A local ledger for AI coding runs.
ait records every Claude Code, Codex, Aider, Gemini, or Cursor run as a repo-local attempt so another agent can review, continue, or recover it later.
Package: ait-vcs. Command: ait. Requires Python 3.11+.
Agents can write code. The workflow around them is still fragile.
The next agent starts from zero. The implementer reviews its own work. Failed runs leave clutter. Useful decisions disappear into chat history.
ait turns each run into an attempt: a proposal with prompt, diff, review state, memory, and an explicit apply step.
Try it in 30 seconds.
No API keys. No existing repo. No real agent required.
pipx install ait-vcs
ait demo
# clean up demo repos later
ait demo --cleanSame agents. A clearer apply gate.
Until you run ait apply, a wrapped agent's work is a proposal.
claude ..., codex ..., or aider ...
Prompt, diff, trace, files, commits, review state.
Optional separate reviewer agent for high-risk changes.
Promote the attempt only when you decide it is ready.
Cross-agent handoff
The next wrapped agent can receive prior attempts, accepted facts, notes, and live repo memory files.
Adversarial review
One agent implements. Another agent challenges the attempt before apply.
Memory recall
Search old prompts, findings, and decisions when context matters again.
ait is not another coding agent.
It is the local workflow layer around the agents you already use.
ait adds a CLI-first attempt ledger across tools.
ait adds isolation, review records, memory, and explicit apply.
ait adds an attempt boundary and optional separate reviewer.
ait adds prompts, traces, reviews, handoff, and recover/apply commands.
Do not use ait when you need a hosted product.
ait is CLI-only, alpha, single-machine, and local-first. It is not an IDE plugin, autocomplete engine, hosted dashboard, team sync service, or proof that an AI reviewer will find every defect.