifecycle + consolidation)

Performance: On the LoCoMo benchmark (standard long-conversation memory evaluation), SLM Mode A scores 74.8% with zero cloud dependency -- outperforming Mem0 (64.2%) by 16 percentage points. Mode C reaches 87.7%.

Works with: Claude Code, Cursor, Windsurf, VS Code Copilot, ChatGPT Desktop, Gemini CLI, JetBrains, Zed, and 17+ AI tools via MCP.

SLM-Lite vs SuperLocalMemory

Feature	SLM-Lite (built into QOS)	SuperLocalMemory (standalone)
4-layer memory (working/episodic/semantic/procedural)	Yes	Yes
Full-text search (FTS5)	Yes	Yes
Trust scoring	Linear decay formula	Mathematical: Fisher-Rao + topology
Auto-invoke (proactive recall)	Bandit-tuned, LLM-assisted	6-channel retrieval including Hopfield completion
Behavioral capture	Yes	Yes + pattern learning + soft prompts
Belief graph	Causal edges, 2-hop expansion	Full knowledge graph with contradiction detection
Retrieval channels	3 (FTS5, LIKE, working scan)	6 (FTS5, semantic, temporal, spreading activation, cross-encoder, Hopfield)
Mathematical engine	None (LLM-based scoring)	Fisher-Rao, PolarQuant, algebraic topology
Adaptive lifecycle	Manual cleanup	Auto-decay, smart compression, consolidation
Embedding models	None (text search only)	Local CPU embeddings (all-MiniLM-L6-v2)
Cross-encoder reranking	No	Yes (ONNX, local)
Dashboard	Via QOS dashboard	Standalone web dashboard (23 tabs)
MCP tools	1 (`search_memory`)	35
CLI commands	Via QOS CLI	26 standalone commands
Modes	Single mode	Mode A (zero cloud), Mode B (local LLM), Mode C (cloud-enhanced)
EU AI Act compliance	Inherits from QOS	Built-in compliance features
License	FSL-1.1-ALv2 (with QOS)	AGPL v3

When to Upgrade

SLM-Lite is sufficient when:

You primarily use QOS for task orchestration and need basic memory
Your memory corpus is small (under a few thousand entries)
Text-based search (FTS5) meets your retrieval needs

Upgrade to full SuperLocalMemory when:

You need semantic retrieval (embedding-based similarity, not just keyword matching)
You have a large memory corpus that benefits from mathematical scoring
You want adaptive lifecycle (memories that strengthen with use and fade when neglected)
You need 6-channel retrieval including Hopfield completion for vague queries
You want the standalone dashboard with 23 tabs for memory visualization
You need the full CLI for memory management (slm remember, slm recall, slm decay, etc.)

Install SuperLocalMemory

npm (recommended)

npm install -g superlocalmemory
slm setup     # Choose mode (A/B/C)
slm doctor    # Verify installation

pip

pip install superlocalmemory

MCP Integration

Add to your MCP client configuration:

{
  "mcpServers": {
    "superlocalmemory": {
      "command": "slm",
      "args": ["mcp"]
    }
  }
}

This gives you 35 MCP tools for memory operations, far beyond the single search_memory tool in SLM-Lite.

Using Both Together

SLM-Lite and full SuperLocalMemory can coexist. QOS uses SLM-Lite for its internal orchestrator memory (task context, behavioral capture, belief graph). SuperLocalMemory runs as a separate MCP server for your broader agent memory needs.

They serve different scopes:

SLM-Lite -- QOS-internal memory (task outcomes, agent behaviors, orchestration patterns)
SuperLocalMemory -- Your complete agent memory (project knowledge, personal facts, cross-session context)