haiku.rag

Overview

Haiku RAG is a comprehensive retrieval-augmented generation system built on LanceDB, Pydantic AI, and Docling. It features hybrid search combining vector and full-text search with multimodal capabilities for text and images. The system includes research agents for complex information tasks, analysis agents with Python code execution, and conversational interfaces. Its MCP server implementation exposes tools for document management, search, QA, and research directly within AI assistants like Claude Desktop.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• add_srcadd_src(source: str)

  Index a document source (file path or URL) for retrieval

  Try:Add research paper.pdf to the knowledge baseTry:Index https://arxiv.org/pdf/1706.03762
• searchsearch(query: str, limit: int = 10)

  Perform hybrid vector and full-text search over indexed documents

  Try:Search for information about attention mechanismsTry:Find documents discussing transformer architectures
• askask(question: str, cite: bool = false)

  Ask questions with citations from the indexed documents

  Try:What were the datasets used for evaluation?Try:Explain the complexity of self-attention with citations
• researchresearch(topic: str)

  Perform iterative research planning, search, and synthesis

  Try:Research limitations of the approachTry:Investigate recent developments in multimodal models
• analyzeanalyze(task: str)

  Perform complex analytical tasks via sandboxed Python code execution

  Try:Count how many documents mention transformersTry:Calculate average sentiment scores across documents
• chatchat(message: str)

  Engage in a multi-turn conversational interaction with session memory

  Try:Let's discuss the implications of these findingsTry:Explain this concept in simpler terms

Note: Tool names inferred from CLI examples in the README; actual MCP tool signatures not documented

Comparable tools

Installation

pip install haiku.rag

haiku-rag serve --mcp --stdio

{
  "mcpServers": {
    "haiku-rag": {
      "command": "haiku-rag",
      "args": ["serve", "--mcp", "--stdio"]
    }
  }
}

FAQ

What document formats are supported?

The system supports PDFs and other formats through Docling, with structure-aware context expansion.

Can I run it locally?

Yes, it's local-first with embedded LanceDB, but also supports S3, GCS, Azure, and LanceDB Cloud.

On Hacker News

Recent discussion from the developer community.

• Retrieval Augmented Generation Based on SQLite▲ 88💬 10
  Story by emzo · 2025-06-24

Compare haiku.rag with