Whiteboard, an open-source IDE where coding agents draw the diagram they are working from: Reveneau

Why it mattersA reviewer no longer has to reconstruct what the agent thought it was doing from a text trace, since the diagram, the decision log and the code all live on one canvas the agent updates.

A reviewer looking at a pull request written by a coding agent has always had to guess how the agent got there. The commit is one artefact, the chat trace is another, and the reasoning that connects them is usually gone. Whiteboard, a new open-source macOS IDE launched on Show HN today, tries to put those three back in one place.

The project is from devdotfast, a Y Combinator W26 startup, and its README calls it an open-source desktop app where humans and agents can architect software together in a common workspace. Its GitHub repository has 192 stars against a 5-week window, is MIT-licensed, and the Show HN post reached 45 points in the first hour.

The SDK the agent draws with

Whiteboard plugs into Claude Code, Codex or another coding agent from the welcome screen. The agent is then given an SDK to draw on an in-app canvas: sequence diagrams, entity-relationship diagrams, or quotes from its own trace. Clicking on any visualisation jumps to the underlying code in the editor, which the README says gets keybindings and LSP support from VS Code out of the box.

The pitch, in the README's own words, is diagrams that lead to code. Design tools that live outside the repository have never linked back cleanly, because the tradeoffs only show up after the first implementation. When the diagram lives beside the code and updates with it, the picture stays honest.

A semantic diff viewer, written in Rust

The team wrote its own diff viewer rather than reusing the standard one. It is AST-aware, meaning it groups changes by what the code is doing rather than by the line it touches. The README says large added functions are summarised as pseudocode, and unit tests and documentation changes are collapsed by default. Both defaults are customisable through a WebAssembly plugin system.

For a reviewer, that is the difference between reading fifty files and reading one summary that says here is the new function, here is the changed contract, here is the test that guards it.

The decision log

The third piece is a link between an agent's trace and the requirements it was working from. The README says the tool lets an agent query and link its own traces on the Whiteboard, so a reader can visualise the requirements that were set, understand how they were implemented, and see what decisions the agent made autonomously.

That is the piece missing from most agent-coding workflows today: a durable record of what the agent chose to do without being told.

What it does not do yet

The README is straightforward about the current limits. You cannot edit files inside Whiteboard, so it complements an editor rather than replaces one. Working across multiple repositories in a single review is not well supported. Shared reviews do not update after they are shared, so a collaborator needs to be sent the new version.

Whiteboard runs against local checkouts, and the app is MIT-licensed. The team says a hosted product for teams is planned, and everything will always remain self-hostable.

Source

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.