Quick start

You installed Cadence on Linux. This page is what to do after the window opens.

Canonical URL: /docs/getting-started/quick-start

Your assistant is ready

When hatch finishes, the assistant is a blank slate. No name, no memories, no preferences yet. There is no onboarding wizard. There is a chat box. Type something.

The more you work together, the more context it builds. You name it. There is no seeded house persona.

Where to talk to it

• Linux desktop. The AppImage or apt package from www.aevumcadence.ai.
• CLI. The cadence command on PATH after first launch.

Memory and conversation history stay on that machine. Browser, phone, and other desktops are later. You can hook up channels (Telegram, Slack, email, phone) so the same local assistant reaches you outside the app. See Channels.

What to say first

There is no wrong answer. A few good first moves:

• Give it a name. "Your name is Bob." That name sticks (IDENTITY.md).
• Tell it about yourself. "I'm Alice. I work in marketing, live in Brooklyn, and prefer short, direct answers."
• Ask what it can do. "What tools do you have?" It should look at live skills, not guess. Catalog pages: Skills and tools, Features, Skills reference.
• Shape its style. "Be more casual, less corporate. Don't hedge." Style notes persist (SOUL.md).

None of this is required. You can also dive into a real task.

Give it a real task

Something you actually need works better than a toy prompt:

• "Draft a follow-up email to Alice about the launch meeting we had this morning."
• "Help me think through whether to take the new role."
• "Summarize what's on my calendar this week."
• "Remind me at 4pm to drink water."

If a task needs a connection that is not set up yet (calendar, mail), the assistant should tell you what it needs and walk you through it. Secrets go through the credential prompt, never pasted into chat.

Approvals

The first time the assistant tries something that could matter (sending mail, writing a file outside the workspace, running a command), it stops and asks. You see the action. You Allow or Deny.

You can add trust rules so it stops asking about things you have already approved. See The permissions model.

What it remembers

Facts that are likely to matter later can land as memories and surface in later conversations. Ask "What do you remember about me?" Tell it to forget something: "Forget what I told you about that project."

Details: Memory and context.

Where to go next

• Your first skill: teach it a workflow that should keep working.
• The workspace: IDENTITY.md, SOUL.md, USER.md on disk.
• FAQ for first-week questions.
• Common issues if something does not start.