Workflow:Microsoft Autogen Studio Team Deployment
| Knowledge Sources | |
|---|---|
| Domains | Multi_Agent_Systems, No_Code, Team_Deployment, Web_Application |
| Last Updated | 2026-02-11 18:00 GMT |
Overview
End-to-end process for building, testing, and deploying multi-agent teams using AutoGen Studio's visual team builder, gallery system, and programmatic deployment options.
Description
This workflow demonstrates how to use AutoGen Studio to visually compose multi-agent teams without writing code, test them interactively in the playground, and deploy them as API endpoints or embed them in Python applications. AutoGen Studio provides a web-based interface built on React and FastAPI, with a drag-and-drop team builder, component galleries for sharing configurations, and multiple deployment paths.
The process covers launching the studio, building a team from gallery components, testing it in the chat playground, exporting the team configuration, and deploying it either as a standalone API or through the LiteStudio programmatic interface.
Usage
Execute this workflow when you want to build and deploy multi-agent teams through a visual interface rather than writing Python code directly. Typical scenarios include:
- Rapid prototyping of agent team configurations with immediate testing
- Non-developers building agent workflows using pre-built gallery components
- Deploying a tested team configuration as an API endpoint for production use
- Sharing team configurations across teams via the gallery import/export system
You have AutoGen Studio installed and want to visually compose, test, and deploy an agent team.
Execution Steps
Step 1: Launch AutoGen Studio
Start the AutoGen Studio web application using the CLI command. The studio provides a full web interface for building and testing agent teams. Configuration options include host, port, database URI, authentication settings, and working directory.
Launch options:
- Full UI mode via the "ui" command with configurable workers and auto-reload
- Lite mode via the "lite" command for lightweight single-team operation
- Environment variables control API keys and model configuration
Step 2: Configure Component Gallery
Set up or import a component gallery containing reusable building blocks. The default gallery includes pre-configured agents (AssistantAgent, WebSurfer), model clients (OpenAI, Azure), tools (calculator, web search, image generation), and team templates (RoundRobin, Selector, Swarm).
Gallery capabilities:
- Import galleries from JSON files, URLs, or the built-in defaults
- Each gallery contains agents, models, tools, termination conditions, and workbenches
- Components are serialized as ComponentModel objects for portability
- Custom galleries can be created and shared across organizations
Step 3: Build the Team Visually
Use the drag-and-drop team builder to compose a multi-agent team. The ReactFlow-based canvas lets you add agents from the gallery, connect them to model clients and tools, configure team orchestration patterns, and set termination conditions.
What happens:
- Drag agent components from the library onto the canvas
- Configure each agent's system message, model client, and tools
- Select the team type (RoundRobin, Selector, Swarm, MagenticOne)
- Add and compose termination conditions
- The visual graph shows agent relationships and data flow
Step 4: Test in the Playground
Run the team interactively in the chat playground. Send messages and observe how agents collaborate in real time via WebSocket-powered streaming. The playground shows individual agent messages, tool calls, and the flow visualization.
Testing features:
- Real-time message streaming with agent attribution
- Agent flow visualization showing communication patterns
- Session management for organizing test conversations
- Compare mode for evaluating different team configurations side by side
- Human-in-the-loop input with configurable timeout
Step 5: Export Team Configuration
Export the validated team configuration as a JSON file. This declarative configuration captures the complete team specification including all agents, model clients, tools, termination conditions, and orchestration settings. The JSON can be version-controlled and shared.
Configuration format:
- Component-based JSON with provider class paths
- Nested structure: team contains participants, each with model_client and tools
- All parameters are serialized including system messages and tool schemas
- Compatible with TeamManager.load_from_file for programmatic use
Step 6: Deploy the Team
Deploy the team configuration through one of several paths: as an API endpoint via the "serve" command, programmatically via TeamManager or LiteStudio in Python, or as a Docker container for production environments.
Deployment options:
- CLI serve: Expose team as a FastAPI endpoint from a JSON config file
- LiteStudio: Lightweight programmatic wrapper that loads config and serves it
- TeamManager: Load config and run teams directly in Python applications
- Docker: Containerize the studio for production deployment with full UI or API-only mode