Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.praison.ai/llms.txt

Use this file to discover all available pages before exploring further.

PraisonAI supports three AutoGen-family frameworks: AutoGen v0.2, AutoGen v0.4, and AG2. Each provides multi-agent collaboration with different implementations and capabilities.
Need a framework that isn’t listed here? See Framework Adapter Plugins to register your own via Python entry points.

​
Which AutoGen Option to Choose

PraisonAI supports three AutoGen-family backends with different capabilities:
FrameworkInstallBest For
framework: ag2pip install "praisonai[ag2]"New projects, AWS Bedrock, latest features
framework: autogen_v4pip install "praisonai[autogen-v4]"AutoGen v0.4 stack (autogen-agentchat + autogen-ext)
framework: autogenpip install "praisonai[autogen]"Legacy AutoGen v0.2 projects

​
AutoGen v0.4

AutoGen v0.4 uses the new autogen-agentchat and autogen-ext packages with OpenAIChatCompletionClient and RoundRobinGroupChat.
1

Install

pip install "praisonai[autogen-v4]"
2

Create YAML file

framework: autogen_v4
topic: Research the latest developments in AI agents

roles:
  researcher:
    role: AI Research Specialist
    goal: Find and summarize recent AI agent developments
    backstory: Expert researcher with deep knowledge of AI trends.
    tasks:
      research_task:
        description: Research the latest developments in {topic}
        expected_output: Summary report with key findings
3

Run

export OPENAI_API_KEY=your-key
praisonai agents.yaml --framework autogen_v4

​
How AutoGen v0.4 Works

​
AutoGen v0.4 Features

  • Model Configuration: Reads llm_config[0] for model, api_key, base_url (defaults: gpt-4o-mini, OPENAI_API_KEY, https://api.openai.com/v1)
  • Agent Creation: One AssistantAgent per YAML role with reflect_on_tool_use=True
  • Agent Name Sanitization: Non-alphanumeric characters become _, leading digits get agent_ prefix, Python keywords get trailing _
  • Tool Registration: Tools must expose .run() method, passed to AssistantAgent(tools=...)
  • Execution: RoundRobinGroupChat with TextMentionTermination("TERMINATE") | MaxMessageTermination(max_messages=20)
  • Turn Limit: max_turns = len(agents) * 3 (currently hardcoded)
Agent names are auto-sanitized: "Research Analyst!" becomes "Research_Analyst_". This ensures valid Python identifiers for AutoGen v0.4 internal usage.

​
AG2 (Full)

AG2 is the community fork of AutoGen with enhanced features and AWS Bedrock support.
1

Install

pip install "praisonai[ag2]"
2

Create YAML file

framework: ag2
topic: Write a technical analysis report

roles:
  analyst:
    role: Technical Analyst
    goal: Analyze market data and trends
    backstory: Expert in financial analysis with 10 years experience.
    tasks:
      analysis_task:
        description: Analyze the technical indicators for {topic}
        expected_output: Detailed technical analysis with charts and recommendations
3

Run

export OPENAI_API_KEY=your-key
praisonai agents.yaml --framework ag2

​
How AG2 Works

​
AG2 Features

  • LLM Configuration: Resolution order: top-level llm.* > first role’s llm.* > llm_config[0] > environment > defaults
  • AWS Bedrock: api_type: bedrock skips api_key, uses AWS SDK credential chain
  • Tool Registration: Uses assistant.register_for_llm() + user_proxy.register_for_execution()
  • Agent Names: Sanitized with [^a-zA-Z0-9_\-] → _
  • GroupChat: Standard GroupChat(max_round=12) with GroupChatManager
  • Result Extraction: Prefers chat_result.summary, falls back to scanning messages in reverse

​
YAML llm Block Precedence

For AG2, the llm: configuration can be placed at the top level or under the first role:
LocationPriorityExample
Top-levelHighestllm: { model: "gpt-4o", api_key: "..." }
First roleMediumroles: { analyst: { llm: { ... } } }
DispatcherLowestllm_config[0] from command

​
AWS Bedrock with AG2

framework: ag2
topic: Cloud architecture design

roles:
  cloud_architect:
    role: AWS Solutions Architect
    goal: Design scalable cloud architectures
    backstory: Expert in AWS services and serverless design.
    llm:
      model: bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0
      api_type: bedrock
      aws_region: us-east-1
    tasks:
      design_task:
        description: Design a serverless architecture for {topic}
        expected_output: Architecture diagram with component descriptions
With api_type: bedrock, no api_key is required. AG2 uses standard AWS credentials from ~/.aws/credentials, environment variables, or IAM roles.

​
AutoGen v0.2 (Legacy)

Original AutoGen v0.2 support for existing projects.
1

Install

pip install "praisonai[autogen]"
2

Create YAML file

framework: autogen
topic: Create a movie script about cats on Mars

roles:
  researcher:
    role: Research Analyst
    goal: Gather information about Mars and cat behavior
    backstory: Skilled researcher with focus on accurate information.
    tasks:
      research_task:
        description: Research Mars environment and cat behavior for {topic}
        expected_output: Research findings document with key facts
3

Run

export OPENAI_API_KEY=your-key
praisonai agents.yaml --framework autogen

​
AutoGen v0.2 Features

  • Simple Architecture: UserProxyAgent + AssistantAgent with initiate_chats
  • Configuration: Uses config_list format for LLM configuration
  • Code Execution: Built-in code_execution_config with working directory
  • Termination: Checks for “terminate” or “TERMINATE” in message content
  • Tool Support: Limited compared to v0.4 and AG2

​
Troubleshooting

​
Installation Errors

  • AutoGen v0.4: pip install "praisonai[autogen-v4]" — requires both autogen-agentchat and autogen-ext
  • AG2: pip install "praisonai[ag2]" — requires AG2 distribution
  • AutoGen v0.2: pip install "praisonai[autogen]" — requires legacy autogen package

​
Import Errors

  • v0.4 missing autogen_agentchat: Install autogen-agentchat package
  • v0.4 missing autogen_ext: Install autogen-ext package
  • AG2 not detected: Ensure both ag2 distribution and autogen namespace are available

​
Agent Name Issues

  • AG2: Research Analyst! → Research_Analyst_ (sanitization is intentional)
  • v0.4: Names become valid Python identifiers (e.g., "123Agent" → "agent_123Agent")

​
Bedrock Configuration

  • Missing AWS credentials: api_type: bedrock requires standard AWS credential setup
  • Wrong region: Ensure aws_region matches your Bedrock model availability

​
Message Generation

  • v0.4 “No messages generated”: MaxMessageTermination(20) was hit; consider the hardcoded max_turns = len(agents) * 3 limitation
Not-yet-implemented adapters:
  • framework: autogen_v4 raises NotImplementedError("AutoGen v0.4 adapter is not yet fully implemented. Please use 'autogen' framework for AutoGen v0.2 support.")
  • framework: ag2 (the placeholder AG2Adapter, distinct from the working ag2 backend) raises NotImplementedError("AG2 adapter is not yet fully implemented. Please use 'autogen' framework for AutoGen/AG2 support.")
Use framework: autogen (v0.2) or the fully-registered framework: ag2 backend instead. Plugin authors: Framework adapters now accept dispatcher kwargs (tools_dict, agent_callback, task_callback, cli_config). See Framework Adapter Plugins for custom adapter development. If the framework is not installed, PraisonAI now fails fast at CLI entry with: 
Framework 'autogen' was requested but is not installed.
Install it with:
    pip install 'praisonai[autogen]'  # or: pip install pyautogen
The error appears immediately, before YAML parsing — so a typo in --framework is caught before any expensive setup runs.

​
Best Practices

Use AG2 for new projects with Bedrock needs, AutoGen v0.4 for existing v0.4 stacks, and AutoGen v0.2 only for legacy compatibility.
Both AG2 and v0.4 sanitize agent names differently. Test your role names to understand the final agent identifiers.
AG2 supports top-level llm: blocks for shared configuration. AutoGen v0.4 uses llm_config[0] from the dispatcher.
AutoGen v0.4 requires tools with .run() methods. AG2 uses register_for_llm/register_for_execution pattern.

CrewAI

CrewAI framework integration

PraisonAI Agents

PraisonAI native agents framework