rekal

Git-anchored intent ledger for AI development teams

TLDR

Initialize rekal in the current git repository

$ rekal init

Capture current AI session context after a commit

$ rekal checkpoint

Search for context related to a query

$ rekal "[search query]"

Search scoped to a specific file or directory

$ rekal --file [src/billing/] "[discount logic]"

View recent checkpoints

$ rekal log

Drill into a specific session

$ rekal query --session [session_id] --full

Push rekal data to the remote branch

$ rekal push

Sync team context from remote branches

$ rekal sync

SYNOPSIS

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 (turns, tool calls, checkpoints, files touched — distributed via a dedicated git orphan branch) and a local index.db (full-text index, vector embeddings, file co-occurrence graphs) used only for search.The tool installs git hooks so checkpoints are written automatically on every commit. Retrieval is a hybrid of BM25 lexical search, LSA, and Nomic vector embeddings, returning scored JSON with the best-matching turn index for progressive drill-down. A session of 2–10 MB of raw transcripts compresses to roughly 300 bytes on disk. Embeddings ship with the binary — no API keys, accounts, or external services.

PARAMETERS

init

Initialize rekal in the current git repository

clean

Remove rekal setup from the repository

checkpoint

Capture current AI session context at the latest commit

push [--force]

Push rekal data to the remote branch

sync [--self]

Sync team context from remote branches

index

Rebuild the local search index database

log [--limit n]

Show recent checkpoints

query --session id [--full] [--offset n] [--limit n]

Drill into a specific session with optional pagination

query "SQL" [--index]

Execute a raw SQL query against the data or index database

version

Print CLI version

--file path

Scope search to a specific file or directory

--role human

Filter search results to human turns only

CAVEATS

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.

HISTORY

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.