Whiteboard: An Open-Source IDE for Agentic Software Design

Whiteboard enables collaborative software architecture between humans and AI agents

Whiteboard is an open-source desktop application designed to move software design beyond text-based prompts and raw diffs. It provides a common workspace where humans and AI agents can architect software together, using a visual canvas to describe system layouts, flows, and logic. By plugging into existing coding agents like Claude Code and Codex, Whiteboard gives agents an SDK to draw directly on an in-app canvas, allowing them to visualize their proposed changes before or during implementation.

Key Technical Features

Diagrams Linked to Code

Whiteboard solves the disconnect between architectural specifications and actual implementation. In this workspace, visualizations—such as sequence diagrams and entity-relationship diagrams—are semantically linked to the underlying codebase. Users can click on a visualization or a quote from an agent's trace to jump directly to the corresponding code. To maintain a high-quality editing experience, Whiteboard vendors Code OSS, providing VS Code's keybindings and LSP support out of the box.

AST-Aware Semantic Diff Viewer

To reduce the noise associated with traditional line-by-line diffs, Whiteboard implements a semantic diff viewer written in Rust. This viewer is AST-aware, meaning it understands the structure of the code rather than just the text. By default, the tool summarizes large added functions as pseudocode and collapses non-essential changes, such as documentation or unit tests, to help reviewers focus on the logic that matters.

Agent Decision Logs

Whiteboard includes tools for agents to query and link their own traces. This creates a decision log that allows humans to visualize the requirements they set and understand the autonomous decisions the agent made during the implementation process, reducing the "black box" nature of agentic coding.

Operational Details and Limitations

Local-First Architecture and Privacy

Whiteboard is MIT-licensed and operates against local checkouts of code. According to the project's privacy policy, anonymous telemetry does not include code, diffs, Whiteboard text, prompts, or model outputs.

Current Limitations

  • No File Editing: Users cannot currently edit files directly within Whiteboard; it functions primarily as a review and design tool.
  • Single Repository Focus: Browsing files across multiple repositories in a single review is not currently supported.
  • Static Shared Reviews: While reviews can be shared between machines, updates made after the initial share do not sync automatically.

Community Feedback and Analysis

Community discussion on Hacker News highlights both the potential for reducing cognitive load and concerns regarding the tool's place in the development lifecycle.

Value Proposition

Many users noted that visual memory is more efficient for understanding complex systems than reading long traces or plans. One user highlighted the utility for corporate environments:

"I just use my coding agent and this is 'plugin' for visualisation... No reasons to not try it for me!"

Critical Counterpoints

  • The "IDE" Definition: Several users questioned the use of the term "IDE" given that the tool currently lacks the ability to edit files.
  • Source of Truth Divergence: A common concern is that diagrams may become an "N+1 source of truth," diverging from the actual code as the project evolves.
  • Platform Availability: There is significant demand for Windows and Linux versions, as the current release is focused on macOS.
  • Accuracy Concerns: Some users expressed skepticism regarding the accuracy of LLM-generated diagrams, noting that hallucinations in a visual flow could lead to wasted development time.

Integration with Existing Workflows

Some developers argued that the tool's scope is narrow, suggesting it might be a feature (such as an MCP with a GUI) rather than a standalone product, or that existing tools like Mermaid.js within Markdown files already satisfy the need for agent-generated diagrams.

Sources

Related

  • Project
    ganbo-gab/open-storyboard-canvas

    Open Storyboard Canvas is a cross‑platform desktop app (macOS/Windows) built with Tauri, React and Rust that provides a node‑based canvas for AI‑generated images, videos and 3‑D storyboard scenes. It includes a chat‑based Agent (beta) that can manipulate the canvas, submit generation jobs to multiple providers, track task status and show diagnostics. The tool bundles provider configuration, prompt libraries, director‑studio 3‑D scenes and full project persistence, targeting creators who want an integrated local workflow for AI‑driven visual storytelling.

  • Project
    NanmiCoder/cc-haha

    Claude Code Haha (cc‑haha) is an open‑source, cross‑platform desktop app that lets you run Claude‑style (or any OpenAI‑compatible) coding agents. It provides multi‑session workspaces, Git branch/worktree launch, diff viewer, built‑in browser preview, permission controls, skill marketplace, sub‑agent/agent‑team workbench, dynamic workflow orchestration, computer‑use (click/type), IM integration, remote H5 access, and scheduled tasks. Built with TypeScript, Electron, React, and Bun, it works on macOS, Windows, and Linux.

  • Project
    freestylefly/wesight

    An open-source desktop control console that provides a visual workspace and unified management for local AI coding agents.

  • Project
    777genius/agent-teams-ai

    A desktop orchestration layer for AI agent teams that allows multiple agents from different providers to collaborate autonomously on tasks via a visual Kanban board.

  • Project
    coulsontl/ai-toolbox

    AI Toolbox is a Tauri‑based cross‑platform desktop app that centralises configuration, credential and prompt management for dozens of AI coding assistants (OpenCode, Claude Code, Codex, Grok, Gemini, Kimi, etc.). It offers visual editors, a local proxy gateway, WSL/SSH sync, backup options, and a system‑tray quick‑switch, all stored in a local SQLite database.