--- title: "Need a hand? (the help assistant)" description: "The built-in help chat: what it can answer, which model it uses, what it can and can't do, and how your conversation is kept." slug: guides/help-assistant section: guides canonical_url: https://cerevisor.com/docs/guides/help-assistant last_verified: 2026-09-27 last_verified_version: "3.6.0" updated_at: 2026-10-01T14:38:17.486149+00:00 --- Every plan includes a small help assistant, always one click away. ## Opening it Click **Need a hand?** at the bottom-right of the window. A panel opens titled "Cerevisor help" with a box to type a question. ## What it can answer - **How Cerevisor works.** It reads from the same help pages you're reading right now and tells you what it found, or says plainly when a question isn't covered. - **Your own worlds and workflows.** The one on screen, any others open or parked, and worlds you've used recently — including what's in them: agents, connections, settings. - **A workflow's last run.** When it ran, how long it took, what it cost, and how each agent did. - **You, through your memory** — but only when you've asked it something about yourself or how you like to work. It doesn't hold any of this in its head ahead of time. It looks things up only when your question actually needs it, the same way you'd search a manual or open a file — not by having everything memorized from the start. ## What it can't do It can look things up and take you places, but it never changes anything. It can't edit a workflow, save a file, close a world, or start a run. If you ask for a change, it tells you how to make it yourself. ## Getting you there Ask it to take you somewhere and it will: a different view (Analytics, the Skill Workshop, Memory, your Organization), another one of your worlds, or a specific workflow. It just changes what's on screen — nothing more. ## Which model answers It uses **your own default model connection** — the same one your other work runs on. It's never the chat builder's own model choice, and never your pinned Background model. If your default is a subscription sign-in (Claude Code, Codex, Cursor, Antigravity, or Grok Build), answers come a little slower: each step it takes with a tool starts that harness fresh. It still ends every turn with a real answer. If you haven't connected a model yet, the panel tells you so and offers a **Connect a model** button that opens the Provider Hub. ## Citations When an answer is based on a help page, it names the page. If that page hasn't been checked since an older version of Cerevisor, the citation says which version it was last checked against — so you know to double-check anything version-specific. ## Your conversation The conversation is kept across restarts, so you can pick up where you left off. **Clear conversation** (in the panel's header) deletes it for good. There's one conversation for the whole app, not one per workflow or per world. ## What is sent, and where Asking a question sends it to your own model provider, the one you connected — the same place your workflows already send their work. When a question needs it, what the assistant looks up to answer goes to that same provider too: a help page, a description of one of your workflows, a summary of a workflow's last run, or a part of your memory. It's sent only for the question that needs it, and only to that provider — nothing goes anywhere else, and nothing about the help chat is sent to Cerevisor itself. Your worlds, workflows, run history and memory are stored on your device; see [Data inventory & consent](./memory-and-learning/data-inventory-and-consent.md) for the complete picture of what's stored and where. ## Still want a person? Click the envelope button (**Send an e-mail to Cerevisor**) inside the same panel to write directly to us — it opens a draft in your email app, same as before. You choose whether to send it. Or click the speech-bubble button next to it (**Send feedback form on Cerevisor website**) to open the feedback form on cerevisor.com in your browser.