Triple Lab
Checked, grounded, then sent.
Every reply your Agent sends runs the same governed steps, inside the rules you set.
How does TripleTick's AI Agent work?
TripleTick's AI Agent handles each message in eight steps: it reads a burst as one turn, checks the messaging window, opt-out and your limits, grounds the reply in the contact's memory and your knowledge, writes it with frontier models from Anthropic, acts through live tools, filters it, stops when a teammate takes over, and updates the vitals.
Follow one message.
Emma, a lead on Instagram, confirms a showroom visit, and each step shows what the Agent does with her message.
Step 1 of 8
Arrive
A burst of messages becomes one turn, and voice notes and images become text first.
From Emma, on Instagram
11 is perfect. My partner has to approve it before we buy though.
Voice note
Image
One turn
What each step guarantees.
01
Arrive
- A burst of messages is answered once.
- Voice notes are transcribed before it reads them.
- Images are described before it reads them.
02
Check
- The messaging window and opt-out come first.
- Your spending limits can stop a turn.
- A classifier reads intent and sentiment.
03
Ground
- Recalls facts learned about the contact.
- Answers from a search of your knowledge.
- Never reads a failed search as a no.
04
Reply
- Written by frontier models from Anthropic.
- Shaped by your Playbook's instructions.
- English, Arabic or Turkish, never posing as a person.
05
Act
- Moves deals through your Pipeline.
- Sets fields on the contact.
- Sends templates and locations.
06
Filter
- Runs on every outbound message.
- No prompt can switch it off.
- Banned phrases are rewritten out.
07
Hand off
- A teammate can take over any conversation.
- After that, the Agent stays silent there.
- A customer who asks for a person gets one.
08
Update
- Intent, sentiment and blocker follow each message.
- Momentum and qualification follow the conversation.
- All six dials sit beside the thread.
Built on the official rails.
- Official Meta Tech Partner
TripleTick runs on the official WhatsApp Business Platform.
- Meta's Embedded Signup
You connect your number through Meta's own official signup flow.
- A public changelog
Every change to the Agent is written down.
Read the changelog- Security and data
How your conversations are stored and protected.
See security and data- Every AI vendor named
The companies that process your data, listed.
See sub-processors
Questions about the engine.
- Does TripleTick train its own AI model?
No, replies are written by frontier models from Anthropic.
Your Playbook, your knowledge and what the Agent remembers about each contact shape every reply.
- What happens when the Agent cannot find an answer?
It searches your knowledge first.
If that search fails, it tells the customer it could not check, instead of claiming you do not offer something.
- Can a prompt get past the safety filter?
No, the filter runs on the finished message after the reply is written, so no instruction or customer message can switch it off.
Your own brand voice settings decide which phrases it removes.
- What stops the Agent once my team takes over?
Taking over a conversation stops the Agent there.
Whatever the customer sends next, the Agent does not reply in that conversation.
- Which languages can the Agent reply in?
English, Arabic and Turkish.
Each language has its own honesty check, so the Agent never claims to be a person.
- Will the Agent message someone who opted out?
No, the opt-out check runs before any AI does, and an opted-out contact gets no AI reply.
Let the Agent take the first reply.
Set the rules in a Playbook, and watch every step in your Inbox.
Free for 7 days