ツール

NVIDIA's open-source LLM vulnerability scanner that probes for prompt injection, jailbreaks, data leakage, hallucination and harmful content, supporting OpenAI, Hugging Face, Bedrock and most REST-accessible models. Scanning cloud models needs their API keys.

Distributed data integration tool and Apache top-level project with 160+ connectors for batch, real-time and CDC sync. Define jobs in config files and run them on the built-in Zeta engine or on Flink or Spark. Requires Java.

High-accuracy retrieval-augmented Q&A over paper PDFs, text and Office documents, filling in citation metadata, checking retractions and attaching traceable sources to answers. Provides the pqa CLI and a Python library. Needs an LLM API key by default; local models can be configured.

Terminal tool for exploring and wrangling tabular data: opens dozens of formats including CSV, JSON, Excel and SQLite for filtering, sorting, pivots and frequency counts. --batch mode replays actions headlessly or converts file formats.

Say build a harness for this project and it designs a domain-specific agent team, generating subagent definitions, companion skills and orchestration, validates them and keeps evolving them from feedback. Supports workflow orchestration, persistent agent teams and subagent delegation.

PDF command line written in pure Go with almost no external dependencies: dozens of subcommands to merge, split, crop, rotate, watermark, encrypt and decrypt, extract images and fonts, and work with forms.

Manage database schemas as code: declare the desired schema in SQL, HCL or your ORM and Atlas plans, lints and applies migrations, and can export existing databases as HCL, SQL or Mermaid ER diagrams. Supports PostgreSQL, MySQL and a dozen other databases.

Converts the output of 100+ commands such as dig, ps, ls and ifconfig, plus common file formats, into JSON so agents can parse it reliably with jq or scripts. Also usable as a Python library.

Lightweight OCR built on PaddleOCR models converted to ONNX, recognizing Chinese and English by default on CPU alone. The rapidocr CLI reads text from screenshots, photos and scanned page images. Needs Python; downloads models on first run, then works offline. Convert PDFs to images first.

Offline speech toolkit for Apple Silicon Macs: CLI text-to-speech (including Chinese and voice cloning) and speech-to-text, for voicing lessons and explainer videos or transcribing lectures and interviews, with data staying local. Needs an M-series Mac and Python 3.10+; MP3 export needs ffmpeg.

Query, modify and convert JSON, YAML, TOML, XML, CSV, HCL, INI and more with one selector syntax, editing config files in place or converting between formats, handy for agents handling structured data in scripts. A single dependency-free binary that works much like jq or yq.

Turns Peng Sida's public research-writing notes into a skill: revises abstracts, introductions, methods and experiments of ML/CV/NLP papers from a reviewer's view, checks that each claim has experimental evidence, and runs a pre-submission self-check. Pure methodology, no API key.

Diffusion model inference in pure C/C++: one sd-cli command runs SD, FLUX, Qwen Image, Wan and more locally to create or edit images and short videos, with Metal acceleration on Mac. No cloud needed; bring your own model weights and use a release binary or build from source.

Scrapes web pages into Markdown or structured data, with full-site crawling, sitemaps, web search and interaction with dynamic pages. Maintained by Firecrawl; the cloud needs an API key (self-hosting is possible), and the keyless hosted endpoint only offers scrape, search and parse.

OpenAI's official Codex plugin for equity investors: earnings previews and recaps, DCF, three-statement and comps valuation, model updates, long/short pitches, catalyst calendars, sizing and hedging, and investment memos. Works with FactSet and PitchBook (paid) or uploads and public sources.

Build with OpenAI best practices inside Codex: call the OpenAI API, build and deploy agents with the Agents SDK, create ChatGPT apps and prepare submission materials, and create API keys directly. Maintained by OpenAI; Codex only.

Creates, edits and validates reproducible Python and SQL Jupyter notebooks: organizes cells as reports or tutorials, checks conclusions against actual output and reruns top to bottom before handoff. Maintained by OpenAI in the Data Analytics plugin of openai/plugins.

Open-source LLM evaluation platform: one opencompass command picks the model and datasets and runs the evaluation across hundreds of benchmarks. Works with local Hugging Face models and API models, with vLLM and LMDeploy for faster inference.

Converts and two-way syncs Jupyter notebooks with plain Python scripts or Markdown, so agents can edit research and teaching notebooks like ordinary text and diffs stay readable. Needs a local Python environment; install with pip.

Cleans a LaTeX project before arXiv submission: removes comments, auxiliary files and unused .tex files and images, and can compress images, so internal notes are not published with the source. From Google Research; installable with Homebrew on macOS.

Python library and CLI for extracting main web content: fetches pages, crawls via sitemaps and RSS feeds, strips headers, footers and other noise, and outputs TXT, Markdown, JSON, XML and more with metadata. Also processes downloaded HTML offline.

Anthropic's official OS-level sandbox CLI: restricts file and network access for agents, MCP servers and any command without containers, using sandbox-exec on macOS and bubblewrap on Linux. Currently a research preview.

Veteran suite of CSV command-line tools: in2csv converts Excel and JSON to CSV, csvcut, csvgrep and csvsort filter and sort, csvstat computes statistics, and csvsql runs SQL directly against CSV files.

Live stock and crypto quotes in the terminal with portfolio profit and loss tracking, using Yahoo Finance and Coinbase data. The main view is an interactive dashboard; ticker print exports holdings from your config at current prices as JSON or CSV for scripts and agents.