CLI

Bedrova ships a command-line interface for scripting and headless use. It talks to the same local runtime as the app, so a model loaded from the CLI is the same instance the UI and the server see.

The exact command set is being finalized for launch. Commands marked (planned) below describe the intended surface; check bedrova --help in your build for the authoritative list rather than relying on this page for flags.

When to use it

Reach for the CLI when you want to start the server on login, pull a model in a setup script, or run a quick generation without opening the app. For day-to-day chat, the app is the better tool.

Commands

Command	What it does
bedrova serve	Start the local API server (port from settings, or --port).
bedrova ls	List installed models and their loaded state.
bedrova pull <repo>	Download a model by user/model or full Hugging Face URL. (planned)
bedrova load <id>	Load a model into memory (runs the fit-check first). (planned)
bedrova unload <id>	Evict a loaded model. (planned)
bedrova run <id>	Run a one-off prompt and print the result. (planned)
bedrova diagnostics	Write a redacted diagnostics bundle to a path. (planned)

Headless use

bedrova serve is enough to use Bedrova as a backend with no windows open — point your OpenAI/Anthropic clients at http://127.0.0.1:1337 exactly as in Integrations. Exposure and auth still follow Settings → Server/API; the CLI does not bypass the LAN-exposure guard.

Exit codes and scripting

Commands return non-zero on failure (model not found, fit-check block, download error) so they compose in scripts. A blocked load reports the same reason the fit-check shows in the UI.

See also

• Server overview — the API the CLI starts.
• Models & downloads — what pull and load operate on.
• Diagnostics & support bundles — what diagnostics collects.