SuperLocalMemory

LIVE

Local-First AI Agent Memory

SuperLocalMemory gives AI agents persistent, local-first memory with mathematical rigor. Using Fisher-Rao information geometry for memory importance scoring and 5-channel retrieval (temporal, semantic, episodic, graph, embedding), SLM ensures agents remember what matters across sessions — without sending data to any cloud.

$npm install superlocalmemory

Read the Paper →GitHub →Website →

5,117+

Monthly Downloads

Published Papers

17+

IDE Integrations

Retrieval Channels

Features

Fisher-Rao Geometry

Information-geometric importance scoring. Memories weighted by their statistical significance on the information manifold.

5-Channel Retrieval

Temporal, semantic, episodic, graph, and embedding retrieval channels. Each optimized for different query patterns.

Zero Cloud Dependency

All data stays local. SQLite + local embeddings. No API calls, no cloud storage, no privacy concerns.

EU AI Act Compliant

Full data sovereignty. Audit trail. Explainable memory decisions. Designed for regulatory compliance.

17+ IDE Integrations

Works with Claude Code, Cursor, VS Code, Windsurf, Cline, and any MCP-compatible tool.

Cognitive Consolidation

Automatic memory lifecycle: observe, consolidate, forget. Sheaf cohomology for contradiction detection.

Use Cases

Persistent memory for Claude Code sessions

Cross-session context for any AI agent

Privacy-preserving enterprise agent memory

Research experiment tracking and recall

Research Paper

SuperLocalMemory V3.3: The Living Brain — Cognitive Memory for AI Agents

Varun Pratap Bhardwaj, 2026

Read on arXiv→

Licensed under Elastic License 2.0