FAQ

Is Bedrova free? Yes — free, forever. It’s made by ideius.

Do I need Apple Silicon? Yes. Bedrova is MLX-first and native to Apple Silicon (M1–M4), and requires macOS 14 or later. Intel Macs aren’t supported.

Does it work offline? Yes. Once a model is downloaded, everything runs offline. The only network calls are model downloads, optional update checks, and opt-in analytics — all listed in Network calls.

Is my data private? Your prompts, chats, documents, audio, and generated images never leave your Mac, and there’s no telemetry by default. See Privacy & security.

What models can I run? Local Mac-runnable models supported by Bedrova’s provider packs: MLX for the preferred text/model path, Core ML for perception tasks, GGUF/llama.cpp for secondary text compatibility, and provider-specific packs for image, voice, vision, and embeddings/search. Some brand-new architectures stay capability-gated until a native Mac runtime exists — Bedrova tells you when a model can’t run yet rather than failing cryptically.

Can it run big models? If your Mac has the unified memory. The fit-check tells you what fits before you download; see What your Mac can run.

How fast is it? That depends on the model, quant, context, and your Mac. Public benchmark numbers land at launch from a formal, reproducible run — see Benchmarks. We don’t publish cherry-picked figures.

Can I run several models at once? Yes — that’s the point. Text, vision, embeddings/search, perception, voice, and image workloads run in parallel through one API, sharing one memory budget. See Concepts.

Is it open source? No — Bedrova is closed-source and free. We earn trust with published network-call docs and verify-it-yourself instructions instead of source access. See Privacy & security.

Can my other apps use it? Yes. It exposes a local OpenAI- and Anthropic-compatible API, plus native management routes. See the Server overview and the API reference.

Does it do agents or automation? No, by design. Bedrova supports tool calling and MCP for its own chat and API, but it isn’t an agent platform — durable, team, or scheduled automation is Packwolf’s job. See Tool calling & MCP.

Can Bedrova generate video? Yes. Text-to-video and image-to-video run on-device through MLX. Video is the heaviest thing Bedrova does, so plan on 32 GB of unified memory or more, and expect clips measured in seconds. See Video generation.

Does it run Stable Diffusion? Yes, for the SD1.5/2, SDXL and SD3.5 profiles Bedrova has proved, through a dedicated native engine rather than a FLUX fallback. You supply the weights yourself, because most common SD checkpoints carry licences Bedrova will not redistribute. Support is profile-specific rather than a promise that any checkpoint will load, and anything unproved stays disabled with a reason. See Image generation.

What is the Loom? It’s Bedrova’s chat surface. Unlike a normal chat window, a single thread can span text, vision, voice and image models at once, each in its own visible lane. See Chat.