Trust Centre
AI works within the commercial boundaries you define.
Written for whoever has been asked to sign off on letting software contact your customers by email and telephone. Every section below says what is not true as well as what is, because a trust page listing only reassurances is the one nobody believes.
Calling policy
Agents identify themselves as AI at the start of every call, and that is not configurable. Hours, attempt caps, do-not-call lists and recording notices are evaluated by the platform before a call is placed — not described to a model and hoped for.
- AI identification at the start of every call, on every plan, not switchable off
- Calling hours enforced in the contact's own timezone, not the account's
- Attempt caps per contact per period, applied across every motion at once
- Do-not-call and suppression lists honoured before the number is dialled
- Recording notices played per the policy you configure for each market
- An agent asked for a person offers one immediately, without a second attempt
Consent and permission
Consent states are read from your consent system at the point of use, per contact and per channel. Nothing is cached here and no consent is recorded here. An absent or unreadable state is treated as a refusal.
- Read per contact and per channel, at the moment of use
- Absent or unreadable means no — failing open is not a configuration option
- A withdrawal takes effect at the next read, which is immediate
- Individual-level website behaviour is gated on the visitor's own consent
- An opt-out applies across every motion, channel and agent at once
- No agent can override a consent state, and no setting grants that
Commercial authority
What an agent may offer is enforced outside the model, before an offer is made. An instruction in a prompt is a preference; the refusal has to happen at the point of the offer or it is not a limit.
- Standard pricing an agent may send; a discount ceiling it may offer
- Beyond the ceiling, a named role approves — derived from your own people records
- Contractual terms and service levels are refused to every agent, on every plan
- Signing, committing and taking payment are refused outright
- Exceeding authority produces an approval request with its reasoning, never an improvised number
- An approval cannot be granted by an agent or by an API key
Agent identity
Every agent acts under its own identity, so every message and every call is attributable to a specific agent and a specific version rather than to a shared credential.
- One identity per agent, recorded on every action with its version
- A conversation from March is explained by the rules in force in March
- Pausing one agent affects only that agent
- Every action records whether a person or an agent took it
Credentials and isolation
Credentials for connected systems are held in a vault, scoped per connection, injected at call time and never placed in an agent's context. Isolation between organisations is enforced at the record level.
- Never in a prompt, in configuration or in agent code
- Redacted from logs, traces and error messages
- Rotation updates the vault and changes no motion
- Cross-organisation reads are not a permission anybody can be granted, including us
- Card data is held by our payment processor and never by us
Audit
Every consequential commercial action records its actor, the agent version, the inputs, the approvals obtained, the policy applied and the result — written as it happens rather than reconstructed afterwards.
- Every message sent and every call placed, with which agent sent it
- Approval decisions record who decided, when, and what they were shown
- Policy evaluations are recorded, including the calls that were not placed
- Append-only for its retention period, including for administrators
- Filterable by contact, company, motion, agent, channel and date
Data, recordings and retention
Retention is set per category, because the obligations genuinely differ between engagement records, call recordings, transcripts and attribution records. Call recordings usually warrant the shortest period of anything here.
- Per-category retention, configured by you
- Recordings default to the shortest period; transcripts are usually sufficient
- Recordings and transcripts are excluded from model training entirely
- Export of opportunities, motions, engagement, transcripts and attribution on every plan
- Your customer records stay in your customer system — they were never ours
- Records under legal hold are retained regardless, visibly and attributed
Models, and what we will not do
Model and voice providers are engaged per task under terms that prohibit training on account content. Which provider handles which task is a platform decision; on some plans you can supply your own.
- Your conversations, transcripts and commercial data do not train models available outside your account
- That is a contractual commitment, not a setting, and it binds our providers
- Content from outside — replies, attachments, event payloads — is data and cannot instruct an agent
- No view ranks your own people by emails sent, calls made or activity volume
- No hidden rating of the people you contact; priority is always stated as evidence
What this does not do.
It does not make you compliant with any marketing or telecommunications regime. It enforces the rules you configure — hours, disclosure, caps, consent, recording — and configuring them correctly for each market you operate in is your decision and needs your own advice. In a category where the penalties are real, a product implying otherwise would be selling you a false sense of security.
It does not establish a lawful basis for contacting anybody. The platform can read a consent state and refuse where there is none; it cannot tell you whether you were entitled to contact somebody in the first place.
It cannot judge tone. An agent permitted to email customers can send something clumsy, and no boundary catches that — which is why draft-first exists and why we suggest reading the first fifty calls and the first hundred emails.
It can only hold outreach on what is recorded. A complaint made by phone to somebody's mobile and never logged will not stop anything, and no amount of connected data fixes an unrecorded conversation.
It cannot unsend. Pausing a motion stops what has not happened yet; an email that has been sent is sent and a call that was placed happened. The platform records exactly what completed so you know what to follow up.
If an agent contacted you.
It was acting for the company whose name it gave, not for us. They decide who it contacts and what it may say, and they are the controller of your data — so ask them, and they must act on it. If you cannot reach them and believe something has gone wrong, write to [email protected] and we will take it seriously.
Reporting something.
If you believe you have found a vulnerability, write to [email protected] and we will acknowledge it. Testing against your own workspace and its demo data is welcome; testing against another organisation’s data is not, and is a breach of the acceptable use policy rather than research.
Make sure every opportunity gets a next action.
Let AI handle the repetitive engagement, keep follow-up moving, and bring your team in when a customer needs a real conversation.
Agents identify themselves on every call · Nothing commits without you · [email protected]