RAGFlow Workbench

Prerequisites: This skill requires uv (Python package manager) and docker (container runtime).

Prefer using scripts from scripts/. Use --json by default to ensure structured output for automation.
User-facing output follows references/output-format.md.

Smart Execution Flow

Determine the RAGFlow environment readiness from the .env file to avoid redundant checks:

┌─ Check if .env contains a valid RAGFLOW_API_KEY?
│
├─ ✅ Yes (connection established)
│    Skip environment checks, use API directly
│    bootstrap_admin.py / configure_default_models.py do not need re-execution
│    Start from API workflows: datasets.py / upload.py / search.py etc.
│
├─ ❌ No (first-time use)
│    ⚠️ Note: The following steps involve environment detection and privileged operations;
│       you MUST confirm with the user before executing.
│
│    1. Ask the user for their intent (present these options; add more as context allows):
│       a. "Have you not yet downloaded and installed RAGFlow locally? Would you like me to help you download and install it?"
│       b. "Do you already have a local RAGFlow instance running? Please provide the address and login credentials, or an existing API Key — I will test the connection automatically and let you know the result."
│
│    2. Execute based on user response:
│       ┌─ User chooses fresh install (a)
│       │   Guide user to download Docker Desktop → deploy RAGFlow → wait for readiness
│       │   Then continue with check → bootstrap → configure workflow
│       │
│       ├─ User provides existing instance info (b1)
│       │   1. Test connection with provided address/credentials
│       │   2. Success → write to .env, skip env checks in future sessions
│       │   3. Failure → report specific error, guide troubleshooting
│       │
│       ├─ User provides existing API Key (b2)
│       │   1. Test key validity
│       │   2. Valid → write to .env, skip all bootstrap steps
│       │
│       └─ Other cases → handle flexibly based on actual user response
│
│    3. Automated execution after confirmation (choose as appropriate):
│       - check_windows_install.py --json    Environment check (after fresh install)
│       - bootstrap_admin.py --json          Bootstrap + API Key
│       - configure_default_models.py --json Configure default models
│       → .env now contains RAGFLOW_API_KEY; future sessions skip env checks
│
└─ 🚀 Enter API Workflow

How it works: After bootstrap_admin.py succeeds, it writes RAGFLOW_API_URL and RAGFLOW_API_KEY to .env. As long as these values exist and are non-empty, the environment is considered validated and check_windows_install.py and other setup scripts will not be re-run.

Use Cases

• First-time RAGFlow setup on Windows, ready out of the box
• Auto-create admin user and obtain API Key
• Configure default Embedding / Chat / Rerank models
• Create knowledge bases, upload files, trigger parsing, check parse status
• Search knowledge bases, manage datasets/documents, create chat sessions

Full Lifecycle Workflow

check_windows_install.py  →  bootstrap_admin.py  →  configure_default_models.py
                                                          ↓
                          datasets.py / upload.py / parse.py / parse_status.py
                                                          ↓
                          search.py / create_chat.py

Quick Start

uv venv
.\.venv\Scripts\Activate.ps1
copy .env.example .env

uv run python scripts/check_windows_install.py --json
uv run python scripts/bootstrap_admin.py --json
uv run python scripts/configure_default_models.py --json

Knowledge Base & Document Management

uv run python scripts/datasets.py create "Sample Knowledge Base" --json
uv run python scripts/upload.py DATASET_ID /path/to/file.pdf --json
uv run python scripts/parse.py DATASET_ID DOC_ID --json
uv run python scripts/parse_status.py DATASET_ID --json

Search & Chat

uv run python scripts/search.py "query text" DATASET_ID --json
uv run python scripts/create_chat.py "Chat Name" --dataset-ids DATASET_ID --llm-id qwen2-7b-instruct --json

Full Command Reference

See references/command-reference.md for all commands and parameters, including:
- datasets.py list / info / create / delete
- upload.py list / delete + file upload
- parse.py / parse_status.py / stop_parse_documents.py
- search.py full retrieval parameters
- update_dataset.py / update_document.py
- list_models.py / create_chat.py

Execution Constraints

• Delete operations must first list candidates and obtain explicit user confirmation
• Only delete using explicit dataset_id / document_id
• Upload does not automatically trigger parsing; only run parse.py when the user requests it
• parse.py only initiates tasks; progress must be checked via parse_status.py
• If parse_status.py returns a progress_msg, echo it verbatim; when status is FAIL, treat it as the primary error and guide the user to references/troubleshooting.md
• bootstrap_admin.py and configure_default_models.py require Docker containers accessible via docker exec

Output Rules

• Follow references/output-format.md
• Use tables for 3+ structured data items
• Preserve api_error, error, message and similar fields as-is
• Do not fabricate progress percentages, failure reasons, or model configuration results