Home

Engram

Persistent memory and cognitive infrastructure for AI agents.

Engram gives AI agents long-term memory, reasoning capabilities, knowledge graph traversal, and tool management.

Quick Links

Section	Description
Architecture	Components, data flow, modules
API Reference	All HTTP endpoints grouped by function
Configuration	Environment variables and defaults
CLI Tools	Command-line interface reference
Deployment	Docker, systemd, encryption setup
Integration Guides	Claude Code, MCP, hooks

Core Features

Memory System

• Episodic memory with categories (task, discovery, decision, state, issue, general, reference)
• Vector search with ONNX embeddings (BGE-M3 by default)
• Structured facts (subject-predicate-object triples)
• Scratchpad for temporary session state

Cognitive Layer (Eidolon)

• Hopfield networks for pattern completion
• Spreading activation across memory graph
• Contradiction detection
• Duplicate detection and deduplication
• Memory consolidation and reflection

Knowledge Graph

• Entity extraction and linking
• Relationship tracking
• Community detection
• PageRank scoring
• Graph visualization

Agent Infrastructure

• Skill management and evolution
• Tool quality tracking
• Gate system for dangerous command approval
• Session management with streaming

Included Components

Crate	Description
engram-server	HTTP API server (default port 4200)
engram-cli	Command-line client
engram-lib	Core library (memory ops, brain, embeddings)
engram-cred	Credential management with YubiKey support
engram-credd	Credential daemon (port 4400)
engram-mcp	Model Context Protocol server
engram-migrate	Database migration tool
engram-sidecar	Lightweight sidecar for agent integration
engram-approval-tui	Terminal UI for approval workflows
agent-forge	Task specification and verification protocol

Getting Started

See the main README for installation and basic usage.