Until now, a candidate reply meant the automation stopped and a human took over. The Conversational Agent changes that: when a candidate replies, the agent continues the conversation for you — it answers questions, works toward a goal you define, and can book a meeting straight into your calendar. You stay in control the whole time from the Inbox.
Where to find it
Open your campaign → Campaign Builder tab. The Conversational Agent block sits in the Actions palette on the left.
If you don't see the block, the Conversational Agent isn't enabled for your workspace yet — contact us via Chat with us.
Adding it to your flow
The agent starts working the moment a candidate replies, so it must be connected to a green response path:
Drag the Conversational Agent block onto the canvas.
Connect a green handle into it:
🟢 Accepted & Replied on a Connection Request, or
🟢 Replied on a Send Message, Send InMail, Send E-Mail, or Send WhatsApp block.
Configure the agent in its settings panel (see below).
The publish check enforces this: "The Conversational Agent must be connected to a response edge (accepted or responded)."
The agent is the end of the automated sequence — from there the conversation runs turn by turn until the goal is reached, the candidate asks to stop, or the agent hands the conversation to you.
Choosing the agent's goal
Agent goal type offers two modes:
Schedule a meeting (Cal.com required) — the default. The agent works toward booking a meeting with the candidate. You can optionally add extra goal instructions as context.
Custom goal — describe your own objective (for example "collect the candidate's salary expectations and availability"). A custom goal is required text — the publish check blocks an empty one.
How the agent replies
Same channel, no setup — the agent replies on the channel the candidate used (LinkedIn message, InMail, email, or WhatsApp). There's nothing to select.
Candidate's language — replies follow the language of the conversation; you can force a specific language in the block settings if needed.
Not instant — each reply is scheduled with a short, slightly randomized delay so the conversation feels natural.
Visible before it's sent — every agent reply appears in the conversation's composer as a scheduled message, where you can edit, approve, send now, or cancel it. See "AI-assisted replies" in the Inbox collection.
Scheduling meetings with Cal.com
With the Schedule a meeting goal, the agent uses the Cal.com calendar of the user whose connection sends the messages:
It proposes a couple of concrete available times spread over different days, and includes your booking link as a self-serve alternative if you configured one.
When the candidate picks a time, the agent confirms the details — and asks for an email address if it doesn't have one yet.
It books the meeting directly in Cal.com and confirms it to the candidate. Goal reached.
Connect Cal.com first under Personal Settings → Calendar integration — see the Calendar integration (Cal.com) article in Settings & Team Management. The publish check reminds you: "The Conversational Agent requires the sending user to have Cal.com connected for scheduling."
Staying in control
Require approval before sending — off by default (replies are scheduled and sent automatically). Turn it on and each agent reply is drafted for your review first; approve it from the Inbox composer before it goes out.
Maximum conversation turns — the agent escalates to a human after this many turns (default 6) without reaching the goal.
Escalation instructions — describe when and how the agent should hand the conversation to a human recruiter (per language).
Take over any time — in the Inbox, an active conversation shows Agent active — take over; click it or simply send a manual reply and the agent steps back.
Opt-out safe — if the candidate indicates they're not interested or asks to stop, the agent stops.
Starting the agent from the Inbox
Campaigns aren't the only way in: the Hand to agent button in any Inbox conversation starts an agent session on the spot, with its own goal — see "AI-assisted replies" in the Inbox collection.
FAQ
Which channels does the agent support? All four: LinkedIn messages, InMail, email, and WhatsApp — always the channel the candidate replied on.
What happens when the maximum turns are reached? The agent stops and escalates: the conversation is yours again in the Inbox, following your escalation instructions.
Can I read what the agent has said? Yes — every agent message is a normal message in the Inbox thread, and upcoming replies are visible (and editable) before they're sent.
Does the agent need approval for every message? Only if you switch on Require approval before sending. By default it sends automatically after the short scheduling delay.