Technical Overview

System Capabilities

From local vectors to hierarchical knowledge graphs. A technical overview of the SuperLocalMemory research implementation.

6-Channel Hybrid Retrieval

Six parallel retrieval channels including query completion — partial queries infer your full intent. Each channel captures what others miss, fused for maximum recall.

View Docs

Mathematical Foundations

Fisher-Rao information geometry for similarity, sheaf cohomology for contradiction detection, Riemannian Langevin dynamics for self-organizing lifecycle. Three techniques never before applied to agent memory.

View Docs

EU AI Act Compliance

Mode A operates with zero cloud calls — data sovereignty by architecture, not policy. GDPR Article 15/17 support. Tamper-proof SHA-256 audit chain. Three privacy modes for different regulatory requirements.

View Docs

Bayesian Trust Scoring

Beta distribution trust per agent and per fact. Trust gates block low-trust agents from writing or deleting. Burst detection for anomaly alerting. Provenance tracking for full data lineage.

View Docs

17+ IDE Integrations

Claude Code, Cursor, VS Code Copilot, Windsurf, ChatGPT Desktop, Gemini CLI, JetBrains, Zed, Continue, Cody — all via MCP. 35 tools and 7 resources. One install, works everywhere.

View Docs

23-Tab Web Dashboard

Real-time visibility: knowledge graph, recall lab with per-channel scores, trust visualization, math health status, compliance audit, learning progress, process health, IDE connections, and settings — all local.

View Docs

11-Layer Architecture

A complete vertical stack from raw storage through to real-time visualization and adaptive learning.

Explore Full Architecture

architecture-stack.sh

6-Channel Retrieval: including query completion and spreading activation

3 Math Layers: information geometry, topology, stochastic dynamics

Adaptive Lifecycle: memories strengthen with use, fade with neglect

Smart Compression: up to 32x savings, precision adapts to importance

Cognitive Consolidation: auto-extracts patterns from related memories

Pattern Learning: soft prompts injected into agent context

Trust & Compliance: Bayesian scoring, ABAC, audit chain, EU AI Act

Web Dashboard: 23-tab real-time visualization with process health

MCP Server: 35 tools, 7 resources, 17+ IDE configs

CLI: 26 commands with structured JSON output

Storage: SQLite schema with WAL mode and profile isolation

Layer 1: Cross-Project

Transferable Tech Preferences

Layer 2: Context

Project Signal Detection

Layer 3: Workflow

Workflow Sequence Mining

Adaptive Learning

Beyond static storage: SuperLocalMemory observes which memories are actually used and re-ranks future results based on observed workflow patterns. All learning is performed locally.

3-Layer Local ML Model
ML-Powered Adaptive Ranking
Zero Telemetry (All Training is Local)

COMMON QUESTIONS

Frequently Asked Questions

What is a knowledge graph in SuperLocalMemory?

The knowledge graph automatically extracts entities and relationships from stored memories, connecting related concepts. This enables graph-based search that finds relevant memories even when keywords do not match directly.

What is memory lifecycle management?

Memory lifecycle management automatically transitions memories through states (Active, Warm, Cold, Archived) based on usage patterns. This keeps the system responsive by prioritizing recent, relevant memories while preserving older ones for long-term recall.

What is behavioral learning?

Behavioral learning tracks what happens after a memory is recalled. If a recalled memory leads to a successful action, similar memories receive higher relevance in future searches. This operates entirely locally with no LLM inference required.

Explore Further

Read the documentation or explore our published research papers.

Documentation Research Papers