Setup guide download, sign in, host. about 5 minutes.
Umbra turns an idle Apple Silicon Mac into a private inference provider. The CLI does the work on your machine: it signs you in, registers the device, opens the required enrollment profile, then pulls and serves a model. The platform sets every price, so you just pick what is worth running.
Before downloading, read the public publisher, system-change, and uninstall disclosure.
Before you install 30-second hardware check
Install the verified CLI macOS 14+ · Apple Silicon
The supported installer downloads the notarized DMG, verifies its SHA-256, Developer ID, nested signatures, entitlements, and Gatekeeper ticket, then installs Umbra Provider.app and the umbra command. On a first install it immediately starts guided setup in the same Terminal.
curl -fsSL https://tryumbra.dev/install.sh | shApprove sign-in
The installer opens the browser device-code flow. Sign in or create an account, approve the code, and the CLI registers this Mac under your account.
You can also open Link a device here and enter the displayed code. Hosting a device means you agree to the Provider Agreement.
Approve attestation enrollment one-time · CLI-driven
The CLI downloads and opens the signed Umbra enrollment profile. macOS requires you to review and install it in System Settings, Device Management; the command waits and verifies that Umbra is the active MDM server.
3LHSL95J9H and the stapled notarization ticket before copying the app.Host a model
Setup lists models that fit your Mac. Pick one and the CLI downloads it, verifies the GGUF digest, and starts the managed host. Run the embedded CLI with status afterward to confirm admission and the verified trust level.
umbra statusLet a coding agent do it paste one prompt
On the Mac you want to host from, paste this into a coding agent (Claude Code, Cursor, Copilot) and it walks through install, sign-in, attestation enrollment, hosting, and admission status.
view prompt
Set up this Mac as an Umbra inference provider so it can serve LLM requests and earn. Umbra runs llama.cpp in-process and never logs prompts. Requirements: Apple Silicon (M-series), macOS 14 or newer, and enough free unified memory for the model.
First open this page and show me the publisher identity, system changes, and removal instructions before doing anything:
https://tryumbra.dev/download/
After I review those disclosures, run the supported installer. Do not bypass any checksum, signature, or Gatekeeper check:
curl -fsSL https://tryumbra.dev/install.sh | sh
On a first install, the installer starts guided setup automatically. It opens browser sign-in, opens the signed Umbra enrollment profile, waits for me to approve it in System Settings, then lists fitting models and starts the model I choose. If the shell is not interactive, run the exact setup command the installer prints. After setup, run `umbra status` (or the exact CLI path it prints) to confirm coordinator admission and trust.
Notes:
- My Hugging Face key is used only to pull models and stays on this machine.
- Cash-out eligibility is described in the provider agreement: https://tryumbra.dev/provider-agreement/
- Manage everything at https://tryumbra.dev/console.Uninstall inspect, confirm, remove
The uninstall script stops and removes the per-user LaunchAgent, app, official CLI wrapper, and local ~/.umbra data after showing the plan and asking for confirmation. It deliberately does not remove the Umbra Device Management profile: review and remove that yourself in System Settings, General, Device Management.
curl -fsSLo /tmp/umbra-uninstall.sh https://tryumbra.dev/uninstall.sh
less /tmp/umbra-uninstall.sh
sh /tmp/umbra-uninstall.shSee the complete install and removal disclosure on Download & uninstall.
Troubleshooting what the installer prints, and what to do
Intel Mac. Umbra runs inference in-process on the Apple Silicon GPU and cannot serve on an Intel Mac. There is no workaround; use an M1 or newer Mac to host.
Not macOS. The provider is a signed macOS app bundle and only installs on macOS. Run the command on a Mac.
Download unavailable. During alpha the notarized build is not always published. This is expected. Try again later, or paste the provider prompt above into a coding agent to retry. Nothing was installed.
Notarization / Gatekeeper. The installer verifies the download's checksum, Umbra Developer ID (Team 3LHSL95J9H), and Apple notarization ticket before copying anything, and stops if any check fails. Do not attempt to bypass Gatekeeper or remove the quarantine flag by hand: a failing check means the artifact is not the trusted build. Re-run the command on a normal network; if it keeps failing, the published build is being re-notarized, so try again later.
Integrity check failed. The download did not match the expected checksum or signature, so it was discarded. A flaky proxy or captive-portal network is the usual cause. Re-run on a direct connection. The installer never installs an artifact that fails these checks.
Permissions. The standard install asks for administrator approval so umbra is available globally from /usr/local/bin. A normal Terminal uses the hidden sudo prompt; an agent shell without a TTY opens a native hidden macOS authorization dialog. If admin access is unavailable, export writable locations before the pipe: export UMBRA_INSTALL_DIR="$HOME/Applications" UMBRA_BIN_DIR="$HOME/bin"; curl -fsSL https://tryumbra.dev/install.sh | sh. The installer starts setup through its exact CLI path and prints the line to add to ~/.zprofile.
What would your Mac earn?
Earnings depend on your chip, unified memory, and how much demand a model sees. The calculator shows what a Mac like yours could earn before you install a thing, with the platform price and projected payout per model.
Figures are projections. Connect payout from the provider wallet when eligible; see the provider agreement.