It proposes, you confirm.
Describe the change instead of clicking through to make it.
An AI assistant that can act on a system is only safe when the boundary between reading and writing is explicit. Here, the assistant reads freely across the workspace and answers questions, but every change it wants to make is presented as a card you confirm. Nothing is executed silently, and confirmed work runs through the same guarded paths a person uses.
- Reads contacts, conversations, campaigns, segments and reports
- Every change proposed as a card you approve or discard
- Confirmed actions run through the same checks as manual ones
- Answers grounded in your knowledge base with citations
How it works.
- 01
Ask in plain language
Ask how many contacts reached a stage, which campaign is ready to send, or what the channel health looks like. The assistant queries your workspace to answer.
- 02
Watch the work
A timeline beside the conversation shows what it is reading as it reads it, so an answer can be checked.
- 03
Confirm a change
When a change is needed, it appears as a proposal: add a note, assign a conversation, create a segment, schedule a campaign, suppress an address. You confirm or discard.
- 04
It runs the normal way
A confirmed action runs through the same guarded route a person would use, with the same permission checks, consent rules and audit record.
Reading is the easy half
The assistant can look across the whole workspace, so an answer is about your business.
- Contacts, stages and profiles
- Conversations and their state
- Campaigns, readiness, forecasts and recipient results
- Segments, sequences, channel health and the knowledge base
Writing always asks
Every change is a proposal with the details visible before you agree. This is the boundary that makes an acting assistant safe to use on real customer data.
- Contact edits, merges, notes and tags
- Conversation assignment, replies and closing
- Campaign creation, scheduling, pausing and test sends
- Suppression, segment creation and sequence enrolment
Grounded, with its sources
Answers about your business are drawn from the knowledge base and your own records, and the parts used are cited so a claim can be checked.
- Citations back to the source document
- Workspace data read live at the moment you ask
- A cost ceiling per run so a runaway task stops
Rehearse before customers are involved
Every part of the product has a way to try it first: a practice mode for the reply agent, a dry run for automations, and a test send for campaigns.
- Ask it a question and see exactly which of your answers it used
- Simulate an automation flow without sending
- Test send a campaign to yourself before it goes wide
Tools that reach outside
Connections to external systems, so an action can update an order or a record in a tool you already run, under the same confirmation boundary.
- Read-only external tools first
- Writes proposed and confirmed like every other action
- Nothing is enabled by default
What it does not do.
It cannot change anything without your confirmation.
Anything it does in another system runs through a connected integration, under the same confirmation.
Your team uses it. Customers never see it.
Common questions
Can the assistant change things on its own?
- No. Reads happen freely, writes are proposed as cards and nothing executes until you confirm.
What can it actually do once confirmed?
- Contact edits and merges, notes, conversation assignment and closing, replies, campaign creation and scheduling, suppression, segments and sequence enrolment.
Does confirming bypass permissions?
- No. A confirmed action runs through the same route a person uses, with the same role checks, consent rules and audit record.
Can it answer questions about my data?
- Yes. It queries your workspace live, and grounds business answers in the knowledge base with citations back to the source.
Does the session survive a refresh?
- Yes. The conversation is held on the server, so reloading the page or continuing on another device keeps the thread.
Who can use it?
- Anyone who can operate the workspace. What a confirmed action is allowed to do is still bound by that person's role.
Can it connect to my other systems?
- Anything reaching outside your workspace runs through a connected integration you enable, and it follows the same confirmation boundary.
How is a runaway task prevented?
- Each run has a cost ceiling, and every write needs a human confirmation, so an unattended loop cannot spend or send.
The reason a customer comes back a second time.