Tabucom: HTML Plans for AI Agents

I built Tabucom for a very simple reason: I wanted AI agents to share work in a format people would actually read.

Markdown is good for a lot of things. It is quick, portable, and easy for an agent to write. But when the output is a plan, a product spec, a roadmap, or a technical proposal, Markdown often becomes a wall of text. The structure is there, but the signal is not always visible.

HTML changes that. A plan can have sections, diagrams, tables, callouts, status markers, screenshots, and lightweight interaction. It can feel closer to a small internal product than a text file. For many ideas, that is the difference between “I skimmed it” and “I understood it.”

Tabucom is a small open source service for publishing that kind of work. An AI agent can send Markdown, a single HTML file, or a prebuilt static site ZIP, and Tabucom returns a URL. The deployment is immutable and expires automatically.

That is the whole point: remove the ceremony between an agent producing a useful front-end and a human being able to open it.

AgentMarkdownHTML specStatic appShareable URL

Why HTML plans

I started thinking about this around plans.

When an agent writes a plan in Markdown, the content can be correct and still be hard to consume. You get headings, bullets, and code blocks. That is fine for a developer who already knows the context. It is less fine when you want to share the plan with a product person, a manager, a designer, or another team.

An HTML plan can make the shape of the work obvious:

• what is in scope
• what is not in scope
• which systems are touched
• what decisions are still open
• what risks matter
• what the sequence of work looks like

The content does not need to be fancy. It just needs enough layout and visual hierarchy to make the important parts easy to find.

This also works for specs. Instead of asking an agent for a long document, I can ask it for a small single-page spec with a few states, examples, diagrams, and implementation notes. Then I can share the URL and get feedback.

Faster feedback

The interesting part is not hosting. Hosting static files is not new.

The interesting part is making publishing cheap enough that it becomes part of the agent workflow. An agent should be able to produce something, publish it, and hand back a URL without opening a ticket, asking for an account, or setting up a project.

That matters for prototypes. If I want to explore a small UI idea, the agent can create a single HTML page or a tiny React app, publish it, and send it to the people who need to react to it. The discussion moves from “imagine this screen” to “open this and tell me what is wrong.”

It also matters for internal communication. Roadmaps, implementation plans, architecture proposals, migration dashboards, release notes, and workshop material can all be richer than Markdown without becoming full applications.

What Tabucom does

Tabucom keeps the model deliberately small:

• publish Markdown, HTML, or a prebuilt static site ZIP
• return an immutable URL
• expire deployments automatically
• support SPA fallback when a small front-end needs client-side routing
• run as one small container with local or S3-compatible storage
• expose API documentation and agent-discovery endpoints

It is meant for trusted internal environments. It serves uploaded files; it does not build or execute them server-side. If you want isolation for previews, you can configure wildcard preview domains and put it behind the ingress, VPN, or SSO layer you already trust.

That boundary is important. Tabucom is not trying to be a full deployment platform. It is a narrow tool for temporary front-end publishing.

The workflow I want

The workflow I care about looks like this:

idea -> agent creates plan.html -> Tabucom publishes it -> colleagues review it

Or:

spec -> agent creates a small interactive prototype -> Tabucom shares the URL

Or:

roadmap notes -> agent turns them into an HTML presentation -> team discusses it

That is where AI agents become more useful. Not only writing code, but preparing work in a format that helps people make decisions faster.

Tabucom is a small piece of infrastructure for that. It gives agents somewhere simple to put the front-ends they create, and it gives teams a better artifact than another Markdown file in a chat thread.

The project is open source here: github.com/markthebault/tabucom.