deepsense-ai/ragbits

Building blocks for rapid development of GenAI applications

What it solves

Ragbits provides a modular set of building blocks for the rapid development of Generative AI applications. It simplifies the process of building reliable, scalable GenAI apps by providing standardized tools for LLM interaction, data ingestion, retrieval-augmented generation (RAG), and multi-agent coordination.

How it works

Ragbits is organized into several specialized packages that developers can use together or individually:

  • Core: Handles prompts, LLM calls (via LiteLLM for 100+ models), and vector database connections (e.g., Qdrant, PgVector).
  • Document Search: Manages ingestion pipelines for 20+ file formats (using Docling or Unstructured) and retrieval processes.
  • Agents: Provides abstractions for building agentic systems, including role-based collaboration via the A2A protocol and real-time data integration using the Model Context Protocol (MCP).
  • Chat: Offers full-stack infrastructure, including a Chat UI and API for deploying conversational AI.
  • Evaluate & Guardrails: Tools for validating prompts (via promptfoo) and ensuring response safety.

Who it’s for

Developers building GenAI applications who need a flexible, type-safe framework that allows them to easily swap LLMs, scale data ingestion, and deploy multi-agent workflows without building everything from scratch.

Highlights

  • Model Flexibility: Swap between 100+ LLMs or use local models via LiteLLM.
  • Type Safety: Uses Python generics to enforce strict type safety in model interactions.
  • Scalable Ingestion: Supports Ray-based parallel processing for large datasets.
  • Multi-Agent Coordination: Role-based collaboration using the A2A protocol.
  • Observability: Real-time tracking via OpenTelemetry and CLI insights.
  • Comprehensive Tooling: Includes a CLI for testing prompts and managing vector stores, as well as a built-in Chat UI.