chainlit-crew-adapter

chainlit-crew-adapter is a small library for running CrewAI crews inside Chainlit with a cleaner developer experience.

It focuses on two things:

• rendering CrewAI execution as Chainlit steps
• letting CrewAI agents ask the user follow-up questions with a reusable Chainlit-backed tool

Demo

Features

• ChainlitCrewAdapter with async kickoff(...)
• step rendering for crew, task, agent, and tool events
• optional show_agent_steps=False for cleaner traces
• ChainlitAskUserTool for human-in-the-loop follow-up questions
• examples for both step rendering and ask-user workflows

Usage

Installation from PyPI

pip install chainlit-crew-adapter

Setup

Before running the examples, set your OpenAI API credentials:

export OPENAI_API_KEY=your_openai_api_key
export OPENAI_MODEL=gpt-4o-mini  # optional, defaults to gpt-4o-mini

Example 1: Step Rendering

Render CrewAI crew execution as Chainlit steps to see the full task and tool execution trace.

import chainlit as cl
from chainlit_crew_adapter import ChainlitCrewAdapter
from your_crew import build_my_crew

@cl.on_message
async def main(message: cl.Message) -> None:
    crew = build_my_crew()
    adapter = ChainlitCrewAdapter(cl=cl, crew=crew)
    result = await adapter.kickoff(inputs={"user_request": message.content})
    await cl.Message(content=str(result)).send()

For a complete example, see examples/steps_demo.py.

Example 2: Human Follow-up Questions

Enable the crew to ask users follow-up questions using the ChainlitAskUserTool.

import chainlit as cl
from chainlit_crew_adapter import ChainlitAskUserTool, ChainlitCrewAdapter
from your_crew import build_ask_user_crew

@cl.on_message
async def main(message: cl.Message) -> None:
    ask_user_tool = ChainlitAskUserTool(
        timeout_seconds=180,
        author="Crew Assistant"
    )
    crew = build_ask_user_crew(ask_user_tool=ask_user_tool)
    adapter = ChainlitCrewAdapter(
        cl=cl,
        crew=crew,
        show_agent_steps=False  # Cleaner trace without agent-level steps
    )
    result = await adapter.kickoff(inputs={"user_request": message.content})
    await cl.Message(content=str(result)).send()

For a complete example, see examples/ask_user_demo.py.

Running the Examples

# Step rendering demo
chainlit run examples/steps_demo.py

# Human follow-up questions demo
chainlit run examples/ask_user_demo.py

Installation

Local development

pip install -e .

Or, if you prefer the repo requirements file:

pip install -r requirements.txt

Environment variables

At minimum, set:

OPENAI_API_KEY=your_openai_api_key
OPENAI_MODEL=gpt-4o-mini

OPENAI_MODEL is optional. The examples default to gpt-4o-mini.

API

ChainlitCrewAdapter

ChainlitCrewAdapter(
    cl: SupportsChainlit,
    crew: Crew,
    inputs: Mapping[str, object] | None = None,
    *,
    show_agent_steps: bool = True,
)

await adapter.kickoff(inputs=None)

Runs crew.kickoff(...) through Chainlit and mirrors CrewAI progress into the Chainlit step tree.

ChainlitAskUserTool

ChainlitAskUserTool(
    *,
    ask_user=None,
    chainlit_context=None,
    timeout_seconds: int = 120,
    author: str = "Crew Assistant",
    fail_on_timeout: bool = False,
)

Use this tool in CrewAI agents when they need to ask the user follow-up questions during execution.

Example Apps

This repo includes two example apps.

1. Step rendering demo

File:

examples/steps_demo.py

Run it with:

chainlit run examples/steps_demo.py

Try prompts like:

Analyze post 5

Summarize post 12

What it demonstrates:

• crew step rendering
• task, agent, and tool steps
• a real tool-driven CrewAI workflow with JSONPlaceholder

2. Ask-user demo

File:

examples/ask_user_demo.py

Run it with:

chainlit run examples/ask_user_demo.py

Try prompts like:

Help me plan a product launch

I need a landing page brief

What it demonstrates:

• ChainlitAskUserTool
• chat-based follow-up questions during crew execution
• a compact step trace using show_agent_steps=False
• a dedicated clarification step summarizing the questions asked in chat

Legacy alias

For backward compatibility, this file points to the ask-user demo:

examples/manual_crew_integration.py

Example Structure

chainlit_crew_adapter/
├── chainlit_crew_adapter/
│   ├── __init__.py
│   ├── adapter.py
│   ├── events.py
│   ├── human_input.py
│   └── step_renderer.py
├── examples/
│   ├── ask_user_demo.py
│   ├── manual_crew_integration.py
│   ├── steps_demo.py
│   └── crews/
│       ├── ask_user_demo/
│       └── steps_demo/
├── pyproject.toml
└── README.md

Notes

• CrewAI streaming crews are not supported yet in the adapter step renderer.
• The ask-user demo intentionally keeps the back-and-forth in chat and the execution trace in steps.
• The step renderer treats the ask-user tool as a special clarification flow to avoid noisy duplicate traces.

Development Checks

A quick local verification pass:

python -m py_compile chainlit_crew_adapter/*.py examples/*.py examples/crews/*/*.py

Next Step Toward Deployment

The repo now has a basic pyproject.toml, package data marker, examples, and a cleaner separation between library code and demos. The next natural step is publishing metadata cleanup and then building the distribution with:

python -m build