A
Agentable
Get Started GitHub
Open source • MIT licensed

The agent-native
canvas framework.

A tldraw-based workspace where humans and AI agents collaborate safely on shared panels.

View on GitHub
OUR STORY

Built for the agent era

Agentable is a shipped, production-ready MIT-licensed framework that turns any interface into an agent-native workspace. Built on tldraw, it gives both humans and AI agents a shared canvas governed by clear rules and human-in-the-loop approval.

We believe UI should be data. Hosts define validated panel specs. Agents can open, fill, compose, and mutate panels at runtime — always with human oversight. The result is a multi-agent world model that feels natural and safe.

Agentable canvas workspace showing shared panels between humans and agents
PRINCIPLES

Human-in-the-loop by design

Human-in-the-loop

Every agent action that mutates state passes through explicit human approval gates. Safety is not optional.

Multi-agent world model

Multiple agents operate simultaneously with a shared understanding of the canvas state and panel catalog.

Four panel tiers

Static, schema-driven, agent-composed, and custom React panels. The right abstraction for every use case.

Built on proven foundations

tldraw + Lit + React. Production tested.

Core
Agentable is built on tldraw for the infinite canvas substrate, Lit for lightweight web component embeds, and React for deep host integrations. The entire system is shipped and running in production today.
Embed
Drop a single <agentable-canvas> Lit component into any page. Configure tenant, primary color, welcome message, and canvas mode in seconds.

Ready to explore the source?

Agentable is fully open source under the MIT license.

Visit GitHub Repository