Overview

Knowl gives coding agents durable engineering context across sessions. It stores decisions, architecture, goals, constraints, facts, current state, and reusable skills as structured knowledge atoms in a repository-local SQLite database under .knowl/.

Agents access the same governed memory through the knowl CLI or the Model Context Protocol (MCP). Items carry status, freshness, confidence, tags, history, and optional evidence pointing to files, commits, tests, commands, URLs, users, agents, or indexed symbols. Core storage and retrieval do not require an external AI provider.

Knowl can also link related repositories into a workspace. Each repository keeps its own database and ownership boundary; only explicitly promoted knowledge is visible to peers.

A normal workflow queries memory first, verifies misses or stale results against repository evidence, records durable corrections, and resumes from lifecycle context or a task checkpoint.