metainflow-studio-cli

metainflow-studio-cli is a Python CLI toolkit. The current implemented commands are parse-doc, search-summary, web-crawl, enterprise-query, enterprise-search, and enterprise-balance.

Skills

Project skills live in metainflow-skills/ inside this repository so they stay versioned with the CLI implementation.

For local OpenCode discovery, symlink skills into ~/.agents/skills/ from the repo root:

ln -sfn "$(pwd)/metainflow-skills/metainflow-doc-parse" "$HOME/.agents/skills/metainflow-doc-parse"
ln -sfn "$(pwd)/metainflow-skills/metainflow-web-search" "$HOME/.agents/skills/metainflow-web-search"
ln -sfn "$(pwd)/metainflow-skills/metainflow-web-fetch" "$HOME/.agents/skills/metainflow-web-fetch"
ln -sfn "$(pwd)/metainflow-skills/metainflow-enterprise-query" "$HOME/.agents/skills/metainflow-enterprise-query"

More agent-facing setup details are in docs/agent-usage.md.

Available repo-local skills:

• metainflow-doc-parse
• metainflow-web-search
• metainflow-web-fetch
• metainflow-enterprise-query

Quick Start

python -m pip install -e .[dev]
pytest -q
python -m metainflow_studio_cli.main parse-doc --file ./sample.txt --output json
python -m metainflow_studio_cli.main search-summary --query "React 19 新特性" --output json
python -m metainflow_studio_cli.main web-crawl --url https://example.com --output json
python -m metainflow_studio_cli.main enterprise-search --keyword "示例智能" --output json
python -m metainflow_studio_cli.main enterprise-query --type business --keyword "示例智能（深圳）科技有限公司" --output json

Search Summary

Use search-summary when you need keyword-based web search plus AI-generated summary.

The current routing strategy is:

1. Zhipu-compatible provider web search
2. SearXNG fallback
3. Baidu Playwright fallback

After results are collected, the configured summary model generates the final answer.

Web Crawl

Use web-crawl when you already have a target URL and need page extraction with optional summarization. This command uses Crawl4AI for page retrieval.

Enterprise Query

Use enterprise-query for exact enterprise detail lookup, enterprise-search for fuzzy candidate search, and enterprise-balance for balance diagnostics. The recommended agent routing is exact-first, fuzzy-fallback for strong full names and fuzzy-first for ambiguous fragments. Pass --session-id when you want cache reuse across repeated lookups for the same agent session.

Supported parse-doc extensions

• .pdf
• .doc (converted via LibreOffice soffice)
• .xls (converted via LibreOffice soffice)
• .docx
• .pptx
• .xlsx
• .csv
• .txt, .md
• .html

Environment variables

Copy .env.example to .env and fill in your values:

cp .env.example .env

Variable	Default	Description
PROVIDER_BASE_URL	https://api.openai.com/v1	Base URL for provider-compatible requests. Set this to https://open.bigmodel.cn/api/paas/v4 if you want the Zhipu provider path for search-summary.
PROVIDER_API_KEY	(required)	API key
PROVIDER_TIMEOUT_SECONDS	60	Request timeout in seconds
PROVIDER_MAX_RETRIES	2	Max retries on failure
PROVIDER_MODEL_DOC_PARSE	gpt-4.1-mini	Model used by parse-doc
PROVIDER_MODEL_WEB_SEARCH	glm-4-air	Reserved search-provider model setting
PROVIDER_MODEL_WEB_FETCH	gpt-4.1-mini	Model used by web-crawl summarization
SUMMARY_BASE_URL	(falls back to PROVIDER_BASE_URL)	Optional separate endpoint for search-summary summarization
SUMMARY_API_KEY	(falls back to PROVIDER_API_KEY)	Optional separate key for search-summary summarization
SUMMARY_MODEL	glm-4-flash	Model used to summarize search results
SEARCH_PAGE_TIMEOUT_SECONDS	30	Baidu Playwright page timeout
WEB_SEARCH_BACKEND	auto	auto / zhipu-web-search / searxng-web-search / baidu-playwright
SEARCH_PROVIDER_ENGINE	search_pro	Zhipu search engine tier
SEARCH_RESULT_COUNT	10	Number of search results to request
SEARXNG_BASE_URL	http://localhost:8080	Endpoint for the SearXNG fallback
METAINFLOW_WEB_FETCH_VERIFY_SSL	1	Whether web-crawl verifies SSL certificates
METAINFLOW_ENTERPRISE_API_BASE_URL	https://test.jszypt.com:42211/admin/api/getTenantApi	Enterprise query endpoint
METAINFLOW_ENTERPRISE_BALANCE_URL	https://test.jszypt.com:42211/sys-tenant-hehe/query	Enterprise balance endpoint
METAINFLOW_ENTERPRISE_API_APP_ID	(required for enterprise commands)	Enterprise API appid
METAINFLOW_ENTERPRISE_API_SECRET	(required for enterprise commands)	Enterprise API secret
METAINFLOW_ENTERPRISE_API_VERIFY_SSL	1	Whether enterprise requests verify SSL
METAINFLOW_RUN_SAMPLE_MATRIX	(unset)	Set to 1 to enable real sample matrix integration tests

Playwright fallback

The Baidu fallback path in search-summary requires the optional undetected-playwright extra and browser binaries:

pip install -e ".[playwright,dev]"
python -m playwright install chromium

Search provider note

If you want provider-backed search instead of pure fallback mode, set:

export PROVIDER_BASE_URL="https://open.bigmodel.cn/api/paas/v4"
export PROVIDER_API_KEY="your-api-key"

You can still keep SUMMARY_BASE_URL and SUMMARY_API_KEY separate for summarization.

Ubuntu dependencies

Install system packages for full .doc / .xls conversion and OCR support:

sudo apt-get update
sudo apt-get install -y libreoffice tesseract-ocr tesseract-ocr-chi-sim tesseract-ocr-eng poppler-utils fonts-noto-cjk

Real sample matrix validation

1. Put real fixture files under tests/integration/samples/ so that all required extensions exist.
2. Run:

METAINFLOW_RUN_SAMPLE_MATRIX=1 pytest -q tests/integration/test_real_sample_matrix.py

By default, the sample matrix test is skipped unless METAINFLOW_RUN_SAMPLE_MATRIX=1 is set.

Attribution: This project uses Crawl4AI (https://github.com/unclecode/crawl4ai) for web data extraction.