Implementation:Microsoft Autogen Studio Default Gallery

Knowledge Sources	python/packages/autogen-studio/frontend/src/components/views/gallery/default_gallery.json
Domains	Configuration, Component Library, Preset Definitions
Last Updated	2026-02-11

Overview

The Default Gallery JSON file provides a comprehensive collection of pre-configured AutoGen components including agents, models, tools, workbenches, terminations, and teams that serve as templates and examples for users.

Description

This JSON configuration file defines the default component gallery shipped with AutoGen Studio. The gallery contains:

Gallery Metadata:

• ID: "gallery_default"
• Name: "Default Component Gallery"
• Author: "AutoGen Team"
• Version: "1.0.0"
• Category: "conversation"
• Description: Basic components for human-in-loop conversations

Component Collections:

Agents (4):

• AssistantAgent with calculator tool - Basic AI assistant with arithmetic capabilities
• Web Surfer Agent - Headless browser agent for web scraping and navigation (uses gpt-4o-mini)
• Verification Assistant - Agent specialized in verifying and summarizing information
• UserProxyAgent - Human user proxy for interactive feedback

Models (4):

• OpenAI GPT-4o Mini - Fast, cost-effective model for most tasks
• Mistral-7B Local - Local model via LMStudio/Ollama
• Anthropic Claude-3-7 - Advanced reasoning model (claude-3-7-sonnet-20250219)
• Azure OpenAI GPT-4o-mini - Enterprise Azure-hosted variant

Tools (6):

• Calculator Tool - Basic arithmetic operations (+, -, *, /)
• Image Generation Tool - DALL-E 3 image generation (requires OpenAI API)
• Fetch Webpage Tool - Convert webpages to markdown (requires requests, beautifulsoup4)
• Bing Search Tool - Web search via Bing API (requires BING_SEARCH_KEY)
• Google Search Tool - Web search via Google Custom Search (requires GOOGLE_API_KEY, GOOGLE_CSE_ID)
• Python Code Execution Tool - Local code execution via LocalCommandLineCodeExecutor

Terminations (3):

• TextMentionTermination - Terminate on "TERMINATE" keyword
• MaxMessageTermination - Terminate after 10 messages
• OR Termination - Combination of text mention OR max message (flexible termination)

Teams (4):

• RoundRobin Team - Single AssistantAgent with calculator in round-robin pattern
• Selector Team - Two agents (assistant + critic) with model-based selection
• Web Agent Team (Operator) - Three-agent team with web surfer, verification assistant, and user proxy
• Deep Research Team - Three-agent research team with research assistant (Google search + fetch webpage tools), verifier, and summary agent using GPT-4o

Workbenches (3):

• Basic Tools Workbench - Static workbench with calculator and fetch webpage tools
• Web Content Fetch Workbench - MCP workbench using mcp-server-fetch for web content
• HTTP Streamable MCP Workbench - Example MCP workbench for HTTP-based MCP servers

The gallery demonstrates various component composition patterns, tool integration strategies, and team orchestration approaches. Many tools require external API keys (OpenAI, Bing, Google) documented in their descriptions.

Usage

This file is loaded by AutoGen Studio at startup to populate the default gallery. Users can import components from this gallery into their projects or use them as templates for creating custom components.

Data Structure Reference

Source Location: /tmp/kapso_repo_2mr4n2g4/python/packages/autogen-studio/frontend/src/components/views/gallery/default_gallery.json

Schema Structure:

interface DefaultGallery {
  id: string;
  name: string;
  url: string | null;
  metadata: {
    author: string;
    version: string;
    description: string;
    tags: string[];
    license: string;
    homepage: string | null;
    category: string;
    last_synced: string | null;
  };
  components: {
    agents: Component<AgentConfig>[];
    models: Component<ModelConfig>[];
    tools: Component<ToolConfig>[];
    terminations: Component<TerminationConfig>[];
    teams: Component<TeamConfig>[];
    workbenches: Component<WorkbenchConfig>[];
  };
}

Root Object:

{
  "id": "gallery_default",
  "name": "Default Component Gallery",
  "url": null,
  "metadata": {
    "author": "AutoGen Team",
    "version": "1.0.0",
    "description": "A default gallery containing basic components for human-in-loop conversations",
    "tags": ["human-in-loop", "assistant", "web agents"],
    "license": "MIT",
    "homepage": null,
    "category": "conversation",
    "last_synced": null
  },
  "components": {
    "agents": [ /* 4 agent configs */ ],
    "models": [ /* 4 model configs */ ],
    "tools": [ /* 6 tool configs */ ],
    "terminations": [ /* 3 termination configs */ ],
    "teams": [ /* 4 team configs */ ],
    "workbenches": [ /* 3 workbench configs */ ]
  }
}

Component Inventory

Agents

Label	Provider	Description	Key Features
AssistantAgent	autogen_agentchat.agents.AssistantAgent	AI assistant with tool abilities	Calculator tool, gpt-4o-mini, unbounded context
Web Surfer Agent	autogen_ext.agents.web_surfer.MultimodalWebSurfer	Web browsing agent	Headless browser, Bing start page, vision capable
Verification Assistant	autogen_agentchat.agents.AssistantAgent	Information verifier and summarizer	Verification-focused system message, no tools
UserProxyAgent	autogen_agentchat.agents.UserProxyAgent	Human user representation	Simple name and description only

Models

Label	Provider	Model	Description
OpenAI GPT-4o Mini	autogen_ext.models.openai.OpenAIChatCompletionClient	gpt-4o-mini	Fast and cost-effective OpenAI model
Mistral-7B Local	autogen_ext.models.openai.OpenAIChatCompletionClient	TheBloke/Mistral-7B-Instruct-v0.2-GGUF	Local model via LMStudio/Ollama
Anthropic Claude-3-7	autogen_ext.models.anthropic.AnthropicChatCompletionClient	claude-3-7-sonnet-20250219	Anthropic's latest reasoning model
AzureOpenAI GPT-4o-mini	autogen_ext.models.openai.AzureOpenAIChatCompletionClient	gpt-4o-mini	Azure-hosted enterprise variant

Tools

Label	Provider	Description	Dependencies
Calculator Tool	autogen_core.tools.FunctionTool	Basic arithmetic (+, -, *, /)	None
Image Generation Tool	autogen_core.tools.FunctionTool	DALL-E 3 image generation	OpenAI, PIL, base64
Fetch Webpage Tool	autogen_core.tools.FunctionTool	Convert webpage to markdown	requests, beautifulsoup4, html2text
Bing Search Tool	autogen_core.tools.FunctionTool	Bing Web Search API	BING_SEARCH_KEY env var
Google Search Tool	autogen_core.tools.FunctionTool	Google Custom Search API	GOOGLE_API_KEY, GOOGLE_CSE_ID env vars
Python Code Execution Tool	autogen_ext.tools.code_execution.PythonCodeExecutionTool	Execute Python code locally	LocalCommandLineCodeExecutor

Teams

Label	Type	Agents	Description
RoundRobin Team	RoundRobinGroupChat	1 assistant	Single agent with calculator tool
Selector Team	SelectorGroupChat	2 (assistant + critic)	Model-based speaker selection with critique loop
Web Agent Team (Operator)	SelectorGroupChat	3 (web surfer + verifier + user proxy)	Web research with human oversight
Deep Research Team	SelectorGroupChat	3 (researcher + verifier + summarizer)	Multi-agent research with Google search

Terminations

Label	Provider	Configuration	Description
TextMentionTermination	autogen_agentchat.conditions.TextMentionTermination	text: "TERMINATE"	Terminate on keyword
MaxMessageTermination	autogen_agentchat.conditions.MaxMessageTermination	max_messages: 10	Terminate after N messages
OR Termination	autogen_agentchat.base.OrTerminationCondition	conditions: [TextMention, MaxMessage]	Terminate on either condition

Workbenches

Label	Provider	Description	Configuration
Basic Tools Workbench	autogen_core.tools.StaticWorkbench	Calculator and webpage fetcher	2 function tools
Web Content Fetch Workbench	autogen_ext.tools.mcp.McpWorkbench	MCP server for web content	StdioServerParams: uvx mcp-server-fetch
HTTP Streamable MCP Workbench	autogen_ext.tools.mcp.McpWorkbench	HTTP-based MCP server example	StreamableHttpServerParams with auth

Usage Examples

Loading the Gallery

import defaultGalleryData from "./views/gallery/default_gallery.json";
import type { Gallery } from "./types/datamodel";

const defaultGallery: Gallery = {
  id: 1,
  config: defaultGalleryData,
  created_at: new Date().toISOString()
};

// Access components
const assistantAgent = defaultGallery.config.components.agents[0];
const calculatorTool = defaultGallery.config.components.tools[0];
const roundRobinTeam = defaultGallery.config.components.teams[0];

Extracting Specific Components

// Find the Web Surfer agent
const webSurfer = defaultGalleryData.components.agents.find(
  agent => agent.label === "Web Surfer Agent"
);

// Find OpenAI models
const openAIModels = defaultGalleryData.components.models.filter(
  model => model.provider.includes("OpenAIChatCompletionClient")
);

// Find tools requiring API keys
const apiTools = defaultGalleryData.components.tools.filter(
  tool => tool.description?.includes("requires") &&
          tool.description?.includes("env variable")
);

Using Team Templates

// Extract the Deep Research Team template
const researchTeam = defaultGalleryData.components.teams.find(
  team => team.label === "Deep Research Team"
);

if (researchTeam && isSelectorTeam(researchTeam)) {
  console.log(`Team has ${researchTeam.config.participants.length} participants`);
  console.log(`Selector prompt: ${researchTeam.config.selector_prompt}`);

  // Extract the research assistant with tools
  const researcher = researchTeam.config.participants[0];
  if (isAssistantAgent(researcher)) {
    const workbench = researcher.config.workbench?.[0];
    if (workbench && isAnyStaticWorkbench(workbench)) {
      console.log(`Researcher has ${workbench.config.tools.length} tools`);
    }
  }
}

Customizing Components

import { cloneDeep } from "lodash";

// Clone and customize the calculator tool
const customCalculator = cloneDeep(
  defaultGalleryData.components.tools.find(t => t.label === "Calculator Tool")
);

if (customCalculator && isFunctionTool(customCalculator)) {
  customCalculator.label = "Advanced Calculator";
  customCalculator.description = "Calculator with additional operations";
  customCalculator.config.source_code = `
    def calculator(a: float, b: float, operator: str) -> str:
        # Add power and modulo operations
        if operator == "**":
            return str(a ** b)
        elif operator == "%":
            return str(a % b)
        # ... existing operations
  `;
}

Building a Custom Team from Gallery Components

import type { Component, TeamConfig } from "./types/datamodel";

// Select components from gallery
const assistant = defaultGalleryData.components.agents[0]; // AssistantAgent
const webSurfer = defaultGalleryData.components.agents[1]; // Web Surfer
const termination = defaultGalleryData.components.terminations[2]; // OR termination
const model = defaultGalleryData.components.models[0]; // GPT-4o Mini

// Build custom team
const customTeam: Component<TeamConfig> = {
  provider: "autogen_agentchat.teams.SelectorGroupChat",
  component_type: "team",
  version: 1,
  label: "Custom Research Team",
  config: {
    participants: [assistant, webSurfer],
    model_client: model,
    termination_condition: termination,
    selector_prompt: "Select the best agent for the current task...",
    allow_repeated_speaker: true,
    max_selector_attempts: 3,
    emit_team_events: false,
    model_client_streaming: false
  }
};

Component Details

Calculator Tool Source Code

def calculator(a: float, b: float, operator: str) -> str:
    try:
        if operator == "+":
            return str(a + b)
        elif operator == "-":
            return str(a - b)
        elif operator == "*":
            return str(a * b)
        elif operator == "/":
            if b == 0:
                return "Error: Division by zero"
            return str(a / b)
        else:
            return "Error: Invalid operator. Please use +, -, *, or /"
    except Exception as e:
        return f"Error: {str(e)}"

Deep Research Team Selector Prompt

You are coordinating a research team by selecting the team member to speak/act next.
The following team member roles are available: {roles}.

The research_assistant performs searches and analyzes information.
The verifier evaluates progress and ensures completeness.
The summary_agent provides a detailed markdown summary of the research as a report to the user.

Given the current context, select the most appropriate next speaker.
Base your selection on:
1. Current stage of research
2. Last speaker's findings or suggestions
3. Need for verification vs need for new information

You should ONLY select the summary_agent role if the research is complete and it is time to generate a report.

Read the following conversation. Then select the next role from {participants} to play.
ONLY RETURN THE ROLE.

{history}

Related Pages

• Studio Gallery Detail - UI component for viewing and editing galleries
• Studio Component Templates - TypeScript template system
• Studio Datamodel Types - Type definitions for gallery structure
• Studio Team Builder - Uses gallery components in team construction
• Studio API Client - Gallery synchronization and management