ScottRBK/forgetful

What it solves

Forgetful provides a shared, persistent knowledge base for AI agents, solving the problem of context loss across different sessions, devices, or different agentic applications. It allows agents to store and retrieve information about users, projects, and decisions so they don't have to be reminded of the same details repeatedly.

How it works

It operates as a Model Context Protocol (MCP) server that agents can connect to. It uses the Zettelkasten principle, requiring memories to be atomic (one concept per note). When a memory is saved, the system generates a semantic embedding using FastEmbed and automatically links it to similar existing memories based on a similarity score, creating a knowledge graph. It supports both SQLite and PostgreSQL for storage and can be deployed via STDIO or HTTP transport.

Who it’s for

Developers and users of AI agents (such as Claude Code, Cursor, or custom bots) who need a centralized memory system to maintain consistency and context across multiple agentic workflows.

Highlights

• Automatic Knowledge Graph: Automatically links semantically similar memories to build a network of knowledge.
• Atomic Memory: Enforces a one-concept-per-note structure to ensure clarity and precise retrieval.
• Entity Tracking: Supports tracking concrete entities (people, organizations, devices) and their relationships.
• Local-First: Can run entirely locally using FastEmbed for embeddings and SQLite for storage.
• Agent Coordination: Includes features for plans, tasks, and procedural "skills" for multi-agent coordination.