Built for teams who want to move fast
AccessAgent combines an AI builder, a universal widget runtime, and real-time team collaboration — all in one open source package you can self-host.
Everything your team needs
AccessAgent combines an AI builder, a widget runtime, and team collaboration in one open source package.
AI-powered building
Chat with an AI agent to build, modify, and debug your widgets. No drag-and-drop, no code. Just describe what you want.
Build anything
Widgets are standard HTML, JS, and CSS rendered in a sandboxed iframe. Games, dashboards, reports, forms — no constraints.
Real-time team collaboration
Every project has a live team chat channel. Discuss changes, share feedback, and iterate together in real time.
Organised workspace
Three-column layout keeps everything in reach: your project list, the active widget, and team chat — all visible at once.
Self-hostable
Deploy on your own server with a single Docker command. Your data never leaves your infrastructure.
Open source
MIT licensed. Read the code, contribute, self-host, or fork it. No black boxes, no lock-in.
Pluggable auth & storage
Swap in your own auth provider or storage backend. The core ships with sensible defaults you can replace.
Session continuity
Agent conversations persist across sessions. Come back days later and the agent remembers exactly where you left off.
From idea to working tool in minutes
No setup, no learning curve. If you can describe what you want, the AI can build it.
Create a project
Give your project a name. It immediately appears in the left panel for everyone on your team.
Chat with the AI agent
"Build me a sales dashboard with this month's data." The agent starts building immediately, writing HTML, JS, and CSS in a sandbox.
See it live instantly
Every file the agent writes renders live in the center panel. Watch your widget take shape in real time.
Iterate and collaborate
Ask the agent to change anything. Discuss it with your team in the right panel. Ship when it's ready.
Build anything that runs in a browser
Widgets are plain HTML, JavaScript, and CSS rendered inside a sandboxed iframe. The AI agent writes these files directly — it can use any library available on a CDN, any layout, any functionality.
- Data dashboards and charts
- Interactive forms and calculators
- Games and simulations
- Reports and document generators
- Internal CRUD tools
- Data visualizations
<!DOCTYPE html>
<html>
<head>
<script src="https://cdn.jsdelivr.net/npm/chart.js">
</script>
</head>
<body>
<canvas id="chart"></canvas>
<script>
new Chart(document.getElementById('chart'), {
type: 'bar',
data: { ... }
});
</script>
</body>
</html>Add a date range picker that filters the revenue chart
I'll add a date picker using Flatpickr and wire it to the chart data. One moment...
Great — now make it default to the last 30 days
Done. The picker now initialises to today minus 30 days. The chart updates automatically when the range changes.
Iterates as fast as you can type
The agent remembers every file it has written and every conversation turn. Ask it to change colours, add interactivity, connect to an API, or start over — it handles all of it without losing context.
Powered by the OpenAI Agents SDK with sandboxed file access per project. Sessions persist so you can continue building days later.