Quickstart: your first day

This is a concrete first day: get oriented, make the workspace yours, talk to your first agent, run your first routine, and dispatch your first mission. It assumes the install is done and Doctor runs green.

One expectation to set before you start: every agent action runs on your Claude subscription, in terminals you can watch. Nothing in this walkthrough happens invisibly.

1. Get your bearings

The window title bar is the global header — branding on the left, the controls for dictation, Voice, and pop-out on the right, and a status strip that follows the active view. Below it, the sidebar is the map:

Three habits worth forming in the first ten minutes:

• Ctrl/Cmd-K opens the palette — search documents, graph entities, workspaces, and memory from one box, and jump straight to the result.
• Any view pops out to its own window with the header's pop-out button — arrange the desk across monitors; the layout survives a restart.
• Start on the Dashboard. It shows platform health at a glance: which background services are up, what the agents have been doing, what changed on disk.

Depending on your tier, some views show a lock with an upgrade note instead of their content — that is the tier gate, not a fault.

2. Make the workspace yours

Open Documents. This is your Command Center home — the folder you chose during setup, as a browsable, editable tree. It starts as a clean scaffold with a worked example; everything you and your agents produce lands here as plain files.

Then open Identity and fill in the company-identity templates: who you are, what you do, how you sound. Agents read these — ten minutes here makes every future output sound like you instead of like a template.

3. See your knowledge graph

Open Cortex. The root graph mirrors your documents folder — every folder and file a node, kept in sync as the workspace changes. This costs nothing: the default graph pipeline runs without any extra API key or token spend.

There is not much to look at yet, and that is the honest state of a fresh install — the graph grows as your workspace does. When you want more depth, the knowledge graph explains the three depths and when each is worth it.

4. Talk to your first agent

Open Terminals and create a workspace, then add an agent pane. A full Claude Code session starts inside Command Center, working in your home folder with the platform's memory, graph, and tool servers already on hand.

Ask it something real about the system it is sitting in:

Read CLAUDE.md and tell me, in five bullets, how this platform wants
agents to work here.

The answer demonstrates the point of the whole product: the session did not need to be briefed, because the working contract is on disk where every agent reads it. This pane is a regular terminal — persistent, visible, yours to type into at any time.

5. Run your first routine

Open Routines. A routine is a scheduled, repeatable agent job — each run spawns a fresh agent process with a defined task, tracked from start to finish.

Pick a routine and run it once manually rather than waiting for its schedule. Watch the job appear, follow its status, and open the run log when it finishes to read exactly what the agent did. Two honest notes: a first run takes as long as the work it does, and every run spends your Claude usage — routines are real agent work, not canned scripts. The full model: Routines.

6. Dispatch your first mission

Open Missions Board. A mission is a unit of larger work — it gets a folder in your workspace, a dossier, and a card on the Kanban board.

1. Create a mission with a clear goal. Keep the first one modest and concrete — an investigation or a written deliverable beats a refactor.
2. Activate it — activation makes it the working focus; it does not start anything by itself.
3. Dispatch. A commander session opens in Terminals and staffs the work, spawning worker sessions as needed — every one of them a visible pane you can read and type into.

Supervise the first mission deliberately: watch the commander reason, answer when it asks, and use the agent-control pause in Terminals if you want everything to hold — one click freezes all agent terminal control. Anything that would leave the machine still waits for your explicit approval. The coordination model: Missions and the fleet.

Where to go next

• The operator model — the thinking behind the desk you just toured.
• Feature guides — one guide per view.
• Agentic infrastructure — the agents, tool servers, memory, and knowledge graph in depth.