All documentation · Insights & assistant

The Clyentra AI assistant

Open the Clyentra operator with ⌘J anywhere in the app. It reads your workspace, answers questions, draws charts, and creates records — but never writes anything without your approval.

10 min read · Reviewed 2026-07-13

Clyentra ships an assistant that does more than answer questions about the product — it can read your actual workspace and change it. Because it can change it, the central design decision is this: it never writes anything without asking you first. Everything else in this guide follows from that.

There are two surfaces. The operator is the one you will use — a panel that opens over whatever page you are on. The Agentic AI page is a full-page chat thread. Both are covered below.

Opening the operator

Three ways in, and the first is the one to learn:

• ⌘J — or Ctrl+J on Windows. Works anywhere in the app.
• The Support Chat button in the header.
• The operator card on the Overview dashboard — press Ask Clyentra, or click one of its suggested questions to open the panel with that question already sent.

The panel header reads Clyentra · operator with the subtitle "workspace assistant". Its tagline states the contract in six words: "ask · analyse · act — confirms before any write".

It knows what page you're on

The operator is page-aware. Open it while you are looking at a lead and the subtitle changes to context: lead, and the starter prompts change with it — "Summarise this lead", "Draft a follow-up task for this lead". Open it on a task and you get task starters. You do not have to describe what you are looking at; it is already looking at it with you.

With no particular context, the default starters under Try asking are:

• "How many open leads do I have?"
• "What tasks are due today?"
• "Show this month's leads by status as a chart"
• "How do I connect my Vercel account?"

That fourth one is the tell: the assistant can search the help documentation as well as your data, so "how do I…" questions are fair game alongside "how many…".

What it can read

Reads are free of consequence, so they happen without ceremony. The operator can look up:

As it works, it streams its steps into an activity rail — Searching leads, Checking calendar, Crunching analytics, Searching docs. You watch it think rather than staring at a spinner, and when the answer arrives you know what it was built from. Ask for a chart and it renders one inline.

What it can write — and how it asks

The operator can create and update records. Specifically: create or update a task, create or update a lead, create a calendar event, create or update a category, and create or update a product.

Nothing is written until you press Approve: Every single write stops at a Confirm write card. The card names the action, states what it is about to do, and lists the exact fields it intends to set — title, status, priority, due date, whatever applies. Two buttons: Approve and Dismiss. Until you press Approve, nothing has happened to your data.

1. Ask for something that changes data. "Create a follow-up task for the Acme lead, due Friday, high priority."
2. Read the Confirm write card. It shows the action (create task), a one-line summary, and a field-by-field table of what will be set. This is the moment to check the date is the Friday you meant.
3. Approve or dismiss. Approve performs the write and the activity rail records it (Created task). Dismiss cancels it, and the thread confirms in plain language: Dismissed "{summary}" — no changes were made.

Dismissing is genuinely free. No partial write, no draft record left behind, nothing to undo. If the assistant misunderstood you, dismiss it, correct it, and let it try again.

It acts only with your permissions

The assistant is not a back door. It runs inside your permissions, exactly as you do. Ask it to touch something your role cannot reach and the activity rail says so — — no access to {resource} — rather than quietly doing it. The panel footer repeats the promise: "acts only with your permissions".

Two guarantees, worth stating together: The assistant cannot do anything you are not allowed to do, and it will not do anything you have not approved. Giving a teammate the assistant does not widen what that teammate can change — see roles-and-permissions.

Using the composer

The composer placeholder is "Ask anything about your workspace…". The key hints sit under it: ⏎ send · ⇧⏎ newline. While it is working you can stop it — "generating — press stop to cancel". Rate any answer with 👍 or 👎, copy it, or press New chat to start over. The panel is resizable: drag its left edge if you want more room for a chart.

The assistant does not cost AI credits: Asking it questions and approving its writes are free. Credits are consumed by other AI features — image and video generation, and the opt-in deep read on a lead. See ai-credits.

The Agentic AI page

The second surface is a full page at /home/agentic-ai — a persistent chat thread rather than a panel over your work. It carries an Agentic AI badge and a model toggle with Fast (active) and Pro (marked Soon — pressing it tells you "Pro model coming soon!").

It opens on a welcome screen — "Good to see you" / "Ready to assist you with your tasks today" — with three quick actions: Add Lead ("Help me add a new lead"), Search Lead ("Search for a lead") and Generate Report ("Generate a report for me"). The composer placeholder is "Message Agentic AI".

The thread persists across sessions, and you can wipe it with Clear Chat History. Clyentra asks first, because it means it: "Are you sure you want to clear AI chat history? This action cannot be undone."

Verify what matters: The page carries the standard disclaimer, and it is not boilerplate: "AI can make mistakes. Please verify important information." The Confirm write card exists precisely so that a mistaken instruction costs you a glance rather than a corrupted record. Read the card before you approve it.

Which one should you use?

In practice: reach for ⌘J. It is one keystroke away from wherever you already are, and it already knows what you are looking at.

What it isn't

The assistant is not a contact in your chat list — it does not appear alongside your DMs and groups in team-chat. And it is a different thing from the AI Insights button, which analyses whatever data is on screen and returns a summary with Key Insights and Recommendations. Insights explains a screen; the operator answers a question and can act on the answer. Both are free. See business-dashboards.