🎨 Whiteboard (YC W26)

An open-source IDE designed for intentional software architecture

Whiteboard is a community-driven, open-source desktop application that creates a shared workspace where human developers and AI agents can collaborate on software architecture. Unlike traditional editors, it provides agents with a dedicated SDK, allowing them to visually map out their logic and design choices on an integrated canvas.

Whiteboard Demo

Watch the 1-minute demo video here

🚀 Getting Started

To begin using Whiteboard, follow these steps:

Install: Download the app for macOS or Fedora.
Integrate: From the welcome screen, link your preferred coding agent (e.g., Claude Code or Codex).
Execute: Request that your agent review your current branch against the latest main and launch the results within Whiteboard.

🧠 Model Guidance

The developers suggest using high-intelligence models to balance performance and cost. The ideal model selection can be thought of as an optimization problem:

$\text{Optimal Model} = \max\left(\frac{\text{Intelligence} \times \text{Speed}}{\text{Cost}}\right)$

Recommended Model	Strength	Use Case
`GPT-6 Sol`	Intelligence/Speed	Complex architectural shifts
`Claude Opus 5.5`	Nuance/Reasoning	Deep code reviews & design

🛠️ Effective Usage & Prompting

While the system prompt is being refined to bake in best practices, you can use the following patterns to get the most out of the canvas:

For API Modifications:

"This commit stack is designed to let [api] achieve [objective]. 
Please provide: 
1. Proposed API examples 
2. Motivations (based on repo context) 
Then we can move to implementation."

For Telemetry Integration:

"Explain the telemetry changes in this commit: 
https://github.com/devdotfast/whiteboard/commit/4837e107946e27ebad50c282eb0f2585210d2a35 
What is being tracked, and how should we structure our product waterfalls/dashboards?"

Iterative Design: If a visualization isn't quite right, simply highlight the section, send it to your agent, and it will ~~keep the old version~~ re-draw the canvas to match your requirements.

✨ Key Features

1. From Diagrams to Implementation

Traditional HTML tools fail to bridge the gap between a specification and the actual code. Whiteboard solves this by making visualizations interactive.

2. Rust-Powered Semantic Diffing

To eliminate the noise of raw diffs, Whiteboard utilizes a Rust-based, AST-aware diff viewer.

Summarization: Massive new functions are condensed into pseudocode.
Filtering: Documentation and unit test changes are automatically collapsed.

3. The Decision Log

To solve the "black box" problem of autonomous agent changes, Whiteboard allows agents to query and link their own traces. This creates a visual map of:

Initial requirements $\rightarrow$ Implementation steps $\rightarrow$ Autonomous design decisions.

💻 Technical Foundation & OSS

Whiteboard is MIT-licensed and operates directly on your local checkouts. While a hosted team version is in the works, the tool will always remain self-hostable.

The `Code-OSS` Approach

The team chose to vendor a fork of Code-OSS (the base of VS Code) rather than maintaining a patch-based fork.

Why?

Agent Compatibility: Coding agents struggle to navigate complex patches.
Bloat Reduction: Much of modern VS Code is heavily integrated with Copilot ( $\approx 45\%$ of the codebase), which is unnecessary for this specific review-centric tool.

The team continues to monitor upstream Code-OSS to merge critical security and feature updates.

⚠️ Current Limitations

While powerful, Whiteboard has some known gaps:

~~Direct file editing~~ $\rightarrow$ Read-only (Please file an issue if you need editing!).
Limited support for browsing/working across multiple repositories in one review.
Shared reviews are static; updates made after sharing do not sync to other users.

🔒 Privacy & Contribution

Telemetry: Anonymous data is collected, but it strictly excludes your code, diffs, prompts, model outputs, or canvas text. You can disable this in settings.
Contributing: Please refer to CONTRIBUTING.md for the PR workflow and adhere to the Code of Conduct.

Inspiration: This project was heavily influenced by Geoffrey Litt's essay, "Understanding is the new bottleneck".