Coming soonOOMU is launching soon. Downloads will open at launch.
← All documentation

Tutorials

Setting Up Custom Assistants

OOMU lets you create specialised assistants; each with its own name, role, model, and thinking depth. A "Research Analyst" assistant that reasons deeply and a "Quick Notes" assistant that answers instantly can live side by side, and you choose which one to talk to.

This tutorial walks you through creating your first assistant from scratch.


Before you start

You only need OOMU installed and launched (see Installation & Your First Launch). You can build an assistant on top of the local engine with no API key at all, or point it at a cloud provider you've configured. Either works.


Where assistants live now. Assistants no longer have their own sidebar door. You meet them where you use them: the agent control at the top of the Chat screen. Open that control and choose Manage agents… to reach the full assistants screen; create, edit, import, and browse templates there.


Step 1; Name your assistant and give it a role

Open the agent control in the Chat header, choose Manage agents…, and create a new assistant. You'll give it two things:

  • A name. This is how OOMU refers to itself in conversation. Pick something that fits its job; *Briefing Desk*, *Contract Reviewer*, *Drafting Assistant*.
  • A role / purpose. A short description of what this assistant is for. OOMU turns this into the assistant's working instructions, so be direct: *"Summarise long documents into three clear bullet points and flag anything that needs a human decision."*

Why the name matters. OOMU binds the assistant to the name and purpose you give it; not to the underlying model. When asked who it is, your assistant answers with the name you chose, not "Claude" or "Gemini." Your assistant is yours.


Step 2; Choose the engine behind it

Each assistant runs on a provider and a model:

  • Local; runs entirely on your Mac. Private, free per-question, and always available offline. This is the default.
  • Cloud; a provider you've configured in the Models area with your own API key. OOMU supports Anthropic, OpenAI, Google Gemini, xAI, Mistral, OpenRouter, and more. Best for the heaviest reasoning.
  • Auto; let OOMU decide per message. It keeps easy work on the free local engine and escalates only the harder prompts to a cloud model you've chosen. See Smart Model Routing.

You can change this at any time. For now, the default local engine is a perfect place to start.


Step 3; Adjust the reasoning setting

This is the most useful dial in OOMU. The reasoning level controls how much effort your assistant puts into thinking before it answers. What you see depends on the engine:

On the local engine, reasoning is a simple switch:

SettingWhat it does
OffFast, direct answers.
OnTurns on the local model's deliberate, step-by-step thinking for harder problems.

On cloud models, reasoning is a graded scale, so you can spend exactly as much effort as a task deserves:

LevelWhat it's good for
OffFast, direct answers. Everyday questions and quick drafts.
LowA little deliberation, still snappy.
MediumThe balanced default; thoughtful without being slow.
HighCareful, multi-step reasoning for harder problems.
XHighDeep analysis where accuracy matters more than speed.
UltraMaximum deliberation for the most demanding work.

Set it to match the task: off for quick correspondence, higher for analysis you'll act on.

OOMU is honest about what each model can do. Not every model supports every level. If you ask for *Ultra* on a model that only supports up to *Medium*, OOMU automatically steps down to the highest level that model genuinely offers; rather than pretending. You always get real reasoning, never a fake setting.


Step 4; Save and manage your assistant list

Save the assistant. It now appears in your list, ready to select whenever you start a conversation.

From the list you can:

  • Switch between assistants for different kinds of work.
  • Edit a saved assistant's name, role, model, or reasoning level at any time.
  • Delete assistants you no longer need.

Your assistants are stored in your private, encrypted workspace on your Mac; nothing about them leaves your machine unless you explicitly use a cloud model.


You've built your first assistant

You now have a named, purpose-built assistant with a thinking depth tuned to your work.

Where to next