Features
Useful work, with the boundaries kept visible.
Nyx AI helps with documents, data, research and code. Start with the outcome below; the Manual and Security pages carry the implementation detail.
Bring the model that suits you
- Use local Ollama models, Ollama :cloud or -cloud, OpenAI, Anthropic, or supported compatible endpoints. Nyx AI includes no model or inference runtime of its own.
- A loopback Ollama model can keep model prompts on this PC. Remote Ollama, any model ending :cloud or -cloud, and direct cloud providers are off-device and use the provider account or key you configure.
- Auto uses suitable local models for fast helper work and normally prefers an eligible configured cloud route for heavier roles. Prefer local is a preference with cloud fallback, not an egress block.
- The model you pick is primary for the turn. An unavailable or stalled exact model may visibly fall back within the same provider, but it is never silently swapped to a different provider off this machine: if the id you chose is advertised by two configured providers, or has moved, or is no longer offered by the provider you chose it from, Nyx AI refuses the call rather than rerouting it.
Documents, spreadsheets and decks
- Create polished DOCX, XLSX, PPTX and PDF files. Nyx AI Lab can also create ODT and LaTeX source.
- Build formulas, formats, tables, charts, slide layouts, speaker notes, equations and branded reports instead of pasting plain text into a file.
- Read attached office files in Chat, or work with files in an opened workspace in Nyx AI Lab. Review every result before relying on it.
Code and project work
- Read and edit files in the selected workspace, search a codebase, run shell and Python tools, and use Git and language-server help.
- Chat works without a project folder. Nyx AI Lab adds workspace tools, previews, project memory and automation.
- The app can suggest edits, run checks and help diagnose failures, but AI output can be wrong or insecure: inspect changes and test them yourself.
Research and browsing
- Search and fetch public web pages, use a guarded headless research browser, and turn findings into local reports.
- Web tools, remote images and browser pages make network requests. Prompt injection remains possible, so sources and conclusions still need human checking.
- Scheduled research can run without you watching. After it accesses workspace files, later web-search, web-fetch and remote PDF-image requests are blocked for that run; configured model-provider calls are not blocked.
Modes, approvals and automation
- Default can run eligible low-risk writes and shell commands automatically and asks before higher-risk ordinary model-issued actions. Auto-accept accepts more; Bypass asks least. Enabled Auto-verify has a documented bounded prompt-free path; the guided Build pipeline is not reachable in this free build. Outside a Sandbox isolation level, one Allow for repository-controlled execution covers later commands in that category in the workspace for the rest of the app session, until you close Nyx AI; high/critical commands keep their own confirmation, and designated sensitive categories still require a fresh decision or are refused.
- Plan blocks workspace file changes and shell/Python execution while it drafts a plan. Read/status tools, guarded web access, session state and MCP tools heuristically classified as read-only can still operate; an MCP tool's name is not proof that it has no side effects.
- Create recurring tasks, connect MCP servers and use hooks. A scheduled task you set up in chat mode runs with no workspace, so its artifacts are saved under Documents\Nyxion\Scheduled, the compatibility folder retained from the former product name. Plugins have to be placed on disk by hand in this free build, and there is no installed-plugin screen to enable one from either: a plugin folder that already carries an enabled state loads its MCP servers and hooks automatically when you open a workspace you have trusted. These integrations can run programs or contact services, so enable only ones you trust.
- Action and Forensic logs help you inspect what happened. They are a local self-check, not evidence for an adversarial party, because the key is on this machine, and not a guarantee that every outside effect was observed.
Execution isolation
- Automatic tries Windows AppContainer and may fall back to Standard. Standard adds screening and best-effort Job-Object limits, but is not a filesystem or network jail. On Automatic and Standard, briefly at start-up and where Windows refuses both backends, a screened command can run with no OS jail at all — the tier badge shows what actually ran. An explicit Sandbox level refuses instead.
- Explicit Sandbox levels use AppContainer and fail closed if it is unavailable. Isolation off runs with your normal Windows permissions.
- AppContainer is the stronger write boundary, not a promise that everything outside the workspace is unreadable. No sandbox is 100%.
Privacy and network controls
- Nyx AI sends NYX LIMITED no product usage telemetry, analytics or crash reports. The default-on update check still contacts the update host with the app version and exposes the ordinary request IP; you can turn it off.
- When enabled, the text privacy sanitizer tries to remove recognised secrets. Its default secrets-only setting does not remove ordinary emails, phone numbers, file paths or IP addresses, and it does not cover images, web requests or MCP traffic.
- Local chats, settings and logs are stored under your Windows account. Cloud providers and connected services receive the data needed for the features you choose to use.
What you need
- Windows 10 or 11 on x64 hardware. Windows on ARM, macOS and Linux are not supported, distributed or tested.
- For local AI, install Ollama separately, keep it running and pull a model. For cloud AI, bring an account or key; provider charges and terms may apply.
- Nyx AI is a proprietary beta for adults aged 18 or over. The personal, non-commercial licence is £0; commercial use requires a separate written licence from NYX LIMITED.
Note: Nyx AI is Windows-only. It runs on Windows 10 or 11 (x64) and is not supported, distributed, or tested on Windows on ARM, macOS, or Linux.
Wondering how this stacks up against the usual options? See the comparison.
Try it on your own work.
The personal beta licence is £0, for personal, non-commercial use. No Nyx AI account needed; Ollama at a local or remote Base URL needs no API key in Nyx AI, or use your own cloud-provider key. Remote Ollama routes are off-device.