Using Agentum

Agentum’s window is a command center for the agents running on your Mac. This page walks through each surface and the things you’ll do most often. If you haven’t installed yet, start with Getting Started.

The shell at a glance

  • Sidebar (left) — navigation: chat, your agents, the Marketplace, system settings. Collapsible to a slim icon rail.
  • Main panel (center) — whatever you selected: Agentum Chat by default, or an agent’s page, the Marketplace, the Create flow, the System view.
  • Sidekick rail (right, optional) — a companion panel that appears on agent pages when you toggle it; see Improving Agents.

Agentum Chat — the home screen

The app opens into Agentum Chat: one conversation surface for everything your agents want to tell you and everything you want to ask them.

  • The pinned ✨ Agentum thread is the delivery channel. Agents post here proactively — updates, reminders, questions — each message labeled with the agent it came from. When an agent asks something with fixed options, the options arrive as real buttons you tap; otherwise just type a reply. This is also what drives the unread badge on the sidebar and macOS notifications.
  • The composer supports slash commands with autocomplete: type / to see built-ins (/help, /model, …) plus every command your installed agents registered.
  • If your AI isn’t configured yet, the chat shows a Set up your AI banner instead of failing — one click to the LLM settings.

Chat isn’t only for reading what your agents say — you can run the app from it. Ask it to change an agent, install one, turn one off, or tell you what an agent is set to, and it does the work (asking you to tap Approve first for anything that changes something). See Running Agentum from chat.

The same conversation is available in your browser and on your phone via Agentum Web Chat.

The sidebar

Top to bottom:

  • Search filters your agent list as you type.
  • Nav rows, under a Navigation heading:
    • Today — everything waiting on a decision from you, sorted into the bands of a day. Off by default; see below for how to switch it on.
    • Create Agent — the guided build flow (Creating Agents). A dot means a build is in progress.
    • Marketplace — browse and install. A chip appears when updates are waiting.
    • Routines — prompts your assistant runs on a schedule. See below.
    • Knowledge — pages your agents keep about your world. Only appears once there’s something in it. See below.
  • Favorites — star any agent to pin it here; drag to reorder.
  • Your sections — group agents however you like: create sections, rename them, drag agents between them, collapse the ones you don’t need today. Applications with sub-agents nest as expandable trees.
  • Integrations, Tools and Paused Agents — three auto-managed, collapsed-by-default groups. Integrations are the connectors to outside services (Google, and so on). Tools are small building blocks that ship with the app and that your other agents use — you rarely open one directly. Paused Agents is anything you’ve switched off.
  • Footer — the Settings gear, a ? for sending feedback, a health badge that appears when agents are erroring (click it to jump to the problem), and your Account button (license status, trial countdown, cancel/reactivate).

An agent’s page

Click any agent to open its page. The header shows its name, a provenance pill — Editable (yours) or Marketplace (managed) — and status banners when something needs attention (disabled, failing, recent errors with an expandable detail ring).

The toolbar tabs:

Tab What it is
Main The agent’s own dashboard — its live panel, stats, and its own settings controls.
Guide The agent’s documentation: a rendered guide (some agents ship full multi-page HTML docs) plus a setup checklist. A dot on the tab means unfinished setup steps.
Improve The AI editing surface — your agents only; see Improving Agents.
Sidekick Toggles the companion rail — ask questions about this agent in plain English.

Everything else lives in the ⋯ menu: the Enable/Disable switch, Integrations (which services this agent may touch — see below), Advanced (add-on capabilities and cross-agent data access), Share Agent (export, your agents only), and Uninstall.

How integration grants work

When an agent is installed, Agentum auto-grants the integrations it declares (AI, Calendar, chat, …). Open ⋯ → Integrations to see them as toggles. Revoking one doesn’t break the agent — a well-built agent keeps running with reduced capability, and any blocked call gets a clear denial. Flip it back on and access resumes immediately, no restart. The same panel governs which other agents’ data this agent may read.

Enabling and disabling whole agents is just as light-weight — the change applies live to the running system, no restart needed.

Today, Routines, and Knowledge

Three surfaces that aren’t agents but sit alongside them.

Today — what’s waiting on you

Today collects everything that needs a decision from you and lays it out as the shape of a day: Now, Later today, This week, and a tray for anything carried over from before. Each row says what it is and which agent raised it, and carries its own approve and defer buttons — so you can clear the pile without opening anything.

When the pile is big, start the focus run: it takes over the screen and walks you through the items one at a time.

Today is off by default. Turn it on in Settings → Connections → Interface — the toggle there is still labelled Dashboard, which is what this page used to be called — and a Today row appears at the top of the Navigation group.

Routines — prompts on a schedule

A routine is a saved instruction your assistant carries out on a timetable — “every weekday at 7, give me my calendar and anything that came in overnight.”

You normally create routines just by asking in chat, in exactly those words. The Routines page is where you see them all afterwards: when each one next runs, and buttons to edit, pause, or delete it. Pausing is a toggle, not a deletion, so a routine you only want during the week is one click away from coming back.

Routines are different from an agent’s own schedule: a routine is your prompt, run by your assistant, and you can change it in plain English at any time.

Knowledge — what your agents remember about your world

Your agents build up a small library of plain-language pages — a summary of your week, notes on the organizations and topics that keep coming up, pages about people you deal with. Knowledge is the reader for that library: pick a page on the left, read it on the right.

It’s read-only in the app on purpose — each page has exactly one agent responsible for keeping it accurate, and a second editor would fight with it. If you want to edit the prose yourself, the header’s Open folder button reveals the real files, which are ordinary Markdown and safe to open in a notes app like Obsidian.

The row only appears in the sidebar once there’s at least one page, so a brand-new install won’t show it.

The Marketplace

Click Marketplace in the sidebar. Four tabs:

  • All — the full catalog: curated groups (stacks of agents that work together — install a group and its members are set up and granted in one shot) plus a searchable grid.
  • Featured — the editorial picks.
  • Installed — what you have, split into Active and Disabled, each card with an Update pill when a new version is out.
  • Updates — everything with a pending update, plus Update all. Routine updates apply in one click; any update that requests new permissions is set aside for your explicit review — you’ll see exactly what’s being added before you approve.

Installing opens a drawer with the agent’s description, screenshots, and what it needs; one click and it’s live in your sidebar. Uninstalling preserves the agent’s data for one round of reinstall, so a change of heart costs nothing.

Marketplace agents are managed: Agentum keeps them updated and their code sealed read-only (no Improve tab, no export). Your own creations are fully editable. Both live side by side in the sidebar with their provenance pill.

Settings

The gear in the sidebar footer opens Settings. Three tabs across the top: Status, Connections, Advanced.

Status

The runtime view. The Agent card shows the background process that runs your agents — Running (with uptime), Starting, Restarting, Crashed, or Stopped — with Stop and Restart buttons. That one restart button is the fix for most “my agent is acting weird” moments: a few seconds, no app relaunch.

Below it: the Agents folder card (where your agents live on disk — ~/Documents/Agentum/agents by default — with change and reveal-in-Finder controls) and the Software update card (current version, check/download/install).

macOS note: if your agents folder sits somewhere macOS protects, the agent card walks you through granting access in Privacy Settings — a restart alone can’t fix a permissions block.

Connections

Everything Agentum talks to, one card per concern:

  • Connections at a glance — a single card summarizing what’s connected and what isn’t, so you can see the whole picture before scrolling.
  • Profile — your name, your timezone, and your Google account (Gmail, Calendar, Drive, Docs, Tasks behind one Connect button).
  • Sidekick and Sidekick memory — the name and personality of your assistant, and what it’s allowed to remember about you. The memory card lists what it has picked up, by category, and lets you clear it.
  • LLM Provider — the AI behind your agents and the coding assistant that builds them (Choosing Your LLM & Coding Harness). This card also holds the daily credit cap per agent — the safety limit on what any one agent can spend in a day.
  • Voice transcription — turning spoken notes into text.
  • Background worker — an optional helper that runs a model on your own Mac for quiet, unmetered background work. Off unless you turn it on.
  • Interface — appearance options, including switching on Today.
  • End of day — whether Agentum gives you an evening wrap-up, and when.
  • Remote Access — the private tunnel that lets things reach your Mac from outside (Remote Access).
  • Web Chat and Paired devices — browser and phone chat, and the devices you’ve approved (Agentum Web Chat).
  • Public Links — private web addresses an agent can hand to an outside service. Each one needs your approval.
  • Chrome automation, Typing in your browser, and MCP servers — the optional browser extension, which sites an agent is allowed to type into, and any external tool servers you’ve connected.

Cards write through to the running system when you save; anything that needs an agent restart raises a banner with the button right there. There is no configuration file to hand-edit — if you find an old guide telling you to edit .env, it’s obsolete.

Advanced

Three inner views:

  • Health — every agent currently erroring or degraded, with the latest error and a jump-to-agent link. The sidebar’s health badge deep-links here.
  • Schedule — every recurring job your agents have registered, laid out as a day in your own timezone with agents named in plain English. Your first stop when “my agent didn’t run this morning” — a stale last-run or a red error tells you exactly which job to look at.
  • General — agent-data backup, terminal appearance, the authoring kit export, and the Danger Zone (re-run onboarding, reset).

Your account

The Account button in the sidebar footer shows your license at a glance — Trial — N days left or Lifetime — and holds the account actions: cancel trial, reactivate, refresh license. Details in Account & Billing.

Where to next