Launched this week

DrawMode
Give your AI agents software diagramming superpowers
13 followers
Give your AI agents software diagramming superpowers
13 followers
Turn your codebase into editable architecture diagrams with Claude Code or Codex through MCP. Analyze your systems with one prompt and view the diagrams. Edit the design visually, then ask your agents to implement it. Collaborate live and import Mermaid or PlantUML. On the desktop, diagrams live in your Git repo beside the code, and even complex ones merge seamlessly across branches.






Free Options
Launch Team

AWS Startup AdvisorPersonalized AWS guidance that meets you where you’re already building
Promoted
Hi Product Hunt! I built DrawMode because I wanted software diagrams that my coding agents could read and edit as easily as I can.
Point Claude Code or Codex at your codebase and ask for its architecture, a C4 view or the domain model, and it draws real, editable diagrams in DrawMode over MCP. Change the design by hand, and the agent sees exactly what you changed, then implements it.
That works because DrawMode understands the structure of your diagrams, not just how they look. Every shape is a typed element and every line a typed relationship, so an agent reads the model behind the picture: what inherits from what, what owns what, what calls what and what sits inside what.
The same engine that runs live collaboration on the web makes DrawMode work with Git on the desktop. Diagrams live in your repository beside the code, as plain JSON with a log of their changes, so even complex diagrams merge seamlessly across branches, with no conflicts to resolve. Switch branches and the open diagrams update in front of you.
Three short videos show it in use: Introduction, Editing & Working with Agents and Desktop to Web Collaboration.
Community is free forever. MCP and team collaboration are in Professional, with a 14-day trial on every new account.
I'd love to hear what you'd draw with it, and what's missing.
a few technical details and unusual features that may be of interest:
The agent's changes appear on the same canvas as your, and each human collaborator / agent has their own undo/redo stack.
You can also ask it to explain a particular code path visually, with links back to the source - I often ask it to make sequence diagrams of complex code paths.
Recently I used this to build a set of diagrams showing where our system was struggling to scale, then got the agent to work through a refactoring plan.
You can also paste Mermaid or PlantUML or code directly onto the canvas, edit the resulting diagrams visually, and export back to those formats. You can use the editor without AI, too.
Stack and implementation notes:
Python/NiceGUI, with JointJS handling much of the drawing, interaction and link routing.
Separate collaboration and persistence interfaces, so desktop diagrams can live in your Git repository while people collaborate on them from the web. Web versions stores in postgres.
Local JSON files with JSONL change logs, field-level merging for collaboration, and seamless Git automerge support. Switching branches updates the diagrams in the editor immediately.
No full CRDT but most of the benefits - tombstoned shape deletes with field level merging for local undo/redo in a collaborative setting. Lighter weight, very fast.
You can export simplified "pure" data models, stripped of visual formatting. Feed it to your downstream tools using the cli tool.