A community based AI agent platform


🤖 Agentsflow

Easily create, connect & run autonomous autogen AI agents from a streamlined web interface.

With Agentsflow it is possible to create your own agents and connect them in a flow, so you can directly see which agent is running and also interact with each of them in runtime.

It is designed to make it easy for beginners to create software, but we also aim to give advanced users the tools to explore deeper possibilities.

AgentFlow Design


Ensure that git, nodejs, python 3.11 and poetry are installed, and then...

Download, build, and start agentsflow with the following commands:

git clone
cd agentsflow
npm install
npm run agentsflow

Open the following URL in your browser: localhost:4200 to load the UI.

If you want to make changes run the development server:

npm run agentsflow:dev


Goals may change and are unsorted

  • Add websockets to autogen agent.
  • Let the agent execute code.
  • Simple html websocket client.
  • Add nextjs as a frontend.
  • Implement react-flow.
  • Drag and drop to add agents to the flow.
  • Make the basic application work (connect agents, run agents, select agents, etc.)
  • Vector database for agents (
  • Popup for global settings (OpenAI API Key and model).
  • Agent Settings (model, name, llm_config..)
  • Popup to add own agent.
  • Save custom flow templates.
  • Save custom agents.
  • Export flow as a python script.
  • Add group chat functionality.
  • Add a documentation website.
  • Ability to use local LLM model.
  • Implement 3rd party tools (
  • Export/Import application settings/flows/agents.
  • Docker Compose file to quickly get started.
  • Create landing page.
  • Show generated files.
  • Make agents multimodal.

Get Involved

If you want to contribute to Agentsflow, be sure to review the contribution guidelines.

We use GitHub issues for tracking requests and bugs.

Please join the Agentsflow Discord for general questions, discussion and contribution.


This project is licensed under GNU General Public License (Version 3)