LangGraph with PraisonAI

Run LangGraph-based agent graphs through PraisonAI's YAML and CLI

LangGraph integration lets you run LangGraph state graphs through PraisonAI’s YAML configuration and CLI — no adapter code required.

Quick Start

1. Install

   Both extras are required for end-to-end use:

   pip install "praisonai[langgraph]"
   pip install "praisonai-frameworks[langgraph]"

2. Create agents.yaml

   framework: langgraph
   topic: Simple Question Answer
   
   roles:
     researcher:
       role: Helper
       goal: Answer simple questions accurately
       backstory: I am a helpful assistant
       tasks:
         answer:
           description: What is the capital of France? Reply with just the city name.
           expected_output: Paris

3. Run

   export OPENAI_API_KEY=<your-key>
   praisonai agents.yaml --framework langgraph

How LangGraph Works

YAML Format for LangGraph

LangGraph requires the roles format (not the steps workflow format):

framework: langgraph
topic: Planet facts

roles:
  researcher:
    role: Research_Analyst
    goal: Gather concise facts
    backstory: Expert researcher
    tasks:
      research:
        description: Name one planet in our solar system. Reply with only the planet name.
        expected_output: A planet name
      summarise:
        description: Write one short sentence describing that planet.
        expected_output: One sentence
        context:
          - research

Direct Adapter Use

Advanced — most users should use the CLI / YAML flow above.

Call the adapter directly without the CLI or YAML loader:

from praisonai_frameworks.langgraph.adapter import LangGraphAdapter

config = {
    "framework": "langgraph",
    "topic": "Quick test",
    "roles": {
        "helper": {
            "role": "Assistant",
            "goal": "Answer briefly",
            "backstory": "Helpful assistant",
            "tasks": {
                "answer": {
                    "description": "Reply with exactly the word OK.",
                    "expected_output": "OK",
                }
            },
        }
    },
}
llm_config = [{"model": "gpt-4o-mini", "api_key": "<OPENAI_API_KEY>"}]
result = LangGraphAdapter().run(config, llm_config, "Quick test", tools_dict={})
# result starts with "### LangGraph Output ###"

Verify Installation

Use praisonai doctor to confirm LangGraph is detected:

$ praisonai doctor
✓ Runtime 'langgraph' available
  capabilities: agent_creation, tool_execution, sequential_execution
  supports_handoff: False
  supports_tool_loop: True

Probe programmatically via _framework_availability:

from praisonai._framework_availability import is_available

if is_available("langgraph"):
    print("LangGraph is installed and importable")

Pip Extras Reference

Troubleshooting

framework='langgraph' is not a valid choice — Pre-#2415 PraisonAI versions hardcode choices=["praisonai","crewai","autogen"]. Upgrade to a version that includes dynamic registry-driven CLI choices, or omit --framework and rely on framework: langgraph in your YAML instead.

Framework 'langgraph' was requested but is not installed — Install the extras listed above. The CLI emits an install hint from get_install_hint("langgraph") pointing to praisonai-frameworks[langgraph].

tool_retry_policy validation error on praisonai.run() — Fixed in PraisonAI #2495 by normalising the RetryPolicy object into a dict before agent-config merge. If you are on a version before #2495, pass tool_retry_policy as a plain dict:

tool_retry_policy = {"max_attempts": 3, "delay": 1.0, "backoff_factor": 2.0, "max_delay": 10.0}

Best Practices

Related