Whiteboard
whiteboard is a desktop application built for reviewing and discussing software design and architecture, particularly work produced by AI coding agents. As agents take on larger changes, reviewing them as a wall of raw code diff makes it hard to see intent: what problem a change solves, what alternatives were considered, and how pieces of a system relate to each other. whiteboard addresses this by giving both humans and agents a shared visual canvas.
Coding agents such as Claude Code and Codex can connect to whiteboard through an SDK and render diagrams directly onto the canvas as they work, including sequence diagrams and entity-relationship diagrams that document the structure of what they built. Rather than a static diff, whiteboard includes a semantic diff viewer built on an AST-aware engine written in Rust: large functions are summarized as pseudocode instead of shown line by line, and low-signal changes like test or documentation edits are collapsed by default. A WASM plugin system lets teams customize how diffs are rendered for their own codebase conventions.
The tool also logs the decisions an agent makes during a task, so a reviewer can trace a piece of implementation back to the requirement or reasoning that produced it, and click through from a diagram or decision straight to the underlying code, with VSCode keybinding and language-server support built in. It runs locally on the user's machine rather than as a hosted service. Known gaps include no in-app file editing, weak support for browsing multiple repositories in one review session, and shared reviews that do not auto-update for collaborators without re-sharing.
whiteboard is placed in the Tools quadrant. It has not been used first-person in this wiki author's own workflow, so despite its comparatively strong external signals, MIT license, over 2,000 GitHub stars, and nearly 500 commits, it is placed in Assess rather than Trial. Those signals describe a technology worth exploring, not confirmed production use, which is the bar this wiki sets for Trial or Adopt. The semantic diff and decision-trace approach is a notable answer to the growing problem of reviewing agent-authored code, and is worth a hands-on trial once the noted gaps around file editing and multi-repo review are less likely to be blocking.
