rekal

rekal \<command\> [options] [query]

rekal captures AI session context at every git commit, creating a permanent, append-only record of development reasoning and conversations alongside code changes. It stores two databases: a shared data.db containing sessions, turns, tool calls, checkpoints, and files touched, and a local index.db with full-text indexes, vector embeddings, and file co-occurrence graphs for search.The tool integrates with git hooks to automatically capture checkpoints when commits are made. Search uses hybrid retrieval combining FTS5 keyword search with semantic vector search. Data is shared across teams via a dedicated git branch, keeping conversation history decentralized with no external services required.

init

Initialize rekal in the current git repository

Remove rekal setup from the repository

Capture current AI session context at the latest commit

Push rekal data to the remote branch

Sync team context from remote branches

Rebuild the local search index database

Show recent checkpoints

Drill into a specific session with optional pagination

Execute a raw SQL query against the data or index database

Print CLI version

Scope search to a specific file or directory

Filter search results to human turns only

Currently works with Claude Code sessions only; support for other agents is planned. Requires Git and runs on macOS or Linux. The append-only design means records cannot be edited or deleted after capture. Data never leaves your local machine or git repository.

rekal was created by the rekal-dev team and written in Go. It was first released in 2026 as a tool for preserving the reasoning context behind AI-assisted development, addressing the problem of lost conversation history when AI coding agents are used across team workflows.

git(1), sqlite3(1)