OpenOS Momo AI — a local, offline compression AI that speaks (DE/EN), inviting independent verification

Description and use

Momo AI is an offline PWA that offers lossless compression and an on‑device chat model, with verifiable SHA‑256 round‑trips and local data provenance.

Problem

Users need a privacy‑preserving, on‑device compression and chat AI that can be independently verified for correctness and performance.

Outcome

The app achieves bit‑exact lossless compression with SHA‑256 round‑trips, realistic compression ratios (−18% on text, −57% on logs, +0.6% on random data), and on‑device chat responses that reference local data. All traffic stays on the device, ensuring privacy.

Workflow

• 1. Install the PWA on a device. 2. Use the compression feature to compress a file; the app reports SHA‑256 before/after. 3. Query the chat model with a question; the response includes provenance tags. 4. Run the verification protocol via the provided curl commands to confirm engine fingerprint and round‑trip integrity. 5. Optionally fine‑tune the model locally using the night‑training script and review perplexity metrics.

Evidence

Nothing leaves the user's device — privacy by architecture, not by promise.

Source

Expected, measured evidence: bit‑exact lossless roundtrips with identical SHA‑256 hashes; realistic compression (−18% on text‑like data, −57% on repetitive logs, +0.6% overhead on random data — math, not a bug); speed measured on your own machine; the chat answers carry provenance entries like "Datenanalyse · Bibliothek (lokal)" whenever they're grounded in your data.

Source