Your data stays yours.
AI you can put in front of patients, clients, and the books, without giving up control of your data, your brand, or the decision.
Six things we promise, on every line.
Never sold, never trained on
Your data is never sold, and not used to train our models.
Provider training is disabled where the provider offers that setting
Each AI provider is used with training on customer data disabled where the provider offers that setting (see /subprocessors).
Human-checked
A person reviews what the AI does.
You stay in control
Nothing sends, pays, or commits on your behalf without your say-so.
A full audit trail
Every action, and the reasoning behind it, logged and reviewable any time.
Tested before it's live
We test it against your real calls and records before it goes live.
On your own ground, if needed
For regulated or sensitive work, the AI can run on hardware you control.
Where it lives, who touches it, how long.
The short version is on the trust page.
The detail
The platform runs on one server in Germany (Hetzner, Falkenstein) and the website is delivered by Cloudflare. Canadian and US regions are on the roadmap, not available today. On a Sovereign deployment the storage is yours.
The detail
Everything is encrypted in transit (TLS). At rest: server-side encryption of stored secrets and backups. The application server's disk is not full-disk encrypted today; we say so rather than round it up to "encrypted at rest".
Model-backed drafting and natural voice are separate paths
The platform's rules-based web, messaging and Twilio voice workflows answer through one shared resolver, with no model call. Model-backed drafting and ElevenLabs natural-voice agents are separate paths. The providers used by your workspace are disclosed in the product.
The detail
We use a named, limited set of providers (hosting, voice, models, payments). The full list is published, with what each one does and where it processes, and the product shows you which of them your own workspace uses. Our data processing agreement is a template, signed on request.
The detail
Model-backed drafting runs on a model key you bring, stored encrypted on your workspace, or on no key at all, in which case that drafting stays off and the product says so. The rules-based resolver makes no model call. Natural-voice agents use a separate provider path.
The detail
Owner, approver and member roles, invited by email from inside the product. A draft is approved by an owner or an approver; a member cannot approve. Only an owner manages the team, and the last owner cannot be demoted or removed.
The detail
Consent is recorded per person, per channel. STOP on any channel revokes every channel, HELP is answered, and a message with no consent record is refused and logged rather than sent. Every outbound channel starts switched off.
Consent, opt-out and audit records are retained separately
You set a retention period for eligible workspace records. Consent records, opt-outs and audit records are retained separately to preserve the record of permissions and decisions. See the trust page for account-deletion details.
Deletion runs after a 30-day window
Export everything as JSON from inside the product at any time, secrets redacted. Deletion is scheduled from inside the product after you have taken an export, and runs after a 30-day window in which you can cancel it.
The detail
Every workflow runs on your timezone. Business hours and schedules are read in your local time, not ours.
The detail
Call recordings, bookings, and records are yours. They are not pooled, not resold, and not used to improve a shared model.
The detail
On the first turn of a call or text our agents say they are AI and name your business, in line with the EU AI Act's transparency rule (Article 50) and good practice everywhere. You can turn the line off, at your own liability.
Every channel lands in the same resolver. Safety is checked before anything a client can configure.
Scroll the diagram sideways to see the whole path.
How to reach us, and how fast.
These are commitments, not aspirations.
Outside those hours
Monday to Friday, 9am to 6pm Eastern. Anything that has stopped a client's business working is handled outside those hours too.
Larger deployments
Clients on a larger deployment get a named contact and a direct number as part of the engagement.
The status page and the public checks
The website, the platform and the public demos are monitored every five minutes. What is covered, how incidents are communicated and how to report one are on the status page; the live checks are public at stats.uptimerobot.com. You do not have to ask us whether something is down.
SLA terms are published on request
Business tier SLA terms are published on request until the status page carries ninety days of measured history. We do not print a percentage we cannot yet show the history for.
We would rather commit to numbers we control than to one we cannot enforce against our own providers. Larger deployments can have availability and response terms written into the statement of work.
Controls your team can review.
Specific requirements are assessed during scoping.
We state what we actually do, not badges we haven't earned.
The on-premise option
For health, legal, and financial work where data sensitivity is highest, the on-premise Sovereign option keeps everything inside your own walls.
Health, legal, and finance: the sensitive ones.
You get the strictest setup: human review on everything sensitive, the on-prem option, and AI that captures and routes but never advises.
The security questionnaire, answered.
The CAIQ v4 domains and a SIG Lite-style short set, 96 answers marked Yes, No, Partial or Not applicable with the evidence for each. Including the ones where the answer is no.
Where draft-only is enforced
Anything that pays, posts or sends is drafted for a named person to approve, and draft-only is enforced in the server and the agent container, not only in the prompt.
All 96 answers, by domain, with a CSV you can paste into your tool → Anything not answered there, ask and we will answer it in writing, including where the honest answer is that we don't do it yet. Send your requirements → Last reviewed 2 September 2026.
Have a security question? Ask it directly.
Send us your requirements, your industry's rules, or your IT team's checklist.