Funcchain provides a Pythonic framework for building cognitive systems by integrating Large Language Models (LLMs) with Pydantic models for structured output. It targets developers looking to seamlessly embed LLM capabilities into their applications, offering a simplified syntax that compiles into Langchain runnables for efficient pipeline execution.

How It Works

Funcchain leverages Pydantic models as output schemas, enabling LLMs to generate data in a structured, type-safe format. It supports OpenAI Functions and LlamaCpp grammars for efficient structured output generation. The library compiles its custom syntax into Langchain Expression Language (LCEL) runnables, allowing for easy invocation, streaming, and batch processing of LLM-driven workflows.

Quick Start & Requirements

• Install via pip: pip install funcchain
• Requires Python 3.7+
• OpenAI API key or Ollama setup for local models.
• Documentation: https://funcchain.ai/docs
• Examples: https://github.com/shroominic/funcchain/tree/main/examples

Highlighted Details

• Pythonic syntax with native Python features for prompt definition.
• Supports dynamic output types, including nested Pydantic models and union types.
• Integrates vision model capabilities for image analysis.
• Seamlessly supports local LLMs via Ollama and GGUF downloads from Hugging Face.
• Offers Langsmith support, auto-retries, and composability with Langchain LCEL.

Maintenance & Community

• Active development with contributions from a community.
• Development setup script available: ./dev_setup.sh
• Contributing Guide: https://github.com/shroominic/funcchain/blob/main/CONTRIBUTING.md

Licensing & Compatibility

• MIT License.
• Permissive license suitable for commercial use and integration into closed-source applications.

Limitations & Caveats

The library relies heavily on the LLM's ability to adhere to the specified output schema; complex or ambiguous schemas may lead to parsing errors or unexpected outputs. While it supports local models, performance and reliability will vary based on the chosen model and hardware.