About WHY
WHY is a local-first desktop app for people who review code written by coding agents, across many repositories and git worktrees. It watches where your agents work and shows the live diff, and, since v0.2, what Claude Code and Codex did there.
The problem it starts from
Agent output is fast, but understanding it is slow. git diff shows what changed. It does not show the task the agent was given, why it made a change, whether a block was necessary, or whether it drifted from scope, so people paste diffs back into the agent's chat to recover that context.
WHY tries to answer three questions in one place: what changed, how it changed, and why. It is not meant to explain every line. An agent's reason can be an after-the-fact rationalisation, so "why" never replaces reading the code; the aim is to tell you where to look, and to say how well each explanation is backed.
What it is, and what it is not
WHY is not an IDE, an agent harness, an automated code reviewer or an orchestrator. It complements the tools you already use, and you own the approval. Line-level provenance, attention signals and closing the loop with the agent are planned, not shipped; the roadmap says what is next and what the limits are today.
Principles
- Local-first
- Your repositories, the agent events WHY captures and their transcripts stay on your machine. There is no telemetry and no network code.
- Harness-agnostic
- WHY is meant to work across agents, not inside one. Claude Code and Codex are covered today.
- Evidence-first
- Every explanation is labelled Observed, Declared, Inferred or Unknown, and missing evidence is shown as unknown rather than guessed.
- Passive-first
- WHY observes. It never writes to the repositories it reads, and it never takes a lock your agent needs.
- No fake reasoning
- WHY claims no access to an agent's private chain of thought. An agent's own explanation is shown as its claim, not as proof of why the code changed.
Not the first tool in this space
Each piece of WHY exists somewhere else. The bet is the combination: harness-agnostic, many repositories and worktrees in one place, and evidence-backed "why". Agent vendors can also show "why" next to the diff inside their own apps; WHY's angle is to work across agents.
- LGTM: live diff and line-anchored feedback, terminal-first.
- Glassbox: a GUI with agent-authored, line-anchored rationale.
- Docket: replays session edits for per-line and per-hunk attribution.
- Entire: git-native agent session checkpoints.
- Atomic: observed events and an inferred causal graph.
- Daintree: a multi-project, multi-worktree agent workspace with review.
- Muster: desktop worktree discovery and review.
- Prufs: decision trails and line links.
Conductor, Crystal and Vibe Kanban are multi-worktree agent GUIs with diffs that lean towards orchestration. These descriptions come from WHY's own research notes and may be out of date; check each project for yourself.
Who is behind it
WHY is maintained by one person in their spare time, as open source under MIT OR Apache-2.0. Bug reports and ideas are welcome as issues; security problems go through GitHub's private reporting, described in the security policy. The changelog lists every release, and the source is public.