AI assistant

Press ⌘I (Ctrl+I on Windows) anywhere in the app, or open Assistant in the sidebar. Ask a question about your pipelines, buyers or leads in plain English and it answers from your own data and this documentation.

It can also make changes for you — never on its own. Ask for one and it shows you exactly what would happen, with a button. Nothing changes until you press it.

What it's good at

Diagnosing one lead. "Why wasn't LD-A7X2K9 delivered?" gets you the actual routing trace: which rules were tested, which buyers were skipped and for what reason, what the score was and where the price came from. This is the thing it does best — the answer is spread across rules, caps and operating hours, and it reads all of them at once.

Explaining a number you're looking at. Several screens carry an Explain button — the unrouted counter on the dashboard, the routing trace on a lead, the intake figures on a pipeline's sources. It sends the question for you, with the context of what's on screen.

Answering product questions. "How do buyer caps work?", "What's the difference between held and deferred?" It answers from this documentation and links the page it used, so you can check it.

Finding things. "Which buyers are close to their cap?", "Which pipelines have no active buyer?", "Do I have unrouted leads, and why?"

Every answer that used your data or a doc page carries a Sources disclosure. Open it to see exactly what was read.

Making changes from the assistant

Ask for a change in plain English — pause Acme, they haven't paid, raise Bright Solar's daily cap to 80, approve LD-A7X2K9 — and the assistant does not make it. It works out what would happen and puts a card in the conversation:

• what would change, field by field, exactly as the activity log will show it afterwards
• what it would mean for your volume — would have matched 114 of 380 leads in the last 30 days
• anything worth knowing first, such as a rule left pointing at a buyer you are about to pause
• whether it can be undone, and for how long
• Apply and Discard

Nothing changes until you apply. Apply and Revert are buttons on the card: typing go or undo that in the chat does not apply or undo anything, and the assistant points you to the button instead of proposing the card again. It has no way to apply a card itself, so it never tells you a change is done.

One card at a time. Ask for two changes and it proposes the first, then the second when you answer. Somebody approving a change should have read it, and two cards stacked up is how the second one goes unread.

Applied, and still reversible. Once you apply, the card turns into Applied, with a link to the activity log and a Revert button — the same undo you get on the activity row, for the same seven days. A change that cannot be undone says so on the card, before you apply it. A change that emails the workspace owners, like a new delivery endpoint or a wallet payment, says that under its title too.

The page updates by itself. Apply a card while a pipeline, a buyer or your leads are open behind the panel and they refresh on the spot: the rule that changed is briefly outlined on the pipeline, and a buyer's status, caps and banner move without a reload. A Revert does the same in the other direction. Other open tabs catch up when you reload them.

A card goes stale. It describes your account as it was ten minutes ago, so past that it expires and you ask again. If somebody edits the same buyer in between, applying is refused and the card offers Refresh preview rather than writing something you never saw.

A buyer already routed is changed, not added again. Ask Acme should only get grade A and B leads when Acme already has a rule in that pipeline, and the card changes that rule's conditions rather than adding a second rule next to it. A second rule for the same buyer in the same pipeline (another tier, another price, another segment of leads) is only proposed when you ask for one in so many words: add a second tier-2 rule for Acme at $10, keep the current one.

Conditions are spelled out. The assistant writes a rule's conditions as field, operator and value (qualityGrade is one of A, B), stored exactly as if you had built them in the pipeline editor. A condition on a field the pipeline does not collect is flagged on the card, with the closest field name when there is one. Changing a rule's conditions replaces the whole set, so the card lists any condition the change would drop before you apply it.

Cards stay in the conversation. Reopen the thread a day later and you see what you approved, what you discarded, and what is still reversible.

What it can propose, and what it cannot

Exactly what your own seat could do by hand. An admin can propose buyer, pipeline, rule and endpoint changes and act on leads; recording a payment on a prepaid balance is an owner's. A member gets answers, not cards.

Your organization can switch it off without switching off the answers: Settings → Developers → Assistant access has one switch per kind of change, and they cover connected assistants and this panel at once. With one off, the assistant says so rather than pretending the product cannot do it.

It still never charges a card, never refunds money, never deletes anything, and never touches credentials — whatever it is asked. See what a connected assistant can and cannot do for the full list.

Pointing at something specific

Type @ followed by at least two letters to point at a pipeline, buyer or lead by name — @Auto Insurance, @SunWest. Picking one attaches it to your question so you don't have to describe it.

On a lead, buyer or pipeline page, the assistant already knows which one you're looking at. "Explain this lead" is enough.

What it won't do

• Change something without showing you first. Every change is a card you apply.
• Tell you a change is done. It is not told whether you applied one, so it never claims you did.
• Discuss billing, invoices or your plan. Use Settings → Billing, or plans and quotas.
• Answer for another organization. It only ever sees your own.
• Invent a number. If the data isn't there it says so rather than filling the gap — an empty window is reported as empty.

Conversations and history

Each thread is kept. The clock icon opens your recent conversations, and reopening the panel within 24 hours offers Continue where you left off. Cards come back with what really became of them: Applied, Reverted, discarded or expired, even when you acted on them from another tab or from the activity log, and the assistant knows it too. The pencil icon starts a fresh thread — worth doing when you change subject, since it carries the earlier turns as context.

Thumbs up and down sit under every answer, and Report incorrect lets you say what was wrong. Both are read.

Using Claude or ChatGPT instead of the in-app assistant

You can connect Claude, ChatGPT, Claude Code, Cursor or a script to your workspace and ask the same questions from there, and propose the same changes. It reads the same data through the same rules, and it is included on every plan.

The same card, in the conversation. In Claude and in ChatGPT a proposed change arrives as the same LeadMove card you see here: what would change, who it affects, whether it can be undone, with Apply, Discard and, once applied, Undo for the same seven days. Saying yes in the chat works just as well. Claude Code and other clients show the same preview as text, and you approve it in the conversation.

Where to set it up. Settings → Developers → Connect an assistant, then pick where you use it. The dialog gives you the link for Claude's connectors, or one command for Claude Code, and waits until the connection says Connected. ChatGPT works through a custom connector: until LeadMove is listed in the ChatGPT app directory, you allow custom connectors in your ChatGPT settings first, and the LeadMove screen shows the steps. A connection has the permissions of your own role, chosen on a LeadMove screen, and can be revoked in one click afterwards. See connect Claude to your workspace, other MCP clients for ChatGPT and Cursor, and what a connected assistant can and cannot do for the permissions and limits.

Owners and admins also see Also available in Claude and ChatGPT under the message box here, which opens the same Settings page, and the Finish setting up list after onboarding has the same entry.

The Let assistants read this workspace switch governs both surfaces at once: turning it off hides the in-app assistant and stops every connection.

Fair use

500 questions per month per organization, on every plan. That is far above normal use — it exists to stop a runaway loop, not to ration you. If you reach it, the assistant says so and comes back at the start of the next month; message support if you need it lifted.

Common questions

Can it set up a routing rule for me? Yes — ask, and it proposes the rule as a card you read and apply. A rule created that way arrives switched off and flagged for review, so nothing is routed by it until you turn it on.

Where is it on mobile? The sparkle icon in the header, and an Assistant entry in the menu. It opens full screen.

It gave me a wrong answer. What now? Use Report incorrect under the answer and say what was wrong — those reports are what new test cases get written from. For anything urgent, Ask support at the bottom of the panel puts you through to a human.

Does it see my leads' personal data? It reads the fields it needs to answer, from your organization only. Contact details are included when the question is about a specific lead, since that is the answer.

Does using it from Claude or ChatGPT count against the 500 questions? No. The allowance covers this panel. A connection has its own ceilings, there to stop a runaway loop rather than to ration you, listed in what a connected assistant can and cannot do. Reading is never affected by them.

Can I turn it off for my team? Yes, for the whole organization: owners and admins can switch it off in Settings › General › AI features, or from Assistant access in Settings › Developers — it is the same switch — and it disappears for everyone. Not per user today. The same screen has a switch per kind of change, if you would rather keep the answers and stop the cards. Nobody outside your organization can see your data through it.