AdalFlow

What it is

AdalFlow is a PyTorch-like library to build and auto-optimize any LM workflows, from Chatbots, RAG, to Agents.

How it works

The project offers trainable task pipelines defined via Parameter objects passed to a Generator, enabling prompt tuning through textual gradients and few-shot demonstrations. It provides components such as AdalComponent and Trainer to manage training, evaluation, and gradient-based optimization of prompts and demonstrations.

Getting started

Install AdalFlow with pip:

pip install adalflow

The README includes a Hello World Agent Example showing how to define tools (calculator, web_search, counter), construct an Agent with a model client, and run synchronous, asynchronous, and streaming modes. The example imports and uses OpenAIClient from adalflow.components.model_client.openai_client and various RunItem types from adalflow.core.types.

from adalflow import Agent, Runner
from adalflow.components.model_client.openai_client import OpenAIClient
from adalflow.core.types import (
    ToolCallActivityRunItem, 
    RunItemStreamEvent,
    ToolCallRunItem,
    ToolOutputRunItem,
    FinalOutputItem
)

1. Synchronous Call Mode

result = runner.call(
    prompt_kwargs={"input_str": "Calculate 15 * 7 + 23 and count to 5"}
)

2. Asynchronous Call Mode

result = await runner.acall(
    prompt_kwargs={"input_str": "What's the weather in SF and calculate 42 * 3"}
)

3. Async Streaming Mode

streaming_result = runner.astream(
    prompt_kwargs={"input_str": "Calculate 100 + 50 and count to 3"},
)

Set your OPENAI_API_KEY environment variable to run these examples.

Recent releases

Latest 5 releases listed with dates:

• v1.1.3 Release v1.1.3 (2025-09-25): AdalFlow v1.1.3
• v1.1.2 Release v1.1.2 (2025-08-16): AdalFlow v1.1.2
• v1.1.1 Release v1.1.1 (2025-08-10): AdalFlow v1.1.1
• v1.0.5a1 v1.0.5a1 Release (2025-07-17): Add agent, runner, mcp tools and modify generator output
• v0.1.0.beta.1 v0.1.0.beta.1 (2024-07-15): Follow the changelog.md

Traction

Stars: 4189 Forks: 383 Open issues: 68

Behind the repo

The project is linked to SylphAI-Inc and collaborates with institutions such as the VITA Group at UT Austin; no external startup/company page is provided in the data.

Caveats

License: MIT Created: 2024-04-19 Last push: 2026-05-29 Language: Python

Notes:

• The repository describes a unified framework for auto-optimization of LM workflows, including prompt tuning (PROMPT) and few-shot demonstrations (DEMO).
• It presents documentation and Colab quickstart links, and references model providers via a generic model_client interface.