Provides a local-first and token-efficient memory server for AI coding assistants, enabling durable knowledge storage.
Sponsored
Memory Lite serves as a dedicated, local-first memory server for AI coding assistants that speak the Model Context Protocol (MCP). It bridges the gap between an AI client's short-term conversation memory and the need for persistent knowledge, storing crucial information like architectural decisions, project conventions, commands, and task states. The system organizes this knowledge in a hierarchical tree (global, project, task, atomic) with optional graph-lite edges for relationships, and employs a token-efficient, tiered retrieval strategy to ensure agents access relevant information without unnecessary cost or context stuffing.
Key Features
01Flexible data storage options including SQLite files, libSQL, and Turso
02Token-efficient retrieval strategy (summaries, light search, full detail)
03Hierarchical memory organization with optional graph-lite relationships
04Local-first durable knowledge storage on disk
050 GitHub stars
06Seamless integration with MCP-compatible AI coding clients (Windsurf, Cursor, Claude Desktop)
Use Cases
01Augmenting AI coding assistants with long-term project context
02Documenting architectural decisions, project conventions, and common gotchas
03Persisting task-specific information to resume work later across sessions
