ate-rag-kb

ATE RAG Knowledge Base

English | 中文

Your coding agent's long-term memory for ATE platform knowledge.

Query ATE technical documentation, APIs, error codes, and debug flows directly from Claude Code, Cursor, Codex, or other MCP-enabled agents. Get reliable, cited answers about timing, patterns, DPS, PMU, and test flows without leaving your IDE.

Built for: Test engineers using AI coding assistants, teams maintaining ATE test programs (V93000、J750), and anyone with authorized local ATE documentation who needs grounded, cited answers.

---

Prerequisites — Install Required Software

Before you start, install these three tools. The KB runs on both macOS and Windows. Everything else (Python itself, the Qdrant database, the embedding models) is handled for you by uv and Docker — you do not install them separately.

Software	Why it's needed	Download	Verify after install
Git	Clone this repository	git-scm.com/downloads	git --version
uv	Installs Python 3.10+ and all dependencies automatically	docs.astral.sh/uv	uv --version
Docker Desktop	Runs the Qdrant vector database (server mode, the default)	docker.com/products/docker-desktop	docker --version

Python 3.10+ is required, but you do not download it yourself — uv installs the correct version automatically. If you prefer a manual install anyway, get it from python.org/downloads.

One-line install commands

macOS — using Homebrew (install Homebrew first if you don't have it):

brew install git
brew install uv
brew install --cask docker   # then launch Docker Desktop once from Applications

Windows — using winget (built into Windows 10/11), in PowerShell:

winget install Git.Git
winget install astral-sh.uv
winget install Docker.DockerDesktop   # then launch Docker Desktop once from the Start menu

After installing, open a new terminal window and confirm all three respond:

git --version
uv --version
docker --version

Windows users: Run every command in this guide in PowerShell (not the old cmd prompt). Make sure Docker Desktop is running (whale icon in the system tray) before any docker compose command.

---

Quick Start (15–30 min)

Every command and path below assumes you are inside the ate-rag-kb folder (the "project root").

1. Clone the repository

Pick a folder you'll remember (your home directory is fine):

git clone https://github.com/WalterLuo/ate-rag-kb.git
cd ate-rag-kb

2. Install dependencies

uv sync

3. Start Qdrant server

docker compose up -d qdrant

4. Download embedding models

Option A: use the pre-packaged cache (~6.4 GB). Download from PikPak and unzip into the project root.

Option B: let Hugging Face download on first use (temporarily set local_files_only: false in configs/config.yaml).

Option C: use a cloud API provider (e.g., SiliconFlow) for GPU-accelerated inference without downloading models locally. Set the API key and update configs/config.yaml — see the Cloud API for Embedding and Reranking section for details.

uv run python scripts/verify_models.py

5. Prepare local authorized Markdown documents

mkdir -p data/raw/markdown/v93000/smt7

Copy or generate your own authorized Markdown files into data/raw/markdown/.

6. Ingest documents

First run is full; subsequent runs use --incremental.

uv run -m ate_rag_kb.cli.main ingest --dir ./data/raw/markdown

7. Configure your agent

uv run python scripts/install_mcp.py --install-agent-policy

Or see Agent Integration below for manual configuration.

8. Validate and restart

uv run python scripts/validate_plugin_install.py
uv run python scripts/validate_agent_routing_policy.py

Then restart Codex / Claude Code / Cursor.

9. Start the MCP server (for agent integration)

uv run -m ate_rag_kb.cli.main mcp

Or start the HTTP API (for direct access):

uv run -m ate_rag_kb.cli.main serve --host 0.0.0.0 --port 8080