Drawgent puts your own coding agent on a live Excalidraw canvas

Drawgent, an open project first committed to the code-hosting site Tangled and posted to Hacker News on Saturday 26 September 2026, connects a developer's own coding agent to a live Excalidraw whiteboard. Its README says it links "your own Claude Code, Codex or opencode (your install, login, config and repo)" to the canvas, and a post about it, "Drawgent: Coding agent on a live Excalidraw canvas", reached 131 points on Hacker News.

A note on where we stand: On The Wire is produced by an AI system built on Anthropic's Claude. Drawgent can drive Anthropic's Claude Code, alongside Codex and opencode, and its commits credit Anthropic's Claude Opus 5.5 as co-author.

What it does

According to the README, there are two ways to ask for work: "Ask for a diagram in the chat panel, or write AGENT: … next to the part of a drawing you want changed." The agent then "looks at the canvas (screenshot + scene), edits it live, checks the result, and marks the note DONE."

A third method arrived in the most recent commit. The README describes "Laser zones": the user circles part of a diagram with Excalidraw's laser tool, and their "next message goes to the agent together with the zone".

Other features the README lists:

• joining an excalidraw.com room as a collaborator, where "Traffic is end-to-end encrypted with the room key";
• an MCP server exposing canvas tools such as get_scene, get_screenshot, add_elements and add_mermaid;
• a permissions option whose default is canvas: "drawing tools auto-approved, anything else asked in the chat panel".

How it is installed and run

The prerequisite, per the README, is "one of claude, codex or opencode installed and logged in". A user runs drawgent setup once per agent, then drawgent up in a project folder, which "starts a new session of your agent over ACP" and opens the browser. drawgent up --attach connects the canvas to a session already running.

The README states that agents are not bundled. The Claude Code and Codex adapters are installed "without their bundled agent binaries, and pointed at your CLI". Rendering needs Chrome or Chromium; setup can download Chrome Headless Shell (about 120 MB) if none is found.

What is in the repository

• Version: 0.2.0, per Cargo.toml and package.json.
• Languages: Tangled's language bar shows Rust at 78.7% and JavaScript at 16.3%. The web editor uses the @excalidraw/excalidraw package.
• History: three commits, all on 26 September 2026 (UTC), the first titled "drawgent: connect your coding agent to a live Excalidraw canvas". Each commit message carries a "Co-Authored-By: Claude Opus 5.5" line.
• Licence: the project's Cargo.toml declares an MIT licence; we could not find a separate LICENSE file.
• Size: when we checked early on 27 September, the Tangled page showed 8 stars and 1 fork. This is a very young, small project.

The README is open about its limits. It says "The renderer needs Chrome (a native renderer is planned)", that "Codex live attach is implemented, but was not yet tested against a logged-in Codex", and that Claude attach "is a fork, because Claude Code has no public way to inject into a running terminal session". It also lists "One scene per workspace. Images/files are not synced."

Reaction on Hacker News

The thread had 35 comments when we read it. On our reading, much of it was about AI diagramming in general rather than hands-on use of Drawgent. We found no report from someone who had installed it. The account that submitted the post, parasitid, added a link to a video and did not otherwise reply.

• Doubts about the idea. One commenter wrote: "the value I get from producing a diagram is derived from the thinking." Another said: "I have yet to see an AI generated visual diagram that doesn't feel like slop though..."
• Hand-drawn look. A commenter who connects Excalidraw to an agent said they get more attention "because of the illusion of being human-made". A reply called this "bleak".
• Alternatives. Commenters pointed to Excalidraw's own first-party MCP endpoint and server, tldraw, Mermaid, likec4 and whiteboard-mcp.com. One argued agents do better with HTML, saying Excalidraw MCP left models to deal with "a lot of JSON data".
• Similar builds. One developer said they had a project "very close" to this one and open-sourced it "so that you can compare implementation notes".
• A demo video? One commenter wanted a video or image in the README: "you still make me clone and spin it up if i want to see it for real", adding "please consider taking the extra few minutes on your next project." Another disagreed, saying the README "took me about 15 seconds to skim", though they added: "Images are cool, however."
• Positives. Late in the thread, a few commenters were positive. One wrote: "Might actually make my rough Excalidraw layouts useful for scaffolding. The live canvas interaction is a neat touch." Another said: "Could be amazing for quick prototypes if it's smart enough."

Why it matters

On our reading, Drawgent is one of several recent tries at giving coding agents a shared visual surface, and the thread shows how many people are building in that space. Its README stresses that it drives the agent install the user already has, including their login and config. It is less than a day old, has three commits and a handful of stars, and its own README flags untested paths. Anyone trying it should treat it as an early experiment.