Skip to content
Bedrova bedrova
FeaturesCompatibilityDevelopersCompareDocs Coming soon

Docs

Getting started

Install Bedrova and run your first local model — system requirements, install, first-launch, and your first chat.

Bedrova runs every kind of AI on your Mac — locally, in parallel, through one API. This guide takes you from download to your first chat. There’s no terminal step, no Python to install, and no server to start by hand.

What you need

  • Apple Silicon Mac (M1 through M4). Intel Macs are not supported — Bedrova is MLX-first, with Core ML and GGUF provider packs where appropriate, and builds on Apple’s local GPU/ANE stack.
  • macOS 14 (Sonoma) or later.
  • Unified memory decides which models you can run, not which Mac you own. A 16 GB Mac runs 7–8B chat models comfortably; a 256 GB Mac Studio runs DeepSeek V4 Flash. See What your Mac can run for the full picture.
  • Free disk for the models you download. A 4-bit 7B model is a few GB; a frontier model is tens to hundreds of GB. Models can live on an external drive — see Data & file locations.

1. Install

Download the .dmg from the download page, drag Bedrova to Applications, and open it. Or use Homebrew:

brew install --cask bedrova

Screenshot: the welcome screen — added at launch.

2. First launch

Shipping builds are signed and notarized by ideius, so they open normally. If you downloaded a build that Gatekeeper doesn’t recognize yet, right-click the app and choose Open, then confirm. You do this once. If you hit a wall here, see Troubleshooting.

3. Let Bedrova read your Mac

On first run Bedrova checks your chip, unified memory, and free disk, then recommends a starter model that fits with headroom — plus an optional larger pick if you have the RAM. This is the same fit logic described in What your Mac can run: it reserves roughly a quarter of unified memory for the rest of the system before deciding what fits.

4. Download a model

Click the recommended model to fetch it. Downloads are resumable, run a free-space check first, and verify each file by SHA-256 when it lands. You can pause, resume, or repair a download at any time. More in Models & downloads.

5. Chat

When the model finishes loading, type a message and send it. You’re now running a model fully on-device — nothing you type leaves your Mac. Open the token inspector to watch time-to-first-token and tokens/sec as it generates.

Where to next