Skald provides a production-ready Retrieval Augmented Generation (RAG) system accessible via a plug-and-play API, designed for developers and power users needing to integrate RAG capabilities into their infrastructure. It offers a flexible, configurable engine that allows fine-tuning of various components, enabling users to ship, configure, and evaluate RAG performance efficiently.

How It Works

Skald implements a RAG architecture with a focus on production readiness and customizability. It offers a core API for chat and semantic search functionalities, allowing users to interact with their knowledge bases. The system is designed to be highly configurable, enabling adjustments to vector search parameters, reranking models, LLMs, query rewriting, and chunking strategies (with chunking planned for future release). This approach allows for tuning the RAG engine to specific needs beyond solid defaults, facilitating performance evaluation directly within the platform.

Quick Start & Requirements

• Primary install/run command:

  git clone https://github.com/skaldlabs/skald
  cd skald
  echo "OPENAI_API_KEY=<your_key>" > .env
  docker-compose up
  

• Non-default prerequisites: Requires an OpenAI API key (or self-hosted LLM/embeddings for advanced usage).
• Estimated setup time: Self-hosted deploy can be live in less than an hour.
• Links: Docs, Website, Demo video, Slack.

Highlighted Details

• Production-ready RAG via plug-and-play API.
• Turnkey configuration with tunable parameters (vector search, reranking, models, query rewriting).
• Built-in evaluation tools for RAG performance.
• SDKs available for major programming languages (Node.js, Python, Ruby, Go, PHP, C#, MCP).
• Option to run with no third-party dependencies by hosting own LLM inference and embeddings.

Maintenance & Community

The project actively encourages contributions and provides a Slack community for users to ask questions and engage. A list of contributors is maintained.

Licensing & Compatibility

Licensed under the MIT license, permitting commercial use and integration into closed-source projects.

Limitations & Caveats

Advanced usage, such as running Skald without third-party dependencies (e.g., OpenAI), requires significant effort in hosting custom LLM inference servers and local embeddings. The chunking feature is marked as "soon" and not yet available.