AdalFlow

// Overview

AdalFlow is an open-source Python library for building and auto-optimizing language model workflows, from chatbots and RAG systems to tool-using agents. It takes a PyTorch-like approach: you assemble your pipeline from model-agnostic building blocks, then let the framework tune the prompts across those components instead of hand-editing them.

It is aimed at developers who are tired of manual prompt engineering and want a structured way to improve accuracy. AdalFlow provides an auto-differentiative framework for both zero-shot and few-shot prompt optimization, so prompts are optimized end-to-end against your task rather than guessed by hand.

Within the prompt-programming and LLM-orchestration space, AdalFlow sits between a raw SDK and a higher-level framework. You can swap the underlying model through configuration, and the library ships agent, runner, and tracing primitives with human-in-the-loop support built in.

// What it does

• Build chatbots, RAG, and agents from model-agnostic building blocks
• Auto-optimize prompts with a unified auto-differentiative framework (zero-shot and few-shot)
• Switch the underlying LLM via configuration without rewriting your pipeline
• Agent and Runner primitives with synchronous, asynchronous, and streaming call modes
• Built-in tool calling, tracing, and human-in-the-loop, with no extra API required
• 100% open-source and lightweight

// Getting started

Install AdalFlow with pip, set your model provider key, then define tools and run an agent.

pip install adalflow

export OPENAI_API_KEY=your-key-here

from adalflow import Agent, Runner
from adalflow.components.model_client.openai_client import OpenAIClient

def calculator(expression: str) -> str:
    """Evaluate a mathematical expression."""
    try:
        result = eval(expression)
        return f"The result of {expression} is {result}"
    except Exception as e:
        return f"Error: {e}"

agent = Agent(
    name="MyAgent",
    tools=[calculator],
    model_client=OpenAIClient(),
    model_kwargs={"model": "gpt-4o", "temperature": 0.3},
    max_steps=5,
)

runner = Runner(agent=agent)
result = runner.call(
    prompt_kwargs={"input_str": "Calculate 15 * 7 + 23"}
)
print(result.answer)

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Building a tool-using agent that can call functions like a calculator or web search and stream its steps
• Replacing manual prompt engineering with automatic prompt optimization for a RAG or classification task
• Prototyping an LLM pipeline that needs to swap between model providers through configuration
• Adding tracing and human-in-the-loop review to an existing LLM workflow

// How AdalFlow compares

AdalFlow alongside other open-source prompt programming tools AI/TLDR tracks, ranked by GitHub stars.