UltraRAG

Use it when

• You are prototyping iterative RAG logic with loops and conditional branches and writing the orchestration code by hand is dominating the work
• You need to reproduce and compare RAG baselines on standard benchmarks with unified metrics instead of ad-hoc per-paper evaluation scripts

What it solves

Not the fit when

• The retained README does not state a software license; verify reuse and redistribution rights before incorporating the source into another product
• A hosted turnkey RAG SaaS; the project is positioned for research exploration and industrial prototyping
• The all-extras Linux GPU environment is locked to CUDA 12.9 and the official vLLM cu129 wheel; use core or on-demand extras when that stack is unsuitable
• Assuming every Docker image is GPU-only: separate base CPU, base GPU, and full GPU images are documented
• A standalone vector database or general-purpose orchestration framework outside RAG pipelines
• hosted turnkey RAG SaaS
• standalone vector database
• general agent orchestration outside RAG workflows

Install

git clone https://github.com/OpenBMB/UltraRAG.git --depth 1 && cd UltraRAG && uv sync for core/UI features, uv sync --all-extras for the full CUDA 12.9-locked environment, or uv sync --extra retriever / --extra generation for selected modules

Invoke

ultrarag run examples/experiments/sayhello.yaml to verify install, then define pipelines with loops and conditional branches in YAML; the UI at http://localhost:5050 provides a visual pipeline builder and one-click chat demo

Alternatives

No reviewed alternatives recorded yet.