← All documentation

Tutorials

Creating an agent

Give OOMU focused roles, instructions, and model choices for repeatable work.

An agent is a saved role. Instead of re-explaining who you want OOMU to be every time (the contract reviewer who cites clause numbers, the researcher who never asserts without a source) you define it once and assign it to a chat.


Where agents live

Open Chat and select the Agents tab across the top of the workspace. Agents are part of the chat workspace, not a separate destination in the sidebar.

You can also jump straight there with ⌘K and typing an agent's name.


Step 1: Identity

Choose New agent. The first panel is "Who this agent is."

FieldWhat it is
NameWhat the agent is called.
Job titleThe role, for example Research lead.
Identity"How this agent should understand its role and perspective."
Avatar colorOne of the colored badges on the agent grid.

Keep the name and role concrete. Contract Reviewer beats Assistant 2.


Step 2: Instructions

Instructions is the working brief: "Describe how this agent should work, decide, and respond." This is where the real substance goes.

Review contracts clause by clause. Cite the exact clause number for every claim. Flag anything that shifts liability. Never assert a legal conclusion without quoting the text it rests on.


Step 3: Personality

Personality shapes how the agent sounds. Start from a template and adjust; the moment you change anything, the template becomes Custom.

Template
Everyday assistant · Researcher · Builder · Strategist · Teacher · Custom

Then tune four things:

  • Communication style: "For example: warm and conversational, or brief and direct."
  • Traits: pick from Friendly, Concise, Professional, Curious, Methodical, Creative, Skeptical, and Supportive.
  • Values: "What should guide this agent's choices?"
  • Boundaries: "What should this agent avoid or leave for you?"

Boundaries are the underused one. An agent that knows to stop and ask rather than guess is more useful than one that is merely fast.


Step 4: Model choice

Under Model choice, pick how this agent reasons:

ChoiceBehavior
AutomaticThe router decides per request, between this Mac and your default cloud model.
On this MacA specific local model. Nothing leaves your machine.
CloudA specific connected service and model.

Mods run only on this Mac. If this agent needs Mods, choose Automatic or On this Mac. A cloud-pinned agent cannot use them, and the editor says so.

Model limits lets you cap two things per agent:

LimitRange
Context budget: how much conversation context the agent may use2,048 – 1,000,000 tokens
Maximum response: the upper limit for each reply256 – 8,192 tokens

Leave both on Model default unless you have a reason.


Step 5: Mods

If you have installed any Mods, choose which this agent may use. All enabled Mods is the simple option: "Newly enabled Mods are available to this agent automatically." Or select specific ones.


Step 6: Test before you trust it

Save the agent, then use the Scratchpad at the bottom of the editor: "Try a real message with this agent before you leave the editor." Send it something representative and read the reply.

Testing requires a saved agent. If the button is disabled, save first.


Using an agent

From the agent card, Start chat opens a new conversation already bound to it. Inside any chat, the Agent control in the header switches the agent, or clears it back to No agent.

Star an agent to favorite it. Favorites sort to the front of the grid.


Importing an agent from Beta 1

Import reads an agent exported from OOMU Beta 1. Because Beta 2 is a different application with a narrower and more honest capability model, some older settings have nowhere to go. OOMU tells you exactly what happened rather than silently dropping it:

Imported {name}. Some settings from the older format weren't carried over. Check the agent before you start.

The report names what came across (name, role, identity, tone, instructions, values, boundaries) and what did not, which can include older response-checking instructions, saved retrieval state, older model and channel preferences, older approval preferences, older repeat-work instructions, and other Beta 1 runtime settings.

Review an imported agent before you rely on it.


Archiving and deleting

Archive takes an agent out of the active grid but keeps it: "Archived agents remain available until you remove them." Restore brings it back. Delete permanently removes it for good.