/
checking…
ES
A multilingual toolkit for responsible AI

One ethical-AI toolkit, many surfaces — glossary, audits, evals, and SDKs.

EthicalStack ships an ICAIRE-grounded multilingual glossary alongside an HTTP API, CLI, MCP server, Python & JS SDKs, a browser extension, a live model evaluator, and an ethical-AI usage auditor — all backed by the same Chroma-indexed dataset.

—
Glossary terms
EN · AR · FR
3
Languages
English · Arabic · French
8
Surfaces shipped
API · CLI · MCP · SDKs · Ext · Auditor · Evaluator
—
Audit backend
checking…

What's in the toolkit

Click any surface to jump in
📚

Multilingual glossary

1,200+ AI & data terms with EN / AR / FR definitions, aliases, and source attribution.

🛡️

Ethical AI auditor

Score text, URLs, or files (PDF/HTML/MD) against ten responsible-AI dimensions.

📊

Live model evaluator

Pick a model and benchmark it term-by-term against the Arabic glossary, judged 1–5.

🌐

Browser extension

Auto-highlight glossary terms on every page with hover translations.

⚡

FastAPI backend

REST endpoints for lookup, prefix & semantic search, annotation, and audit.

🧠

MCP server

Plug the glossary and auditor into Claude Desktop, Cursor, or VS Code.

💻

CLI & SDKs

Terminal-first lookups plus zero-dep Python and TypeScript clients.

🧬

Vector database

Persistent Chroma index with MiniLM embeddings powering every semantic lookup.

Quick start

Two-minute setup
  1. 1
    Spin up the API (After Cloning Repo)
    cd api && uvicorn app.main:app --reload --port 8000
  2. 2
    Hit any surface
    curl http://localhost:8000/terms/Bias
python cli/ethicalstack_cli.py audit paper.txt
  3. 3
    Open this dashboard
    open http://localhost:8000/dashboard/

Recent audits

No audits yet — paste some text into the auditor to get started.

Ethical AI Usage Auditor

heuristic fallback

Paste an abstract, model card, technical spec, grant proposal, or syllabus. The auditor scores ten ethical-AI usage dimensions — anchored in the ICAIRE multilingual glossary — surfaces the evidence it found, and flags concrete responsible-AI gaps a reviewer should escalate.

Ctrl+Enter to run
Audit results will appear here.

Glossary Explorer

— normalized AI & data terms with English, Arabic, and French definitions and aliases. Click any term for details.

Tools & Downloads

Same backend, multiple surfaces. Pick the one that fits your workflow.

⚡

FastAPI Backend

REST endpoints for lookup, prefix & semantic search, annotation, and the ethical-AI usage audit.

Read more →
🧠

MCP Server

Plug the glossary into Claude Desktop, Cursor, or VS Code as a tool-calling MCP server.

Read more →
💻

CLI

Terminal-first lookups, semantic search, annotation, and audits straight from your shell.

Read more →
🐍

Python SDK

Zero-dependency drop-in client for the EthicalStack API.

Read more →
🟨

JS / Node SDK

Same API, idiomatic JavaScript. Browser- and Node-friendly.

Read more →
🧬

Vector Database

Persistent Chroma index of every glossary entry, embedded with MiniLM for semantic search and concept retrieval.

Read more →

Live Model Evaluation Benchmark

Pick a target model, run the ICAIRE Arabic-glossary benchmark live against it, and watch each term-by-term score land in real time. The judge model grades each Arabic explanation 1–5 against the ground-truth definition.

Browser Extension

Surfaces the ICAIRE glossary directly inside any web page — auto-highlight, hover tooltip, and right-click lookup. Manifest V3 for Chrome & Edge.

🌐

What it does

  • Auto-highlight on every page — terms from the glossary are detected on load.
  • Hover tooltip — definition + Arabic / French translations appear inline.
  • Right-click → Lookup on any selection.
  • Configurable API base URL (defaults to http://localhost:8000).
  • Themes & language pairs (en-ar / ar-en / en-fr).
  • Caches results in chrome.storage.local — one network roundtrip per day.
📦

Install

  1. Download the .zip and extract it.
  2. Open chrome://extensions → enable Developer mode → click Load unpacked.
  3. Pick the unpacked folder, then open the options page to set your API URL.
Download .zip

Audit history

Locally stored on this device — your last 20 audits.

Glossary Data & Contributing

The dataset is open and reproducible. Bring corrections, aliases, or new terms.

Dataset

  • Source: ICAIRE multilingual Excel (EN / AR / FR)
  • Pipeline: scripts/ingest_glossary.py
  • Outputs: data/glossary.json, glossary.csv
  • Versioning: dataset_meta.json
View dataset_meta Sample JSON

How to contribute

  1. Edit the dataset xlsx or add aliases to data/aliases.json.
  2. Re-run python scripts/ingest_glossary.py to regenerate JSON / CSV.
  3. Run the eval suite: python evals/run_benchmark.py.
  4. Open a PR — the dataset hash updates automatically.