AdalFlow

Normal
GitHub Python MIT

Description

AdalFlow: The library to build & auto-optimize LLM applications.

Key Features

  • 100% open-source Agent SDK with built-in human-in-the-loop and tracing
  • Unified auto-differentiative framework for zero-shot and few-shot prompt optimization (LLM-AutoDiff)
  • Model-agnostic building blocks for RAG, Agents, and classical NLP tasks via config switching
  • PyTorch-like API design for building and auto-optimizing LM workflows from chatbots to agents
  • Sync, async, and streaming execution modes with Runner and step history tracking

Use Cases

πŸ’‘ Building RAG pipelines with automatic prompt tuning and retrieval optimization
πŸ’‘ Creating AI agents with tool-use capabilities and multi-step reasoning
πŸ’‘ Automated few-shot example selection to improve in-context learning accuracy
πŸ’‘ Prototyping and evaluating chatbot workflows across different LLM providers

Quick Start

1. Install: pip install adalflow. 2. Define tools (functions with docstrings), create an Agent with a model client, and wrap it in a Runner. 3. Call runner.call() for sync, runner.acall() for async, or runner.astream() for streaming.

Related Projects