Why ait¶
AI coding agents are useful, but their work is easy to lose track of.
ait exists for engineers who already use Claude Code, Codex, Aider, Gemini
CLI, Cursor CLI, or similar tools and want every run to leave a local,
reviewable record in the repo.
The short version¶
Without ait, an agent run is usually just a chat transcript plus whatever
changed in your working tree.
With ait, an agent run becomes an attempt:
- what you asked
- what the agent changed
- what another agent found during review
- what decision should be remembered later
- whether you explicitly applied it
The attempt lives under .ait/ in the repo. It is local. No telemetry. No SaaS.
Problem 1: the next agent starts from zero¶
Yesterday Claude investigated a bug and found a constraint. Today Codex opens the same repo and repeats the investigation because the useful context lived in a closed chat.
With ait, the next wrapped agent can receive a compact handoff from prior
attempts, accepted facts, notes, and live repo memory files such as
CLAUDE.md, AGENTS.md, .claude/memory.md, .codex/memory.md, and
.cursor/rules.
Try the demo:
See also: Pain-point demos.
Problem 2: the implementer reviews its own work¶
An agent finishes and says the tests pass. That is useful, but it is not the same as an independent review. The same model, in the same context, often misses the same blind spots.
With ait, you can ask a different agent to review the attempt before apply:
ait review attempt latest-reviewable --mode adversarial --review-adapter claude-code
ait review finding list --severity high
This is not proof of correctness. It is a separate review pass with recorded findings.
Problem 3: failed runs leave debris¶
An agent is interrupted, times out, or makes a risky edit. Without isolation, you may be left with a confusing working tree.
With ait, the wrapped run lands in an isolated workspace. Your root checkout
should not change until you decide to apply:
If the result needs inspection:
Problem 4: useful decisions disappear¶
You decided the retry budget should be three. Three weeks later, a new agent proposes five because it never saw the earlier reasoning.
With ait, you can search prior attempts and repo memory:
You still decide which recalled context matters. ait gives you the record; it
does not outsource judgment.
What ait is not¶
ait is intentionally narrow.
It is not:
- an IDE plugin
- an autocomplete engine
- a hosted dashboard
- a cross-machine sync service
- a replacement for Claude Code, Codex, Aider, Cursor, Cline, or Git
- proof that an AI reviewer will find all defects
When it is worth using¶
Use ait when the cost of losing agent context is higher than the cost of one
extra workflow layer.
It is most useful when you:
- use more than one agent CLI
- want explicit apply/recover instead of direct working-tree mutation
- want a separate reviewer agent for risky changes
- need old prompts, diffs, findings, and decisions to be searchable
- prefer local-first metadata over SaaS provenance tools
Ready to try it: Getting started.