Multi-turn conversations with history
Long-running conversations with full history. Per-conversation settings: temperature, top-P, top-K, max tokens, system prompt. Voice playback of answers, camera capture, code-edit awareness for system owners.
AI assistant
Long-running conversations with full history. Per-conversation settings: temperature, top-P, top-K, max tokens, system prompt. Voice playback of answers, camera capture, code-edit awareness for system owners.
Use an in-house model, a self hosted AI endpoint, a private cloud deployment, or a commercial API on your own key. No imposed provider, ever — you choose where inference runs.
Attach files and images with previews. Toggle web search when you want live information. The in-app tools panel exposes capabilities when available, otherwise the assistant operates on conversation only.
Authentication and key exchange are post-quantum, as everywhere in UltimaOS. The stored content of uAI, however, sits outside the end-to-end encrypted envelope today — the details are below, stated plainly rather than glossed over.
All authentication and key exchange uses NIST-standardized ML-DSA-65 (FIPS 204) signatures and ML-KEM-768 (FIPS 203) key encapsulation. There is zero RSA, zero elliptic-curve, zero classical-only crypto in the authentication path. Defeats store-now-decrypt-later attacks.
AI history is not end-to-end encrypted today. uAI history is encrypted at rest with a key the server holds, and uSages history is stored in plaintext; in both cases the server can read it, because it assembles the model context server-side. End-to-end encrypted AI history is on the roadmap — until it ships, treat AI conversations as server-readable.
Generating an answer requires sending the relevant conversation to a language model for the duration of the request. Where you choose an external provider, that provider processes the context in plaintext at inference time — true of every AI product built on hosted models, and no storage encryption changes it. You choose the model backend, and context is sent per request.
Your private keys are derived in your browser from three secrets only you control, and are held in memory for the session only — never written to browser storage, never sent to the server. A new device derives the same keys locally from the same three secrets.
UltimaOS runs in any modern browser — Chrome, Firefox, Safari, Edge. Nothing to install. Sign in by re-entering the three secrets your keys are derived from, and the same keys are rebuilt locally on whatever device you are using.
uAI sits outside the end-to-end encrypted envelope in this release, so what you put into it is readable by the operator. The connection is protected in transit, and authentication remains post-quantum, but the stored content is not encrypted to your keys.
Anything you would not want the operator to be able to read should go in one of the encrypted apps instead — chat, files, notes or documents — until end-to-end encryption reaches uAI.
Any change you make is uploaded and picked up by every other device you are signed in on, so uAI looks the same wherever you open it.
Rewrite, extend, or summarise a selected passage in Documents. The AI runs against your chosen provider.
Apply transformations to a selection — pivot, normalise, summarise. Spreadsheets keeps the formula chain.
Draft a note from a one-line prompt. Notes inherits the draft's title and color tag.
AI provider configuration lives in Account / AI. Your model choice and API key stay under your control.
UltimaOS runs in Chrome, Firefox, Safari and Edge. Nothing to install. Sign in with your passphrase to derive your private key locally.
The launcher shows every app. uAI is right there with its capsule video preview. Click to open.
uAI works alone, but it shines when combined with the rest of the UltimaOS apps. Same identity, same workspace, one place to work.
Short answer
Not today. uAI conversation history is encrypted at rest with a key the server holds, and uSages history is stored in plaintext — in both cases the server can read your history, because it assembles the model context server-side. Separately, generating an answer sends the conversation to a language model, which processes it in plaintext at inference time. End-to-end encrypted AI history is on the roadmap; treat AI conversations as server-readable until then.
Short answer
uAI uses the same post-quantum authentication and key exchange as every other UltimaOS app — ML-DSA-65 (FIPS 204) and ML-KEM-768 (FIPS 203). Conversation history, however, is not held under the XChaCha20-Poly1305 content envelope: it is readable server-side, because the server assembles the model context.
Short answer
Yes. Sign in on any device by re-entering your three secrets and the same keys are derived locally, so your uAI content follows you. Changes sync live across all signed-in devices.
Short answer
Not today. uAI conversation history is encrypted at rest with a key the server holds, and uSages history is stored in plaintext — in both cases the server can read your history, because it assembles the model context server-side. Separately, generating an answer sends the conversation to a language model, which processes it in plaintext at inference time. End-to-end encrypted AI history is on the roadmap; treat AI conversations as server-readable until then.
Short answer
UltimaOS is free for individuals and families — permanently, not as a trial. Organizations pay per seat once early access ends, and that is what funds the free accounts. There is no per-app add-on.
Short answer
Yes. UltimaOS supports encrypted export of all your content for backup and portability. The export is encrypted to a key you control; you can store it on your own infrastructure or in a personal encrypted backup.
Short answer
No. uAI runs against the AI provider you choose, with the API key or deployment you control. UltimaOS does not train any model on your data, and the conversation is encrypted to your key so the provider you choose sees the prompt in plaintext only during inference, never at rest.
Short answer
Any OpenAI-compatible endpoint, plus Anthropic, Mistral, and your own self-hosted model. Configuration is per-slot in the AI tab of the system Settings.
No ads, no data mining, no trial clock. Individual and family accounts are free — permanently. Organisations pay per seat, and that is what funds it.