<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> OneTwo is a Python library for developing and deploying LLM agents in production. It simplifies the creation of complex LLM applications by enabling users to define custom agents and integrate them with a rich set of pre-built components, abstracting away underlying complexities.

How It Works

The library's core design prioritizes flexibility and ease of use. Its model-agnostic API allows code to run on any LLM backend without modification. Prompting supports formatted strings, Jinja, or modular function calls. OneTwo ensures reproducible experiments via automatic result caching, saving resources and enabling instant replay. It integrates external tools and offers a secure sandbox for executing model-generated Python code. Stateful agents are composed using Python control flow, with OneTwo managing asynchronous execution, batching, and caching.

Quick Start & Requirements

• Primary install / run command: pip install git+https://github.com/google-deepmind/onetwo
• Non-default prerequisites and dependencies: A Python virtual environment is recommended. Requires API keys for LLM backends (e.g., Google Gemini API).
• Links: Official tutorials for getting started, agent building, and RAG are available.

Highlighted Details

• Model-Agnosticism: Uniform API for seamless LLM backend swapping.
• Agent Composition: Build and compose stateful, autonomous agents using Python control flow for complex actions.
• Tool Integration: Seamlessly integrate external tools and securely execute model-generated Python code.
• Reproducibility & Efficiency: Automatic caching for reproducible experiments, quota saving, and efficient handling of async execution, batching, and caching.
• Retrieval Augmented Generation (RAG): Streamlined integration for grounding LLM responses in custom data.

Maintenance & Community

Associated with Google DeepMind. Authors include Olivier Bousquet, Parth Kothari, Nathan Scales, Nathanael Schärli, and Ilya Tolstikhin. No specific community channels or roadmap links are provided in the README.

Licensing & Compatibility

Licensed under the Apache License, Version 2.0. This permissive license generally allows commercial use and integration into proprietary software.

Limitations & Caveats

Explicitly stated as "not an official Google product." Installation via git+... may indicate early development. Version 0.4.0 suggests the library is in an early stage. Usage requires external API keys for LLM services.