Repolode
Submit
SylphAI-Inc

SylphAI-Inc/AdalFlow

A PyTorch-style Python library for building LLM workflows such as chatbots, RAG, and agents, with automatic prompt optimization.

View on GitHubWebsite
Stars
4,224
Forks
389
Open issues
72
Last commit
6 days ago

What is it?

AdalFlow is a Python library for building LLM applications, covering chatbots, retrieval-augmented generation (RAG), and agents. Its API follows PyTorch conventions: pipelines are composed from components, and model providers can be swapped through configuration rather than code changes.

The library includes an optimization layer that treats prompts and few-shot demonstrations as trainable parameters. Developers define a Parameter and pass it to a Generator, then train it with textual gradient descent for instructions or bootstrap few-shot optimization for demonstrations. A Trainer with an AdalComponent interface handles the training and validation loop.

AdalFlow also provides an agent runner with synchronous, asynchronous, and streaming call modes, tool calling, and step history. Human-in-the-loop and tracing are covered in its documentation, and it integrates with MLflow for tracing.

Key features

  • 01PyTorch-style components for composing LLM pipelines.
  • 02Automatic prompt optimization using textual gradients and few-shot demonstrations.
  • 03Model-agnostic configuration to switch between LLM providers.
  • 04Agent runner with sync, async, and streaming call modes plus tool calling.
  • 05MLflow integration for tracing.

Use cases

  • Building RAG question-answering systems over your own documents.
  • Creating tool-using agents that call functions and stream progress.
  • Automatically tuning prompts for a classification task instead of hand-editing them.
  • Prototyping LLM pipelines that can later switch model providers via config.

Quick start

  1. 1

    Install AdalFlow with pip

    pip install adalflow
  2. 2

    Set your OPENAI_API_KEY environment variable

  3. 3

    Run the Hello World Agent example from the README

Things to know

  • The Hello World examples need an OpenAI API key to run.

In collections

Frameworks for building apps and websites

Topics

#llm#rag#agent#prompt-optimization#python#pytorch#retrieval#framework#ai#auto-prompting

FAQ

What is AdalFlow?
A PyTorch-style Python library for building LLM workflows such as chatbots, RAG, and agents, with automatic prompt optimization. AdalFlow is a Python library for building LLM applications, covering chatbots, retrieval-augmented generation (RAG), and agents. Its API follows PyTorch conventions: pipelines are composed from components, and model providers can be swapped through configuration rather than code changes.
Is AdalFlow free to use? What license does it have?
AdalFlow is open source under the MIT license. Check the license terms before using it in commercial products.
Is AdalFlow still maintained?
Actively maintained. The last commit to AdalFlow was on Nov 1, 2026.
What can I use AdalFlow for?
Building RAG question-answering systems over your own documents. Creating tool-using agents that call functions and stream progress. Automatically tuning prompts for a classification task instead of hand-editing them. Prototyping LLM pipelines that can later switch model providers via config.
How do I install AdalFlow?
Install AdalFlow with pip: pip install adalflow Set your OPENAI_API_KEY environment variable Run the Hello World Agent example from the README
Where can I get AdalFlow?
The source code and installation instructions are on GitHub: https://github.com/SylphAI-Inc/AdalFlow

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects