Early access · Paid API credit is live. New accounts still get free starter credit.
← Back to Umbra

Host a model

browse models, see what you would earn, then host from the CLI
Sign in
CLI hosts

Hosting happens in the CLI, not on the web. Review and install the signed app from the download page, then each row below carries the exact umbra host command to run.Pick your hardware, see which models earn the most on it, and host one from the CLI. Pulling, verifying, and serving a model is an on-device job, so this page has no host button. Models you add via the CLI then appear in Models you host. Your device must be online for it to serve and for its state to show as live.

ProjectionThese figures are projections. Connect a payout account from the provider wallet when eligible; see the provider agreement.The floor is model-aware base pay at the 1.0× reference reliability band once your device is attested and online (low uptime scales down to 0.6×; sustained uptime up to 1.2×). The upside is variable on-demand pay (your 90% share per token served). Earnings accrue as withdrawable and become payable once you connect a payout account when eligible. Full terms are in the provider agreement.

Your hardware



Hours per dayprojected at 24h/day
$0$0.50
Subtracted from your projected earnings in the breakdown below (and applied to any model you price in the paste box). It won't re-sort the recommendations — every model on your chip draws the same power.

Earnings breakdown The floor is model-aware base pay at the 1.0× reference reliability band once a device is attested and online (scales with committed unified memory; see the reliability ladder). Your local electricity cost is then subtracted to show projected take-home. The upside is a projection of variable on-demand pay (per output token you generate, your share after competition); its demand basis is stated above. Projected provider share is 90%, matching current coordinator settlement. Figures are live from the coordinator; if it is unreachable this panel shows an unavailable state rather than a guess.

Select hardware to see the top pick.

Recommended to host Operator-reviewed creator demand is promoted first when it matches a compatible ranked candidate. The remaining list is ranked by expected net per month on your hardware: base-pay floor plus scarcity-adjusted on-demand upside. A creator request never bypasses your choice or the normal public-weight, license, architecture, GGUF, integrity, and onboarding checks.

Don't see the model you want? Browse the full catalog or request model hosting. Approved creator requests become reviewed demand signals for providers.

Paste any Hugging Face GGUF link

Price any GGUF model on the hardware above.
Paste a Hugging Face repo URL or an owner/name id. It must ship .gguf files (llama.cpp serves GGUF only); a non-GGUF repo is reported clearly, never silently priced.

Report a bug

Tell us what went wrong. We'll include the page you're on and your browser — no prompts or content.