Montevive Labs

Experiments and research demos from the Montevive team.

Secure AI for secure decisions.

WebGPU onnxruntime-web 100% local EN · ES · CA · EU · GL

Shieldstral Policy Probe

Write a moderation policy as a plain-language yes/no question and watch Mistral's open safety classifier answer, in one token, in your browser. Then run the same sentence through five languages at once — Shieldstral claims Spanish, but not Catalan, Basque or Galician.

Open demo →
WebGPU transformers.js v4 100% local

OpenAI Privacy Filter

PII detection running entirely in your browser. Detects names, emails, phones, addresses, secrets, and more — with character-level spans and confidence scores. No backend, no API calls, no data transmission.

Open demo →
transformers.js v4 100% local ES · CA · EU · GL

ALIA Tokenizer Comparison

Side-by-side: how do the ALIA (Spain), Llama 3, and Mistral tokenizers handle the same text? Demonstrates ALIA's ~1.7× efficiency on Iberian languages. Live tokenization in your browser.

Open demo →