High-performance RAG search tool for AI agents. 160ms keyword search, 5.1MB binary, 110MB memory.

Build Index Guide | API Service | Codebase Overview | 中文文档

Arag CLI is a RAG search tool designed for AI agents — not humans. It provides keyword search (BM25), semantic search (embeddings), and hybrid search (RRF fusion) for agentic RAG workflows. Built in Rust, 50x faster than the Python version.

Features

• Keyword Search — BM25 with multi-keyword support and snippet extraction
• Semantic Search — Embedding-based retrieval via HTTP API
• Hybrid Search — BM25 + semantic RRF fusion with personalized ranking
• Structure-Aware Chunking — Tables and code blocks are atomic, never truncated
• Multi-KB Management — Create, switch, and manage multiple knowledge bases
• Incremental Updates — Auto-detect file changes via mtime/size diff
• Scheduled Auto-Update — Daemon mode for periodic index refresh
• Query Cache & History — JSONL-based cache with followup inference
• User Profile — Hot/cold chunk analysis, blind spot detection, adoption rate
• Agent Integration — JSON schema output, --format text for LLM consumption
• REST API — FastAPI service wrapping all CLI commands

Quick Install

# Linux / macOS
curl -sSL https://github.com/KuaishouGameMind/arag-cli/releases/latest/download/install.sh | bash

# Windows (PowerShell)
Invoke-WebRequest -Uri https://github.com/KuaishouGameMind/arag-cli/releases/latest/download/install.bat -OutFile install.bat
.\install.bat

Or build from source (requires Rust 1.75+):

git clone https://github.com/KuaishouGameMind/arag-cli.git
cd arag-cli && cargo build --release
ln -sf "$(pwd)/target/release/arag-cli" ~/.cargo/bin/arag-cli

Quick Start

# Set up embedding API (optional, semantic search will be unavailable without it)
export ARAG_EMBEDDING_BASE_URL="https://your-api.example.com/v1"
export ARAG_EMBEDDING_MODEL="your-model-name"
export ARAG_EMBEDDING_API_KEY="your-api-key"

# Build index from documents
arag-cli build-index --input-dir ./docs

# Keyword search
arag-cli keyword-search "warrior" "skill" -k 5

# Semantic search
arag-cli semantic-search "What are the sequences of the Warrior pathway" -k 5

# Read a chunk
arag-cli read-chunk "1850"

# Hybrid search (BM25 + semantic fusion)
arag-cli search "warrior pathway sequences" -k 10

Environment Variables

Variable	Description
ARAG_EMBEDDING_BASE_URL	Embedding API address (optional, required for semantic/hybrid search)
ARAG_EMBEDDING_MODEL	Embedding model name
ARAG_EMBEDDING_API_KEY	Embedding API key (optional, required for semantic/hybrid search)
ARAG_INDEX_DIR	Index directory (overridable by -i)
ARAG_KB	Knowledge base name (overridable by --kb)
ARAG_SESSION	Session file path

Global Options

Option	Description
--kb <name>	Specify knowledge base name
-i, --index-dir <PATH>	Directly specify index directory (highest priority)
--format <json|text>	Output format (default: json)
--session <PATH>	Session file path

Citation

This project is based on the original arag project and the A-RAG paper.

@misc{du2026aragscalingagenticretrievalaugmented,
      title={A-RAG: Scaling Agentic Retrieval-Augmented Generation via Hierarchical Retrieval Interfaces},
      author={Mingxuan Du and Benfeng Xu and Chiwei Zhu and Shaohan Wang and Pengyu Wang and Xiaorui Wang and Zhendong Mao},
      year={2026},
      eprint={2602.03442},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2602.03442},
}

License

MIT