haiku.rag

Local-first agentic RAG with citations - hybrid search, reranking, multimodal document retrieval

io.github.ggozad/haiku-rag · Repository · Website · version 0.110.0 · 655 stars · listed from registry

Install

The sweep installs into a clean prefix, but this is the command a user would run:

uvx haiku-rag

Results by platform

Linux

needs configuration2026-11-09 · probed over pypi

6 recorded runs, oldest first.

handshake failed: MCP error -32000: Connection closed
transport: pected token '│', "│ doctor  "... is not valid JSON
Unexpected token '│', "│ history "... is not valid JSON
Unexpected token '│', "│ download"... is not valid JSON
Unexpected token '│', "│ inspect "... is not valid JSON
Unexpected token '│', "│ chat    "... is not valid JSON
Unexpected token '│', "│ mcp     "... is not valid JSON
Unexpected token '│', "│         "... is not valid JSON
Unexpected token '│', "│ tag     "... is not valid JSON
Unexpected token '╰', "╰─────────"... is not valid JSON
--- stderr (tail) ---
Downloading pyarrow (46.6MiB)
Downloading nvidia-cuda-nvrtc (44.8MiB)
Downloading shapely (2.2MiB)
Downloading zstandard (5.3MiB)
Downloading nvidia-nvshmem-cu13 (129.1MiB)
Downloading nvidia-cufile (1.3MiB)
Downloading torchvision (7.2MiB)
Downloading nvidia-cuda-cccl (3.9MiB)
Downloading fastavro (3.4MiB)
Downloading torch (514.2MiB)
Downloading transformers (12.4MiB)
Downloading nvidia-cuda-cupti (11.4MiB)
Downloading docling-slim (1.1MiB)
Downloading nvidia-nccl-cu13 (206.0MiB)
Downloading sympy (6.0MiB)
Downloading faker (2.0MiB)
Downloading scikit-learn (8.7MiB)
Downloading nvidia-cusparse (144.0MiB)
Downloading nvidia-cufft (208.1MiB)
Downloading pydantic-ai-slim (2.7MiB)
Downloading nvidia-cuda-runtime (2.2MiB)
Downloading triton (272.4MiB)
Downloading onnxruntime (22.7MiB)
Downloading nvidia-cusparselt-cu13 (162.3MiB)
Downloading nvidia-cublas (386.1MiB)
Downloading nvidia-cudnn-cu13 (495.1MiB)
Downloading rapidocr (26.0MiB)
Downloading pylance (81.4MiB)
Downloading cuda-bindings (8.1MiB)
Downloading docling-parse (10.3MiB)
Downloading opencv-python-headless (59.1MiB)
Downloading nvidia-nvjitlink (44.5MiB)
Downloading nvidia-cusolver (217.4MiB)
Downloading pydantic-monty-client (4.2MiB)
Downloading pydantic-monty-runtime (14.0MiB)
 Downloaded docling-slim
 Downloaded nvidia-cufile
 Downloaded faker
 Downloaded shapely
 Downloaded nvidia-cuda-runtime
 Downloaded pydantic-ai-slim
 Downloaded fastavro
 Downloaded nvidia-cuda-cccl
 Downloaded pydantic-monty-client
 Downloaded zstandard
 Downloaded sympy
 Downloaded torchvision
 Downloaded cuda-bindings
 Downloaded scikit-learn
 Downloaded docling-parse
 Downloaded nvidia-cuda-cupti
 Downloaded transformers
 Downloaded pydantic-monty-runtime
 Downloaded onnxruntime
 Downloaded rapidocr
 Downloaded nvidia-nvjitlink
 Downloaded nvidia-cuda-nvrtc
 Downloaded pyarrow
 Downloaded opencv-python-headless
 Downloaded pylance
 Downloaded nvidia-nvshmem-cu13
 Downloaded nvidia-cusparse
 Downloaded nvidia-cusparselt-cu13
 Downloaded nvidia-cufft
 Downloaded nvidia-nccl-cu13
 Downloaded nvidia-cusolver
 Downloaded nvidia-cublas
 Downloaded triton
 Downloaded nvidia-cudnn-cu13
 Downloaded torch
Installed 235 packages in 5.43s

macOS

needs configuration2026-11-09 · probed over pypi

6 recorded runs, oldest first.

handshake failed: MCP error -32000: Connection closed
--- stderr (tail) ---
error: No solution found when resolving tool dependencies
  cause: Because onnxruntime>=1.24.1 has no wheels with a matching Python ABI tag (e.g., `cp315`) and haiku-rag-slim[docling]>=0.110.0 depends on onnxruntime>=1.24.1, we can conclude that haiku-rag-slim[docling]>=0.110.0 cannot be used.
         And because haiku-rag>=0.110.0 depends on haiku-rag-slim[docling]==0.110.0 and you require haiku-rag==0.110.0, we can conclude that your requirements are unsatisfiable.

hint: You require CPython 3.15 (`cp315`), but we only found wheels for `onnxruntime` (v1.31.1) with the following Python ABI tags: `cp311`, `cp312`, `cp313`, `cp313t`, `cp314`, `cp314t`

Windows

needs configuration2026-11-09 · probed over pypi

6 recorded runs, oldest first.

handshake failed: MCP error -32000: Connection closed
transport: pected token '|', "| doctor  "... is not valid JSON
Unexpected token '|', "| history "... is not valid JSON
Unexpected token '|', "| download"... is not valid JSON
Unexpected token '|', "| inspect "... is not valid JSON
Unexpected token '|', "| chat    "... is not valid JSON
Unexpected token '|', "| mcp     "... is not valid JSON
Unexpected token '|', "|         "... is not valid JSON
Unexpected token '|', "| tag     "... is not valid JSON
Unexpected token '+', "+---------"... is not valid JSON
--- stderr (tail) ---
Downloading sympy (6.0MiB)
Downloading rapidocr (26.0MiB)
Downloading transformers (12.4MiB)
Downloading faker (2.0MiB)
Downloading docling-slim (1.1MiB)
Downloading torch (120.9MiB)
Downloading pydantic-monty-client (4.1MiB)
Downloading docling-parse (11.3MiB)
Downloading pyarrow (26.1MiB)
Downloading shapely (1.7MiB)
Downloading pydantic-monty-runtime (12.6MiB)
Downloading pydantic-ai-slim (2.7MiB)
Downloading scikit-learn (7.9MiB)
Downloading opencv-python-headless (39.1MiB)
Downloading onnxruntime (14.2MiB)
Downloading pylance (88.3MiB)
Downloading torchvision (1.3MiB)
 Downloaded torchvision
 Downloaded docling-slim
 Downloaded shapely
 Downloaded faker
 Downloaded pydantic-ai-slim
 Downloaded pydantic-monty-client
 Downloaded scikit-learn
 Downloaded pydantic-monty-runtime
 Downloaded sympy
 Downloaded docling-parse
 Downloaded rapidocr
 Downloaded onnxruntime
 Downloaded opencv-python-headless
 Downloaded pyarrow
 Downloaded pylance
 Downloaded transformers
 Downloaded torch
Installed 218 packages in 1m 35s

Badge

Paste this into the project README to show the current result:

[![does it install](https://img.shields.io/endpoint?url=https://doesitinstall.com/badge/io.github.ggozad__haiku-rag.json)](https://doesitinstall.com/s/io.github.ggozad__haiku-rag.html)