Nyx AI

Run a local model on a compatible phone (beta), without a cloud subscription or per-message provider fees. Nyx AI also offers document tools, interactive previews and Wi-Fi or cloud model connections.

Run AI on your own hardware
• Use the optional on-phone model without a cloud AI subscription or per-message provider charges. After downloading it, text generation runs on your phone without internet. Web tools and other online features still need a connection.
• This is not unlimited processing: conversation, reply and runtime limits apply. Speed and availability depend on memory, battery and heat. Download/data and electricity costs may apply.
• On-phone beta: Qwen3.5 4B, Unsloth quantisation from Hugging Face, under its own Apache-2.0 licence. Separate 2.7 GB download; needs a compatible 8 GB-class phone, supported 64-bit processor, at least 4 GiB available RAM and 3.3 GB free storage. Text only. Actual on-phone inference, speed and document generation remain unverified for this release. Resource checks cannot guarantee compatibility or prevent every crash or overheating.

Or connect a model
• Use Ollama or LM Studio running on your computer over your own Wi-Fi, or connect Ollama Cloud, OpenAI, Anthropic or another OpenAI-compatible service.
• Connected models do not require the phone to meet on-phone inference requirements. Provider accounts, API keys and charges may apply; no AI subscription is included. No Nyx AI account is needed.

Create, preview, save
• Make Word documents, Excel workbooks with formulas, PowerPoint slides and PDFs. Preview them in the app, ask for changes, then choose Save to device or Share.
• Attach photos, PDF, Word, Excel and text files. Model capabilities vary; reading photos needs a compatible vision model.
• Slides can use your pictures or approved Wikimedia Commons pictures with credits. Check the image licence for your intended use.
• Open Markdown and SVG previews. Try interactive HTML previews after approving each opening; approved code can contact other websites.
• Plan mode answers in chat without creating files.

Research with control
• Web search (DuckDuckGo) and page reading are on by default, with separate switches. In default mode, approval is required before the first web request in a reply; approved searches also cover pages on returned sites. Auto-accept changes these prompts.

Your data and choices
• Connected-model messages go to your chosen provider under its terms. Local network servers may use unencrypted HTTP; Nyx AI asks you to confirm that first.
• The optional privacy sanitizer is on by default and best-effort, not a guarantee. It filters recognised secrets from hosted-model text, not ordinary personal details by default. It skips local Ollama/LM Studio except cloud models, image pixels and web requests.
• API keys use Android Keystore-backed encryption. Chats and files are stored in app-only storage on your phone. Save and Share create external copies.
• Saved notes are on by default and included with model requests. Review, delete or disable them in Settings.
• Delete all my data clears app data, keys and the downloaded model, not exported copies or provider-held data. Our report service holds submissions up to 30 days. See the Privacy Policy for report follow-up, Google Play reports and Android platform processing.

Made to be readable
• Light, dark or system theme, text up to 200%, screen reader labels, offline read-aloud and a guided tour. English interface; replies can use other languages.

Please note
• Free beta for ages 18+, licensed for personal, non-commercial use. App and model licence terms apply; see Legal in the app.
• AI can be inaccurate, incomplete or biased. Check outputs before relying on them; they are not professional advice. Report problematic responses with Report AI content.
• By NYX LIMITED, England and Wales, company 17261427. Not affiliated with or endorsed by named providers or Microsoft; trademarks belong to their owners.