Session replay, for agents.
Every conversation an AI agent had with your site: what it asked, exactly what it was told, and how far it got. Your team acts on it faster, and you see which research didn’t convert so you can win the next one.
A visit, a form, or their agent’s conversation.
A visit shows that someone came, and a form shows how it ended. Their agent’s conversation shows what they needed, and what you said back.
A page view and a referrer.
Someone came, and from where. When their agent acts from its own servers, there is no page view for your analytics to count.
What they typed at the end.
The last step, in the box you gave them. Nothing of the questions they asked on the way, or of what your site told them.
Their questions, and your answers.
Each question in the agent’s words, your site’s exact answer, whether it was a real answer, and how far it got. The conversation in your workspace is the record.
Their agent does the research. You win the customer.
Each question, and exactly what it was told.
When an AI agent uses your site’s agent tools, each visit is recorded as one conversation. Open any one and replay it turn by turn.
- Your MCP endpoint, which an agent can call from its own servers
- Your /agent/* answers
- The Agent Actions your pages register in the browser
- The agent’s question with contact details masked, your exact answer, and whether it was a real answer
- Which agent it was, as it named itself, and which of your tools it used
- Found
- Understood
- Chosen
- Acted
- Delivered
- 01 get_briefing Answered
Residential roofing across metro Atlanta: the licence, the service area and the projects it takes on, each linked to its source.
- 02 ask_site Answered
“Do you handle hail damage on an older shingle roof, and do you cover 30307?”
Yes. Storm damage is one of the projects this contractor takes on, and 30307 is inside its service area.
- 03 ask_site Partly answered
“Will you work directly with my insurance adjuster?”
Storm-damage estimates include photos you can share with your insurer.
- 04 request_estimate
project_type: storm_damage · service_zip: 30307 · timing: emergency
Reference IA-7Q2KD4, waiting for the person's go-ahead.
- 05 confirm_request
name: [masked] · phone: [masked] · consent_contact: true
Confirmed. Reference IA-7Q2KD4. The business has your request and will respond within one business day.
You’ll never hear what a buyer tells their agent. You can read what their agent asks you.
See every question agents ask your site, how often, and whether your site had a real answer, including from buyers who never became leads.
- Answered: your site had a real answer for this.
- Partly answered: a general answer went out for a specific question.
- No answer: your site had nothing to say.
- One filter finds every conversation that hit a question you could not answer.
- Publish the answer: draft an Inception Point from the question, approve it, and agents can use it.
See where the research stopped.
Each conversation shows how far the agent got: found, understood, chosen, acted, delivered. Each one also ends with what came of it.
| How it ended | What it means |
|---|---|
| Converted | It ended in a lead. |
| Stopped at confirm | The agent reached the confirmation step and the person did not say yes. |
| Informed | The agent got real answers and left without acting. |
| Unserved | Nothing it asked got a real answer. |
| Still open | Recent enough that it may still continue. |
Paths shows the routes agent conversations take through your answers, what comes of each, and how each agent researches your site.
Sales starts where their research ended.
Each lead arrives with their words, the facts, how they said yes, and what their agent asked first. One link replays the conversation, so the first call picks up where the research left off.
- One lead, in your Leads inbox and by email, with the agent and the surface it came from
- What their agent asked first: its questions before the request, each marked answered, partly answered or no answer
- Replay the conversation: one link to the turn-by-turn replay in your dashboard
- Record what came of it, and what it was worth, from your inbox or by asking your own Claude
New lead IA-7Q2KD4
request estimate via mcp_hosted for summitroofing.example
1. Stated facts
- project type
- storm_damage
- service zip
- 30307
- timing
- emergency
2. Verbatim ask
- ask
- Hail came through last night and there’s a drip in the upstairs bedroom. The roof is about 20 years old. Can someone look at it this week?
7. Provenance
- surface
- mcp_hosted
- approval method
- agent_authorized
10. Contact
- name
- Dana Whitfield
- phone
- (404) 555-0142
What their agent asked first
- answered
- “Do you handle hail damage on an older shingle roof, and do you cover 30307?”
- partly answered
- “Will you work directly with my insurance adjuster?”
The same record, in both lanes.
The request, with what happened in their words.
The consult, estimate or tour request arrives with the person’s own words, the facts your intake asks for, and what their agent asked first.
Demo requests that arrive briefed.
When a buyer’s agent asks your site about pricing and plans, each question lands in your workspace with the answer it got. A demo request arrives with what their agent asked first.
How they said yes, on every lead.
Nothing reaches you without their go-ahead, and every lead says how it was given: on a card in your brand, or through their own agent.
| On your page | Through their own agent | |
|---|---|---|
| Who gives the go-ahead | The person, on a card in your brand | The person, to their own agent, which completes the request for them |
| What comes back | A reference, shown on the card | A reference their agent can pass on to them |
| Recorded on the lead as | card_tap | agent_authorized |
| Opts them into your marketing | Only if they tick it themselves | Never |
The contact form is disappearing into the tool call. The card is for the person. The tool is for their agent. Either one gets you the yes, and the lead records which.
Their agent can ask you to call them back. It can’t sign them up for your marketing. Only the person’s own tick does that.
One contact can complete five requests a day with you, and agents that call too fast are told to wait.
Ask your own Claude what came in today.
Through the Inception Agents connector, ask Claude about your agent conversations, your leads and your pipeline, and act on them.
- “What came in today?”
- “What did agents ask us this week that our site could not answer?”
- “For each new lead, what did its agent ask first?”
- “How far are agent conversations getting, from found to delivered?”
- Or run /conversation-insights for one brief, with a lens each for sales, marketing, product and leadership
The record is yours.
Agent conversation recording is on from day one, and the record sits in your workspace.
Masked when it is recorded
Contact details in an agent’s question are masked when it is recorded. Their details reach your team on the lead they approved.
Kept a year, erased on request
Conversations are kept a year by default, and an erasure request removes them.
Never shown to visiting agents
Visiting agents see your public tools only. Your conversations and your analytics answer only to you.
See your first agent lead before a buyer sends one.
Before you go live, we send a test lead through your real site the way a buyer’s agent would. Your team gets it marked TEST, with what the agent asked first and a link to replay the conversation, and it is never billed.
- It asks a few questions, then requests a callback
- One of your people confirms it arrived before your site is marked live
- We took our own website live on the platform on September 25, 2026 by asking Claude, and it has passed every hourly check since
TEST LEAD. This is a test lead, sent while your Inception Agents setup is checked. It is not a real enquiry: nobody is waiting for a reply.
I received this test lead[TEST] New lead IA-T5K3P8
request callback via mcp_hosted for summitroofing.example
What their agent asked first
- answered
- “What do you help with, and how quickly does someone usually get back to a new enquiry?”
- answered
- “I’m looking for help with something new. Is it the kind of work you take on, and do you cover my area?”
What the record covers.
Three things to know about what each conversation holds.
-
Your site’s tools, not every page read
It records conversations with your site’s agent tools. An agent that only reads your pages leaves no questions to read.
-
One sitting, one conversation
Each sitting is its own conversation. If their agent comes back another day, that is a new one.
-
Your inbox, your dashboard and your own Claude
What their agent asked shows in your lead email, your Leads inbox, your dashboard and your own Claude.
This is where you grow the channel.
Grow agents into your best channel.
Run the free audit to see what an agent can do on your site today, or book a demo to see what arrives with every request.