Aivyx PA · ManualContents ↓

Create your agent

This is the guided setup that gives your assistant its identity. You can run it in the terminal (aivyx-pa init) or in the Studio’s Create screen — both do the same thing and write to the same place.

It has three steps: Profile → Persona seed → Access.

Step 1 — Profile (who your assistant is)

The Profile is the identity you declare. It has six parts:

  1. Name — what you’ll call your assistant.
  2. About you — a little about you, so it can tailor how it helps.
  3. How it talks — tone and style (warm and chatty, terse and technical, …).
  4. What it’s for — the kinds of tasks you want help with.
  5. What it tends to do — default behaviors and habits you want.
  6. Lines it must never cross — hard boundaries it will always respect.

You can fill these in two ways:

  • Let the model draft it. Answer a few questions about the relationship you want, and your chosen model writes a full first draft. You then review and edit every line — you’re always the final author.
  • Write it yourself. The same six fields with helpful prompts, fully offline. No model required.

It ends with a friendly “meet your assistant” preview you can confirm, edit, or restart.

The Profile is who your assistant starts as. Its deeper character — the Persona — is earned over time as it learns from working with you. You shape the starting point here; you don’t have to get it perfect.

Step 2 — Persona seed (its starting character)

Optionally give your assistant a short starting personality and a few starter skills — a “seed” for the character it will grow into. You can describe it in your own words, or let the model draft one. Leave it blank and your assistant simply starts as a clean slate and learns from there.

This seed is planted once, at first launch, and is fully auditable and reversible later from the Agents screen.

Step 3 — Access (how far it can reach)

Finally, choose how much of your machine the assistant can touch:

  • Sandbox (the default) — it can only read and write inside one dedicated folder, ~/aivyx-pa-sandbox. Safest; good for trying things out.
  • Workspace — one project folder you choose. (Terminal setup only; in the Studio, pick it later on Settings.)
  • Home — it can reach your home directory (documents, projects, …).
  • Full — the whole filesystem, system files included.

Anything beyond the sandbox shows a warning and asks you to confirm.

Whatever you pick, irreversible actions always stop for your confirmation — deleting or overwriting a file, sending an email, spending money. Access level sets the reach; the confirmation step is your safety net on top.

You can change the level any time later (the Settings screen, or aivyx-pa access).

After setup

That’s it — your assistant is ready. Everything you chose here is editable later:

  • Edit the Profile from the Agents screen.
  • Review and approve the character changes your assistant proposes as it learns — also on Agents.
  • Change access and budgets on Settings.