Trust and control
Control is a feature, not a promise.
Agents act only through the tools, channels and rules you give them. Every send asks first, every decision is on the record, and a person can step in at any moment.
Lucía Morales
Handled by Leo · Sales
Every send asks first
Consent, opt-outs and quiet hours, checked before anything goes out.
Each text and call is checked against the consent your records hold for its purpose, the customer’s opt-outs on that channel, the store’s quiet hours and the sender’s registration. A send that can’t go says why.
- Consent by purposeMarketing needs marketing consent; service messages need at least service consent.
- Opt-outs stop sendsSTOP closes that channel, and the reason is on the customer’s record.
- Quiet hoursA send outside the store’s hours waits, with the time it resumes.
- Registered sendersTexts go from 10DLC-registered numbers in the store’s name.
Only what you allow
An agent can’t do what it hasn’t been given.
Each agent works at one store, through the tools attached to it. Reading openings, booking, texting a caller, transferring, writing a note: each is a tool you can see and switch.
- Reads and writes, labelledYou see which tools read your systems and which write to them.
- Connections confirmed per storeWhat each system may read or write is agreed with you during setup.
Service Agent Active
as Mandy · Service · Your store
Attached tools
An agent can only act through the tools attached to it, at the store it belongs to.
Guardrails
Every agent passes the same tests.
Before an agent talks to customers it runs the scenario benchmark: role-played conversations on every channel, graded against written criteria, three times each. The guardrails are the same for every agent.
- Opt-out is honoredAsked to stop, the agent acknowledges and stops.
- Wrong person, no leaksIt never shares the intended person’s details.
- Identifies itselfAsked who is calling, it names the business and why.
- Tools mocked in testsBenchmark runs never touch your real systems.
Scenario benchmark
Role-played conversations your agent should handle, run on every channel it has and graded against written criteria. Each one runs 3 times.
A person can step in
Nothing waits on an agent that can’t finish.
Anything an agent can’t finish goes to Needs a person with the reason. Your team can tell the agent the answer, reply themselves, or take over the conversation; the agent pauses until they hand it back.
Needs a person
Nothing waiting
Nothing needs a person right now
When an agent can’t finish something, it lands here with the reason.
Access
People see the stores and the work their role allows.
Roles cover the group, each store and each agent, from owners who configure everything to viewers who only read.
Team
Product screens on this page show an illustrative store; customers, numbers and conversations are examples.
On the record
If it happened, you can see it.
Every conversation keeps its own evidence, kept with the customer and readable by your team.
Recordings and transcripts
Calls recorded and transcribed, with moments and a summary.
Decisions
Why a message was sent: what changed, why, what’s next, what was held back.
Event traces
Each step from intake to delivery, for when you need to dig in.
Delivery receipts
Delivered, or not, with the carrier’s reason in plain words.
Who acted
An agent, a teammate or the store, named on every booking and message.
Support from the console
Raise a ticket from the page you’re on; it carries the context.
Questions
What security teams ask.
Controlled.Checked.Answered.
Walk through the controls with us on the workflows you’re considering, and see the record each one leaves.