Akcio provides a framework for building Retrieval Augmented Generation (RAG) chatbots, enhancing LLM responses with custom knowledge bases. It targets developers and researchers looking to create more accurate and context-aware conversational AI, offering a flexible CVP (ChatGPT + Vector DB + Prompt-as-code) stack.

How It Works

Akcio implements the CVP stack by first retrieving relevant information from a knowledge base via semantic search or keyword matching. This retrieved context, along with the user's query, is then fed into a Large Language Model (LLM) to generate a more informed and tailored response. This approach mitigates LLM hallucinations and improves response quality by grounding them in specific data.

Quick Start & Requirements

• Install via pip install -r requirements.txt.
• Requires Python 3.x.
• Supports various LLMs (OpenAI, Llama-2, Dolly, Ernie, etc.) and vector stores (Milvus, Zilliz Cloud).
• Configuration is managed via config.py or environment variables.
• Start service with python main.py --towhee or python main.py --langchain.
• Documentation: https://github.com/zilliztech/akcio/blob/main/docs/README.md

Highlighted Details

• Offers two primary integration platforms: Towhee (for simplified pipeline-based development) and LangChain (for agent-based, more complex orchestration).
• Extensive support for diverse LLMs, embedding models, vector stores, scalar stores (Elasticsearch), and memory stores (SQLAlchemy compatible).
• Provides both offline data loading tools and an online API endpoint (/project/add) for data ingestion.
• Includes a FastAPI backend with Swagger UI for API interaction and a Gradio demo option.

Maintenance & Community

• Project by Zilliz Tech.
• Community contact and roadmap information are available via links in the README.

Licensing & Compatibility

• Licensed under the Server Side Public License (SSPL) v1.
• SSPL v1 has specific requirements regarding offering the software as a service, which may impact commercial use and closed-source linking.

Limitations & Caveats

The online data loading method is not recommended for large datasets. The SSPL v1 license may impose restrictions on certain commercial deployment scenarios.