====== AutoGen Studio ======

**AutoGen Studio** is Microsoft's low-code visual interface for building, testing, and deploying multi-agent AI workflows. Built on top of the [[https://github.com/microsoft/autogen|AutoGen framework]] (AgentChat), it provides a drag-and-drop environment where developers and non-developers alike can compose agent teams, assign tools and skills, and orchestrate complex multi-step conversations — all without writing extensive code. It serves as both a rapid prototyping tool and a gateway to production deployment via Docker and Azure.

===== What Is AutoGen Studio? =====

AutoGen Studio is a web-based application that sits on top of Microsoft's AutoGen framework. While AutoGen itself is a Python library for programmatic multi-agent orchestration, AutoGen Studio provides:

  * A **visual workflow builder** for composing agent teams
  * A **gallery of pre-built agents** with configurable roles and capabilities
  * **Real-time testing** with visibility into agent "inner monologues"
  * **Profiling data** showing tool usage, token costs, and code execution outcomes
  * **Export capabilities** for deploying workflows as JSON, Python apps, or Docker containers

It is built on AutoGen AgentChat, the high-level API for multi-agent applications, and leverages AutoGen's asynchronous, event-driven architecture (v0.4+).

===== Visual Workflow Builder =====

The Studio's **Build** section provides a WYSIWYG interface for composing agent teams:

  * **Drag-and-Drop** — Select agents from the gallery and arrange them into teams
  * **Agent Configuration** — Assign foundation models (GPT-4o, Azure OpenAI, local models), system prompts, and tools
  * **Workflow Types** — Choose between sequential execution (predefined order) or autonomous chat (LLM-driven agent selection)
  * **Skills Assignment** — Attach Python functions as tools (weather APIs, database queries, web search, etc.)
  * **Team Composition** — Build hierarchical or flat agent topologies for different task types

===== Team Types =====

AutoGen Studio supports the group chat patterns from AutoGen AgentChat:

  * **RoundRobinGroupChat** — Agents take turns in a fixed sequential order. Each agent processes the conversation and contributes its perspective before passing to the next.
  * **SelectorGroupChat** — An LLM or custom logic dynamically selects which agent should respond next based on the current conversation state and task requirements.
  * **Custom Workflows** — Define arbitrary agent interaction patterns with conditional routing and branching logic.

===== Gallery of Pre-Built Agents =====

The Studio includes a library of pre-configured agents for common tasks:

  * **Coder** — Writes and debugs code based on specifications
  * **Planner** — Decomposes complex tasks into subtasks
  * **Critic** — Reviews and provides feedback on outputs
  * **Researcher** — Gathers information from tools and APIs
  * **Custom Agents** — Create new agents with specific roles, prompts, and tool access

Users can share and reuse agents, skills, and complete workflow configurations across projects.

===== Skills and Tools Management =====

Skills in AutoGen Studio are Python functions registered as tools:

  * Define functions with docstrings for LLM understanding
  * Register skills through the visual interface
  * Agents invoke skills conditionally based on task needs
  * Results feed back into prompts and agent memory
  * Secure execution via Docker containers prevents untrusted code from affecting the host

===== Deployment Options =====

AutoGen Studio workflows can be deployed through multiple channels:

  * **JSON Export** — Save workflows as JSON for loading into Python apps
  * **Command-Line API** — Run workflows from scripts and CI/CD pipelines
  * **Docker Containers** — Package for Azure Container Apps, Web Apps, or Kubernetes
  * **Azure Integration** — Native connectors for Azure OpenAI, Dynamics 365, and Azure DevOps
  * **GitHub Actions** — Integrate into CI/CD workflows for automated agent tasks

===== Installation and Usage =====

<code python>
# Install AutoGen Studio
# pip install autogenstudio

# Launch the web interface
# autogenstudio ui --port 8081

# Alternatively, use AutoGen programmatically with the same team patterns
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_agentchat.conditions import TextMentionTermination
from autogen_ext.models.openai import OpenAIChatCompletionClient

# Define agents
planner = AssistantAgent(
    name="Planner",
    model_client=OpenAIChatCompletionClient(model="gpt-4o"),
    system_message="You break down complex tasks into steps.",
)

coder = AssistantAgent(
    name="Coder",
    model_client=OpenAIChatCompletionClient(model="gpt-4o"),
    system_message="You write Python code to implement the plan.",
)

reviewer = AssistantAgent(
    name="Reviewer",
    model_client=OpenAIChatCompletionClient(model="gpt-4o"),
    system_message="You review code for bugs and suggest improvements. Say APPROVE when done.",
)

# Create a team with round-robin orchestration
termination = TextMentionTermination("APPROVE")
team = RoundRobinGroupChat(
    participants=[planner, coder, reviewer],
    termination_condition=termination,
)

# Run the team on a task
import asyncio

async def main():
    result = await team.run(
        task="Create a Python function that calculates compound interest."
    )
    print(result)

asyncio.run(main())
</code>

===== Architecture Diagram =====

<mermaid>
graph TD
    A["AutoGen Studio (Web UI)"] --> B["Build Section"]
    A --> C["Test Runner"]
    A --> D["Deploy / Export"]
    B --> E["AutoGen AgentChat Framework"]
    C --> E
    D --> E
    E --> F["Agents Library"]
    E --> G["Teams (Groups)"]
    E --> H["Tools Registry"]
    F --> I["LLM Providers (Azure OpenAI / OpenAI / Local)"]
    G --> I
    H --> I
</mermaid>

===== Current Status =====

AutoGen Studio is under active development as part of the broader AutoGen ecosystem. Key points:

  * Built on AutoGen v0.4+ with async, event-driven architecture
  * Positioned as a **prototyping-to-deployment** tool, not just a research prototype
  * Part of Microsoft's Agent Framework unification strategy
  * Integrates with Azure AI services for enterprise deployment
  * Open-source under MIT license

===== References =====

  * [[https://github.com/microsoft/autogen|AutoGen GitHub Repository]]
  * [[https://microsoft.github.io/autogen/|AutoGen Documentation]]
  * [[https://www.microsoft.com/en-us/research/blog/introducing-autogen-studio-a-low-code-interface-for-building-multi-agent-workflows/|Microsoft Research: Introducing AutoGen Studio]]
  * [[https://autogen-studio.com/|AutoGen Studio Website]]

===== See Also =====

  * [[swarm_openai|OpenAI Swarm]] — Lightweight multi-agent orchestration framework
  * [[composio|Composio]] — Tool integration platform for agents
  * [[e2b|E2B]] — Sandboxed code execution for agent workflows
  * [[modal_compute|Modal]] — Serverless compute for agent deployment