Skip to main content

Smarter Flet Studio with AI Agent

· 5 min read
Feodor Fitsner
Feodor Fitsner
Flet founder and developer

In case you missed it, we have Flet Studio - an online tool for building Flet apps and sharing them with other users. What started as a quick "FletPad" experiment has already grown into a solid app - an online Flet IDE, if you will - with user accounts, a forkable gallery of examples, multiple apps per account, and version history. You can play with Flet and learn the framework before ever running pip install flet on your machine. Flet Studio itself is written in Flet, with a custom FastAPI backend, and it's a fantastic opportunity for us to dogfood Flet ourselves.

Our vision is to shape Flet Studio into a hub of online services that help you build better Flet apps. The first such service was an online editor with the ability to share your apps with other users.

Today we are introducing a new member of the family - the AI agent!

Go to https://flet.app and ask something like:

...and in a few moments you get code you can start working with:

The generated code is not your typical AI slop - it's compact and readable to any Python developer. Maintainable code is an explicit goal for us, and we are building tools for the agent to achieve exactly that. It helps that Flet itself, with its minimal boilerplate, is a framework equally friendly to humans and AI.

Why an AI agent?​

  • Because it's fun! Every program today has an agent, right? :)
  • It helps you start a new project - and once it's off the ground, you can download it and continue developing on your computer.
  • It's a great way to explore and learn the Flet framework: "How do I do this? How do I do that?"
  • You can ask it to fix your own app.

We don't see the built-in agent as just a coding agent for your Flet apps - Claude or Codex would probably do a better job there - but rather as the central "brain" of Flet Studio, coordinating a bunch of tools and data sources. Over time it will help you with app deployments, configuring app backends, and troubleshooting errors.

Today's agent is just the beginning of our long AI journey and, bear with us, it can be dumb sometimes :) Let us know if you see a way to improve its behavior or teach it something!

Is it free?​

Every account on the new free "Explorer" plan gets 1,000 credits per month for free.

Frankly, 1,000 credits is not enough to use the agent as your everyday tool, but it's quite enough to build 3-4 simple apps and get a taste of the framework. We are not planning to make our profit reselling AI tokens - more interesting value-added services are in the works - but if you want to play with AI longer, or need more fuel to finish your app, you are not blocked: you can buy more credits or subscribe to a plan.

There is a "Creator" plan which includes 10,000 AI credits per month, "unlimited" app quotas, and Flet support! We are still fleshing out what exactly "Flet support" means, but in essence it's your way to get closer to the Flet team and move ahead of the line with your feature requests and issues.

Flet MCP server​

We are not going to pretend the Flet agent competes with your favorite local agent - Claude, Codex, and others are way more powerful, including at developing Flet apps. But Flet is a fast-growing framework with an evolving API, and most LLMs lag behind with their training data. Flet 0.86 introduces the "official" Flet MCP server with an up-to-date dataset that helps your agent make smarter decisions when tasked with building a Flet app. In our experiments, adding the Flet MCP server alone reduces Flet API hallucinations next to zero.

Flet MCP supports stdio and HTTP transports and provides tools organized into groups:

  • API - look up any Flet control, property, event, or enum as it exists in the installed Flet version.
  • Icons - find the right icon by keyword.
  • Examples - search and fetch working code examples.
  • CLI - get help for flet CLI commands.

Check the Flet MCP server cookbook for more details and examples.

Flet Skills​

Agent Skills are another layer of Flet AI "wisdom". We are going to work on skills in the coming weeks, so expect the Flet agent to become smarter - today it still struggles in some areas like routing/navigation or declarative apps.

Try it​

Open flet.app, sign in, and ask the agent to build something - your free monthly credits are already waiting. If you'd rather stay with your local agent, plug in the Flet MCP server and enjoy hallucination-free Flet coding right away.

And tell us how it went: what the agent nailed, where it stumbled, and what you'd like it to learn next. Drop your feedback in GitHub Discussions or on Discord - this is the very beginning of the journey, and your reports are exactly what will make the agent smarter.

Happy Flet-ing!