Whiteboard ships as an open-source app for human-agent software design
Whiteboard is an open-source desktop app built for reviewing and designing software together with coding agents rather than alone. It connects to tools developers already use, including Claude Code and Codex, and gives the connected agent an SDK to draw on an in-app canvas: sequence diagrams, entity relationship diagrams, and traces of its own reasoning that link back to the underlying code. Clicking a diagram or a quoted piece of the agent's trace jumps straight to the relevant code, and the editor itself is built on VS Code, so it carries over the familiar keybindings and LSP-based navigation.
The core review workflow is: install Whiteboard, connect a coding agent from the welcome screen, then ask that agent to review the current branch against an up-to-date main and open the result inside Whiteboard. The authors say Whiteboard works best, in their own experience, with models they call GPT-6 Sol and Claude Opus 5.5, citing their balance of intelligence, cost and speed; they give no further detail on these models. Because raw diffs are noisy, the team wrote its own semantic, AST-aware diff viewer in Rust, so a reviewer sees only the changes relevant to them; large added functions are summarized as pseudocode by default, and routine changes like unit tests and documentation are collapsed. The diff system can be customized through a WASM-based plugin system.
Rather than maintaining patches on top of stock VS Code, the team vendors a fork of Code OSS outright, arguing that coding agents struggle to apply patches cleanly and that a large share of the current VS Code codebase, which they estimate at about 45%, is Copilot-related code they don't need. They say they track upstream Code OSS and merge in security and feature patches as they appear.
Whiteboard is MIT-licensed and runs against local checkouts; the vendored Code OSS fork keeps Microsoft's original MIT license and third-party notices. The authors say a hosted product for teams is planned, but that everything will always remain self-hostable. Anonymous telemetry is collected, but the authors state it does not include a user's code, diffs, Whiteboard text, prompts, or model output, and it can be turned off.
The post also lists known limitations: users cannot currently edit files inside Whiteboard, working across multiple repositories in a single review isn't well supported, and while reviews can be shared between machines via a share button, updates made after a review has been shared do not propagate automatically, requiring the review to be re-shared.
Key facts
- Whiteboard is an open-source, MIT-licensed desktop app that lets developers and coding agents like Claude Code and Codex collaborate on a shared visual canvas for software design and review.
- The app gives connected agents an SDK to draw diagrams and link their own reasoning traces on the canvas, with click-through navigation from diagrams straight to the underlying code.
- A custom semantic, AST-aware diff viewer, written in Rust, filters raw code diffs down to relevant changes and can summarize large functions or collapse routine test and documentation changes.
- Rather than patch stock VS Code, the team vendors its own fork of Code OSS, saying coding agents handle patches poorly and that roughly 45% of today's VS Code codebase is Copilot-related.
- Current limitations include no in-app file editing, weak support for reviewing across multiple repositories at once, and shared reviews that don't auto-update after later changes.
Why it matters
As more developers hand coding work to agents like Claude Code and Codex, reviewing and understanding what those agents changed becomes the bottleneck rather than writing the code itself. Whiteboard is aimed squarely at that gap: instead of reading raw diffs, a reviewer gets diagrams, traces of the agent's own decisions, and a filtered view of only the code changes that matter.
Who it affects
Developers and teams who already use coding agents such as Claude Code or Codex for implementation and want a dedicated interface for reviewing and discussing that agent-generated work, rather than reusing a general-purpose text editor for line-by-line diffs.
How to use it
Whiteboard is downloadable for macOS and Fedora and is MIT-licensed and free, running against local repository checkouts. After installing, a user connects an existing coding agent from the welcome screen, then asks that agent to review the current branch against main and open the result in Whiteboard. A hosted, team-oriented product is planned for the future, with the authors saying the tool will always remain self-hostable.
How solid is it
The account comes directly from the project's own Show HN post and README, describing what the team built rather than independent testing. The authors' claim that Whiteboard performs best with models they name GPT-6 Sol and Claude Opus 5.5 is stated as their own experience, with no benchmark or further detail given, and the roughly 45% Copilot-code figure for VS Code is likewise the authors' own estimate rather than a cited measurement.
Risks and caveats
The authors themselves list several gaps: files cannot currently be edited inside Whiteboard, reviewing across multiple repositories at once isn't well supported, and a review shared between machines does not reflect later updates until it is re-shared. As a fresh open-source project built around specific coding agents, its workflow and defaults are still described as being actively tuned.
“Whiteboard is an open-source desktop app where humans and agents can architect software together in a common workspace.”
— Whiteboard project description