Whiteboard

Whiteboard is an open-source, MIT-licensed desktop application built by dev.fast, a YC-backed team, that gives developers and AI coding agents a shared workspace for reviewing and understanding code changes. It runs locally against your own checkouts on macOS and Linux, with a hosted team product planned for the future. The project reached v0.1.3 in late September 2026 and has accumulated over 1,600 GitHub stars since its August 2026 launch.

What It Is

Whiteboard sits at the intersection of code review and AI agent collaboration. Rather than reading raw diffs in a terminal or a generic pull-request UI, developers open a "review" — a structured document generated by a coding agent (Claude Code, Codex, or similar) — inside the Whiteboard desktop app. The app renders that review as a rich canvas: sequence diagrams, entity-relationship diagrams, semantic diffs, and agent decision logs, all linked back to the underlying source code. Clicking a node in a diagram jumps directly to the relevant lines; clicking a quote from the agent's trace surfaces the authoring conversation that produced it.

Core Architecture

Whiteboard is built on a vendored fork of Code OSS (VS Code's open-source base), chosen deliberately so that developers get familiar keybindings and LSP support while reviewing code. The team strips out the portions of the VS Code codebase they don't need and merges upstream security and feature patches regularly. Three subsystems define the product:

• Semantic diff viewer — an AST-aware diffing engine written in Rust (published separately as the diffr library) that collapses noise: large added functions are summarized as pseudocode, unit tests and documentation changes are hidden by default, and the whole system is extensible via a WASM plugin API.
• Diagram canvas — agents use an SDK to draw sequence diagrams, DB diagrams, and other visualizations directly inside the app; each node is a live link to source.
• Decision log / trajectory viewer — agents can query and annotate their own execution traces so reviewers can see which requirements were set, how they were implemented, and which decisions the agent made autonomously.

Agent Integration Model

Whiteboard plugs into existing agent workflows rather than replacing them. The quickstart asks developers to connect Claude Code, Codex, or another coding agent from the welcome screen, then prompt the agent to review the current branch against main and open the result in Whiteboard. The README recommends GPT-6 Sol and Claude Opus 5.5 for the intelligence, cost, and speed tradeoff. Developers can highlight any section of a review, copy it to the clipboard with a keyboard shortcut, paste it into their agent, and ask follow-up questions — the agent can then redraw the canvas to reflect new understanding.

Open-Source Deployment Model

The core platform is MIT-licensed and self-hostable. Whiteboard runs entirely against local checkouts; anonymous telemetry is opt-out and explicitly excludes code, diffs, prompts, and model output. The team publishes a full telemetry reference in the repository. A hosted product for teams is on the roadmap, and the project states that self-hosting will always remain an option.

Update: v0.1.3 (Review Desktop 0.1.3)

The latest release, v0.1.3, was published on September 26, 2026, roughly five weeks after the repository was created on August 18, 2026. The project is in active early development — 34 open issues, 75 forks, and regular pushes — with known limitations including no in-app file editing, limited multi-repo support, and share links that do not propagate post-share updates. The team is iterating quickly on the agent SDK and system prompt defaults, and the open-source nature of the project is explicitly tied to that iteration goal.