Installation & first launch

By the end of this page OOMU will be installed, set up with a model that runs on your Mac, and answering questions without sending anything anywhere.

---

Before you start

You do not need an account, a credit card, or an API key. A cloud provider is optional, and you can add one later.

---

Step 1: Install

Download OOMU, open the disk image, drag OOMU to your Applications folder, and open it from there.

OOMU is signed with a Developer ID and notarized by Apple, so macOS opens it without a detour into System Settings.

Install it properly, in Applications. Two features check whether OOMU is running from a real application bundle: background work, which registers a macOS Login Item, and software updates. Running the app from your Downloads folder quietly disables both.

If you used Beta 3

Beta 3 is retired. Install Beta 4 as above: the first time it opens, it imports your Beta 3 agents, chats, projects, workflows, memories, and settings, and copies your saved provider keys into its own Keychain entry, so you do not enter them again. Model files you already have are reused rather than downloaded again. Once that is done you can move the Beta 3 app to the Trash. See What changed since Beta 3.

Beta 4 imports from Beta 3 only. It does not read a Beta 2 installation.

---

Step 2: Accept the license

The first screen is the OOMU Standard License (Version 2.0). The full terms are on screen, not behind a link. Accept and continue moves on; Decline and quit closes the application and stores nothing.

Nothing else in OOMU works until the license is accepted, including model installation.

---

Step 3: Set up your models

The setup screen is titled Set up OOMU, with the subtitle "Your models stay on this Mac. Cloud service keys are stored safely on this Mac." Two rows are required, and a third is optional.

Getting the model files

Each required row names the exact file it expects and offers two buttons:

• Download fetches the file from its official model source, with progress, speed, and time remaining. You can Pause and Resume Download.
• Locate File… opens the standard macOS file picker, if you already have the file on this Mac.

OOMU picks the assistant from your Mac's memory and shows only the one you need. E4B is the larger and more capable of the two, but on an 8 GB Mac it would leave too little memory for everything else, so those Macs get E2B.

However the file arrives, OOMU checks its exact size and SHA-256 against the manifest built into the application. A file that does not match is rejected and the row stays unfinished. That is what the footer means by "OOMU verifies every selected model before using it."

Image understanding is not part of setup. The optional vision file that matches your assistant (gemma-4-E4B-it-mmproj.gguf or gemma-4-E2B-it-mmproj.gguf, about 990 MB either way) lets the local assistant read images you attach, and you can add it at any time in Settings → Models & Providers. If you attach an image without it, OOMU tells you where to get it.

While your models download

The side of the screen offers to set up access now: "You can allow access now, or choose later when you need it." Calendar and Contacts can be requested here; "Mail asks for access when you first use it." A few short tips cycle underneath. None of this is required to finish.

Connecting a cloud service, if you want one

Open Cloud model (optional), pick a service and a model, paste the key, and choose Connect. OOMU makes a real round trip to the provider before accepting the key. If the round trip fails, nothing is saved. Once it succeeds, the key goes into your macOS Keychain and the row shows the provider and model that verified.

Skip this entirely if you would rather stay local. You can add a service later from Connections.

Choose Start Using OOMU once the understanding model and the local assistant are both ready.

Privacy, in one sentence

OOMU sends no telemetry. The in-app notice puts it this way: "OOMU does not send diagnostics, prompt history, or usage metrics as telemetry. Update checks send the app version and processor architecture, without a device or account identifier." The full Application Privacy & Data Notice is in Settings → Storage & Privacy whenever you want to read it.

---

Step 4: Find your way around

OOMU opens on Chat. The sidebar has four destinations, with two more beneath them.

You will also notice Mochi, a small round character on your desktop. It is OOMU outside the window: click it to ask a question without switching apps. Drag it wherever you like, or press ⌘⇧M to hide it. See Working with Mochi.

Three shortcuts are worth learning immediately:

• ⌘K opens the quick switcher, which searches your chats, agents, workflows, and every destination in the app.
• ⌘⌃⇧Space opens the Quick Launcher from anywhere on your Mac, even when OOMU is not in front.
• ⌘⇧M shows or hides Mochi, and ⌘⇧O brings back the main OOMU window.

---

Step 5: Ask your first question

The empty chat asks "What would you like to do?" and offers three starting points:

• Weekly Brief: summarize recent activity and schedule, from your Apple Calendar.
• Summarize Folder: pick a local folder and analyze it on this Mac.
• Help with Email: triage unread Apple Mail and suggest replies. It does not send anything.

Or type anything into the composer, which checks your spelling as you type. Watch the header while it works: "Understanding your request…" is the router deciding what kind of work this is. Simple questions go straight to an answer, which streams in as it is written.

The Model control above the composer offers three choices:

New chats start on Automatic. Automatic uses a cloud model only if Use cloud models in Automatic is turned on in Settings, and it is off until you turn it on. If you skipped the cloud service, everything runs locally regardless.

While a reply is in flight you are not locked out. Steer injects a correction into the running turn, Queue next lines up the following task, and Stop cancels.

Under every prompt and reply there is a Copy button, and Export as Markdown saves the whole conversation as a Markdown file you can open in Notes, Obsidian, or anything else.

---

Step 6: Tell OOMU who you are

Open your Profile from the button at the top right of Chat. It is worth ten minutes. Your role, expertise, priorities, preferred tone, response length, and boundaries all become standing context, so you stop repeating yourself. The panel puts it plainly: "The more OOMU knows about your work, the less you'll have to repeat yourself."

The same screen holds Memory: the things you have explicitly asked OOMU to remember, each labeled with its scope (This project or Everywhere) and a Forget button.

It also holds Standing memory: "Useful facts, kept privately on this Mac." Update overnight is on from the start. While it is, OOMU reads what you wrote in recent chats with its local model and keeps the facts worth keeping, sorted into People, Projects, and Preferences. It does this between 2 and 5 in the morning, only while your Mac is plugged in, idle, and at a normal temperature. Update now runs it straight away, and also needs the Mac to be plugged in.

• Each fact shows the exact words it came from, under Source turn, and Open source chat takes you to that conversation.
• Edit corrects a fact. Forget removes it along with the quoted words, and "Future synthesis will not restore this entry." Deleting a chat does not remove facts that came from it; Forget does.
• Memory activity shows each update: how many sources it read, how many facts it kept, and how many it left out.
• Anything that looks like a password, a key, or a long encoded string is left out, along with the rest of the text around it. That is cautious on purpose, so it sometimes skips a harmless fact.

Standing memory is never sent to a cloud model. When a request goes to the cloud, these facts stay behind.

---

What you have now

• OOMU installed in Applications, license accepted
• A verified understanding model and local assistant
• An encrypted database on this Mac, keyed to your Keychain
• A first conversation underway

Where to go next

• Working with Mochi: the desktop companion, and what it can watch for you.
• Setting up a project: the right way to keep a case or client together.
• Working fast: the Quick Launcher, @ tags, quick actions, and the quick switcher.
• How OOMU picks a model: what the understanding model is actually deciding.
• Connecting services: Apple apps, Google Workspace, and more.