Knowl

•
112
•

Provides persistent, self-retiring memory for AI agents like Claude Code, Cursor, and Codex, leveraging local SQLite for fact management without requiring API keys.

Knowl addresses the common challenge of AI agent memory becoming bloated and inaccurate over time by implementing a unique "supersession" mechanism. Instead of simply appending new information, Knowl intelligently retires outdated facts when they are replaced, preventing agents like Claude Code, Cursor, and Codex from retrieving conflicting information. This local-first, SQLite-based persistent memory system ensures agents always access the most current and relevant data, significantly improving retrieval accuracy without the need for external API keys, as validated by strong performance in leading AI memory benchmarks.

主な機能

01Integration with Model Context Protocol (MCP) for agent connectivity
02Self-retiring "supersession" mechanism for outdated facts
03Persistent local memory for various AI agents (Claude Code, Cursor, Codex, etc.)
04100% local operation with SQLite database and no API keys required
05Demonstrated high accuracy in AI memory benchmarks for conflict resolution
0627 GitHub stars

ユースケース

01Providing consistent, shared memory across multiple AI agent sessions or parallel worktrees
02Preventing AI agents from retrieving conflicting or outdated information during development
03Improving the reliability and accuracy of AI agent interactions in coding and development workflows