Installation
Download
| Platform | Installer |
|---|---|
| macOS (Apple Silicon) | .dmg — arm64 |
| macOS (Intel) | .dmg — x64 |
| Windows x64 | .exe setup |
| Windows ARM64 | .exe setup |
| Linux x64 | .AppImage, .deb, .snap, .flatpak |
| Linux ARM64 | .AppImage, .deb |
Always-current links are on the latest release page; all releases are listed too.
Internet is required on first launch (to fetch the runtime and any components you enable). After that, everything works offline.
System requirements
| Local models | Remote only | |
|---|---|---|
| Disk | 5 GB+ | ~500 MB |
| RAM | 16 GB+ | 4 GB |
| OS | macOS 12+, Windows 10+, modern Linux (glibc 2.28+) | Same |
Local model memory
Local models need serious RAM — roughly 8 GB for a 7B model, 16 GB for 13B. On a lighter machine, connect to a remote server instead.
First launch
- Open the app. On macOS, allow it past Gatekeeper the first time.
- Either add a connection to an existing Hrida.ai server, or install the local server (Studio) from Settings.
- Optionally enable the local inference engine and download a model.
- Set a default connection so Spotlight and voice input know where to send messages.
Privacy
No telemetry, no tracking, no phone-home. Your conversations stay on your machine. The only outbound requests the app itself makes are: fetching runtime components you explicitly enable, checking for updates, and talking to whichever Hrida.ai server(s) you connect it to. Running models locally keeps everything on-device.