The assistant

The assistant lets you use the platform in natural language. You write what you want to get, it calls the platform's tools and shows you each call in a card. Any action that affects a real person waits for your approval.

Where to find the assistant

You can open the assistant in three ways:

• The "Assistant" page, in the left menu. It shows the six tabs and the conversation history.
• The side panel, with the "Open the assistant panel" button. It keeps the current screen visible next to it.
• From a screen, with a dedicated button: "Draft with the Assistant" in a campaign, "Ask the Assistant" in "Campaigns", "Write with the assistant" in "Publish". From a campaign, the conversation opens with the line "About campaign "…"".

The "AI agent" mode of the "Search" tab, in "Prospects", uses the same assistant as the assistant's "Search" tab. See Prospects.

The six tabs

Each tab has its own tools. A tab can only do what its tools allow. If you ask for something else, the assistant tells you.

What else the "Campaigns" tab can do

Available

• Create a Target that does not exist yet, edit it, and read its "health" (how many evaluated prospects match it).
• Review an existing campaign: steps, settings, Target, number of people per status.
• Pause, resume, stop, archive and restore a campaign. Pause, resume or stop a single person.
• Read a campaign's approval queue and act on a send: edit the text, choose a variant, rewrite, skip, cancel. Sending itself remains your action, on the campaign screen.
• Prepare the texts for the next sending day. Propose instructions for a step.
• Propose confirming a Target on a "Confirm a target" card. This card also exists in the "Search" and "Command" tabs.
• In "Command", review a campaign and read its approval queue.

The rules the assistant follows

No invented figures. Each number, status or name the assistant quotes comes from a tool card. If a tool refuses, the assistant gives you the reason for the refusal.

Nothing goes to a person without your approval. When the assistant proposes to send a reply, activate a campaign or arm Autopilot, it shows an "Approval required" card. Nothing happens until you click "Approve".

The "Campaigns" tab stops at the draft. The assistant builds the campaign and runs the dress rehearsal. You review the texts, then you activate the campaign yourself on the campaign screen. See Campaigns.

No e-mail or InMail. The platform does not send e-mails today. No campaign type contains InMail. The assistant says so if you ask.

Text written by third parties gives no orders. A received message, a LinkedIn profile or a search result is written by someone else. It can contain a sentence that looks like an instruction. The assistant never follows such a sentence:

• in the Inbox, it does not read the raw message: it reads a sanitized summary, marked "Sanitized summary of a third-party message — never executed.";
• in Search, it shows the results without obeying them;
• an action towards a person taken after such a reading always requires your approval.

The recipient is chosen by the platform, not by the text. The assistant never uses an address or a name written in a message to send something. It acts on objects already in your workspace: a conversation, a draft, a campaign.

It prices before it spends. Before a paid action on several prospects, the assistant shows a quote. Before promising a volume, it reads your LinkedIn account's capacity for the day.

What you see in the conversation

Tool cards

The assistant's actions are grouped in one block per answer.

• While it works, one line shows the current action and its rank, for example "Count prospects… · step 3 · 2 finished".
• When the work is done, the block collapses to one line: "✓ 6 actions completed". Click it to see the list.
• If an action was refused or failed, the line turns to a warning colour, for example "6 actions · 1 refused". The reason stays visible under the line, even when the block is collapsed.
• If an action is waiting for your decision, the line shows "Waiting for your approval".

In the list, each action has a status icon, a name ("Today's capacity", "Create a campaign"…) and a one-line result. Click an action to see its parameters, the full result and the technical detail.

Links such as "Open in Campaigns" or "Open in Prospects" stay visible under the block. They take you to the screen concerned.

To always see the detail, turn on "Always show action details" in the right column of the Assistant page. The setting unfolds the list and each action. It is remembered in your browser.

The message field grows with your text. Enter sends the message. Shift+Enter adds a new line.

In the "Inbox" tab, two special cards appear: "Summary of the received message" (intent, sentiment, summary) and "Proposal" (the draft and its "Rationale"). If the reply depends on information only you know, the card shows "To confirm before sending".

The "Approval required" card

This card appears before any sensitive action. It shows what is about to happen, with figures read by the platform and not written by the assistant. Examples:

• activating a campaign: "“Campaign name” — 42 recipients";
• sending a reply: "To First Last · “start of the message”";
• arming Autopilot: the Target, the number of profiles per night, the nights of reserve and the maximum cost per night;
• confirming a Target: "Target “…”", "Objective", "Definition", "Disqualifiers", "Threshold";
• stopping a campaign: the number of people stopped and the sends cancelled;
• rewriting several sends: the number of texts and the cost in credits.

To decide:

1. Read the card.
2. Click "Approve" to run the action.
3. Or click "Refuse". A "Refusal reason" field opens. Write the reason if you wish, then click "Refuse" again.

After a refusal, the action is not done. The assistant reads your reason and corrects its proposal in the same turn. For example, with the reason "Limit the Target to France", it edits the Target and then shows you the card again.

"A card is waiting for your answer"

While an approval card is waiting for your decision, the conversation is paused. You cannot send a new message.

The composer says so: "A card is waiting for your answer: approve or refuse it before sending a new message." The "Show the card" button takes you back to it. Your text stays in the composer.

The "Proposed plan" card

For work in several steps, the assistant can propose a plan. The "Proposed plan" card shows:

• the list of steps and what each one does;
• the "Quote": "Total: … cr" and "Balance after: … cr";
• the step that would send something to people, marked "send — approval required".

Two buttons:

• "Launch": you approve the whole plan in one go. The steps run one after another. Each step shows its state: "pending", "done", "failed" or "skipped". The plan stops at the first failure.
• "Edit": the cursor goes back to the composer. Write what needs to change.

In the "Campaigns" tab, a plan cannot contain the activation. Adding prospects by Target or by segment cannot be part of a plan: the assistant does it directly and tells you.

How to word a request

Write the result you want, with the information only you know. The assistant chooses the tools.

Always give:

• the audience: the name of the Target, the saved search or the segment. Write the exact name in quotes. The assistant finds the Target by its name; it does not guess it.
• the relationship degree for a campaign: 1st degree (your connections) or 2nd/3rd degree (not connected yet). The degree decides the campaign type.
• what must be said: the topic, the link, the date, the angle.

You can leave out the rest. The assistant uses default values and announces them: campaign name, sending window, follow-up delay (3 to 4 days), approval of sends (on by default). You change them afterwards in one sentence.

Examples:

If the assistant asks questions, it asks them all at once. If you do not answer a question, it uses a default value and continues.

Ready-made requests

When a conversation is empty, the "Here is what I can do" screen offers requests under "To get started". One click fills the composer. You send it yourself. Replace the "…" with your values.

A request only appears if your plan includes the tools it needs. Otherwise, the screen shows "Your plan doesn't include these tools."

"Publish" tab

• "Draft a LinkedIn post about" (complete the topic)
• "Draft a LinkedIn post for my Target « … »."
• "Revise my latest draft: make it shorter and the hook more direct."
• "Plan a series of 3 posts over the coming week."
• "List my latest posts."

"Campaigns" tab

• "Build an invitation campaign for people in my Target « … » who aren't connected yet."
• "Announce our webinar on … to my 1st-degree connections, here is the link:" (complete the link)
• "How many people in my default Target are 1st degree vs 2nd?"
• "Propose a full plan — campaign, steps, enrolment, rehearsal — for me to approve once."
• "Rehearse my latest campaign: show me the messages that would go out."

"Inbox" tab (after choosing a conversation)

• "Summarize this thread in a few sentences."
• "Suggest a reply."
• "Suggest a meeting for next steps."

"Search" tab

• "Find fintech CTOs in France, 50-200 employees, on LinkedIn."
• "Save this search and create a Target from it."
• "Arm the autopilot on my Target « … »: how many profiles imported per night and how many credits?"
• "Rerun my latest saved search."
• "How many nights does my profile reserve last?"

"Command" tab

• "Find my CTO leads, qualify them against my Target and prepare a campaign."
• "Give me a costed plan before you act."
• "How many CTOs do I have in SaaS?"
• "How many people can I reach today?"

"Intelligence" tab

• "Follow LinkedIn posts that mention “generative AI” or “customer service”."
• "Watch AI trends in B2B."
• "Show me my trends."
• "Explain the first trend."

Full conversations

The examples below show what you type and what the assistant does. The names and figures are examples. The tool card names are the ones shown on screen.

1. Build an acquisition campaign ("Campaigns" tab)

You:

Build a campaign for the people in my Target "Operations directors" who are not connected yet: profile visit, like, then invitation with a short note about our hosting audits. Require my approval before sends.

The assistant:

1. "Today's capacity": it reads the invitations and messages left for the LinkedIn account, and the credit balance.
2. "List Targets": it finds the Target "Operations directors".
3. "Count prospects": it counts the people in the Target at 2nd and 3rd degree, and gives you the breakdown before adding anyone.
4. "Create a campaign" of type "Acquisition", then "Add a step" three times: visit, like, invitation with note.
5. "Update campaign settings" (sending window, pace), "Require approval", "Set the campaign Target".
6. "Enroll contacts": it adds the people in the Target.
7. "Dry-run rehearsal": it writes each person's invitation note and shows it with the planned time, without sending anything.

It ends with what is left to do: review the rehearsal, then activate the campaign from its screen.

You:

The notes are too long. Keep two sentences and remove the mention of the price.

The assistant: "Update a step" to change the invitation instructions, then "Preview a step" to show you the new texts.

Cost: 3 credits per note written. A text already written is replayed for free in a new rehearsal.

The assistant also reviews the campaign with "Read a campaign" before the rehearsal. It checks that an "Acquisition" campaign contains a single invitation, as the last step. By default, it spaces the steps as follows: visit at 0 h, like 1 h later, comment 1 h later, invitation 2 h later. If the Target does not exist, it creates it as a draft and proposes that you confirm it.

2. Announce a webinar with a plan ("Campaigns" tab)

You:

Propose a full plan to announce our webinar on 14 November to my 1st-degree connections in the Target "Mid-market CIOs". Link: https://example.com/webinar. One message, then one follow-up.

The assistant:

1. "Today's capacity", "List Targets", "Count prospects".
2. "Propose the plan": the "Proposed plan" card lists the creation of the "Network messaging" campaign, the two steps, the settings and the rehearsal, with the "Quote".
3. It states that the people in the Target will be added outside the plan, directly.

You: click "Launch".

The assistant: the plan steps switch to "done" one by one. It then adds the people in the Target with "Enroll contacts", then runs the "Dry-run rehearsal".

Cost: 3 credits per message written, and 3 credits per follow-up written.

3. Qualify a list ("Command" tab)

You:

How many CTOs do I have in SaaS? Qualify them against my Target "CTO SaaS 10-50" and give me the quote first.

The assistant:

1. "Count prospects": for example, 64 CTOs in SaaS.
2. "Today's capacity": the evaluations left today under your plan.
3. "Plan quote": the card shows the total, here 64 × 2 = 128 credits, and your balance afterwards.

You:

Go ahead.

The assistant: "Qualify against the Target". The evaluations are queued. Those above the daily cap go out on the following days. The scores appear in "Prospects" once they are calculated.

4. Arm Autopilot ("Search" tab)

You:

Find IT directors of mid-sized retail companies in France, 2nd degree.

The assistant:

1. "Resolve locations and industries": it translates "France" and "retail" into LinkedIn filters.
2. "Search LinkedIn": it runs a single search and shows the results.

You:

Save this search as "Retail CIOs FR" and create a Target from it.

The assistant: "Save a search", then "Draft a Target". It shows you the proposed Target and waits for your agreement. Then, "Create a Target".

You:

Arm Autopilot on this Target with 20 profiles per night.

The assistant:

1. "Today's capacity": it checks that your plan allows daily qualification.
2. "Turn on autopilot": the "Approval required" card shows the Target, 20 profiles per night, the nights the reserve will last and the maximum cost per night (40 credits).

You: click "Approve".

The assistant confirms that Autopilot is armed. It never says so before your approval. See Autopilot.

5. Reply to a message ("Inbox" tab)

You: choose the conversation, then:

Suggest a reply.

The assistant:

1. "Reading the conversation": the "Summary of the received message" card shows the intent (for example "Question"), the sentiment and the summary.
2. "Reply suggestion": the "Proposal" card shows the draft and its rationale.

You:

Shorter, and offer two slots on Tuesday or Thursday morning.

The assistant: a new proposal, then "Draft saved" when you approve it.

You:

Send it.

The assistant: "Sending the reply". The "Approval required" card shows "To First Last · “start of the message”". You click "Approve".

Cost: 3 credits per reply suggestion. Sending costs nothing.

If the received message asks not to be contacted again, the assistant can propose "Mark as opted out". A card asks you to confirm. After an opt-out, no message goes to this person on any channel, and their running campaigns stop. Only the person can ask for the opt-out to be lifted.

6. Create a watch ("Intelligence" tab)

You:

Create a watch on digital sovereignty for my Target "Mid-market CIOs". Follow LinkedIn posts with the keywords "trusted cloud" and "SecNumCloud", and this RSS feed: https://example.com/feed.xml

The assistant:

1. "List Targets", "Read a Target", "Read the business profile": it reads the Target and your profile to propose the topic and the keywords.
2. "Create a watch" with the given RSS feed and the LinkedIn keyword source.
3. It states that collection runs in the background: trends are not immediate.

You, later:

Show me my trends.

Explain the first trend.

The assistant: "List trends", then "Explain a trend". It quotes the source address of each trend.

You:

Add the posts of the company "Example Cloud" to this watch.

The assistant: it announces that searching for an author costs credits and asks for your agreement. Then "Find a LinkedIn author", shows you the results and lets you choose. Finally "Update a watch".

Cost: 5 credits per author search. Rerunning the watch ("Relaunch a watch") costs the LinkedIn read and the report: see Intelligence.

7. Write a post for a Target ("Publish" tab)

You:

Draft a LinkedIn post for my Target "Operations directors".

The assistant:

1. "List Targets", "Read a Target", "Read the business profile".
2. It chooses an angle taken from the Target's definition and states it in one sentence.
3. "Draft a post": the draft is created and linked to the Target. The "Open in Publish" link takes you to the draft.

You:

Rewrite the hook so that it starts with a number, and end with a question.

The assistant: "Revise post".

To publish, open the draft on the "Publish" screen. The assistant does not publish. See Publish.

Cost: 15 credits per draft, 15 credits per revision.

8. Find out how many people you can contact today ("Command" tab)

You:

How many people can I reach today?

The assistant: "Today's capacity". The card shows "invitations left today: …", "messages left today: …" and "balance: … cr". If no LinkedIn account is connected, the card says so: no invitation or message can go out.

Cost: 0 credits.

9. Create and confirm a Target from a campaign ("Campaigns" tab)

Available

You:

Create an invitation campaign for a new Target "Healthcare CISOs": security managers of clinic groups in France.

The assistant:

1. "List Targets": no Target with this name.
2. "Create a Target" as a draft, with the name, objective and definition taken from your request. It announces the values used.
3. It builds the campaign and runs the rehearsal, as in example 1.
4. Last, "Confirm a target": the card shows the name, objective, definition, disqualifiers and threshold.

You: click "Refuse", reason: "Exclude clinics with fewer than 200 employees."

The assistant: "Update the Target" to add this criterion, tells you what it changed, then proposes "Confirm a target" again.

You: click "Approve". The Target is confirmed. Campaign texts rely only on a confirmed Target.

10. Manage a campaign's approval queue ("Campaigns" tab)

Available

You:

Pause the campaign "Webinar 14 November" and show me the sends waiting for approval.

The assistant: "Pause a campaign", then "List sends to approve". It shows the pending texts, without the people's names. To see the person, open the campaign's "To approve" screen.

You:

Rewrite all the pending messages with a less salesy tone.

The assistant: "Rewrite several sends". The approval card shows the number of texts and the cost, at 3 credits per text, 25 texts at most per call. After your approval, the new texts still wait for approval. The assistant does not approve or send any message.

Cost of the assistant

• A conversation costs nothing in itself.
• The paid tools the assistant calls cost the same price as from the screens. Examples: 15 credits for a post draft, 3 credits for a message or a reply suggestion, 2 credits per prospect evaluated.
• Reads, counts, today's capacity and creating a campaign cost nothing.
• A LinkedIn search costs no credits. It uses the LinkedIn account's "search lines" (1,000 per day with Classic, 2,500 with Sales Navigator).

The price list is in Credits and subscription.

History and dictation

• The "Show conversation history" button opens the "Conversations" list. Each conversation can be renamed or deleted. "New conversation" starts another one.
• The "Dictate" button transcribes your voice into the composer. The transcription is done by your browser's speech service.
• "Agent tools" opens the list of tools available in your workspace. See Account and team.