Customer data under your control
Who sees what, what goes into the log, how projects are separated and what happens to a customer’s data when they ask to delete it.
Team access
Every team member sees and changes only what they need for their job.
Roles and custom roles
Six built-in roles and custom roles made of individual permissions; a member can hold several roles. Nobody can grant more than they have.
Access to what is needed only
A role can be limited to selected journeys, segments or integrations — the rest stay invisible to that member.
Two-factor sign-in
An authenticator app and recovery codes. An owner or administrator can make the second factor mandatory for every project member.
Invitations and revoking access
An invitation is a single-use link valid for seven days for the address it was sent to. Removing a member revokes access at once, open tabs included.
Sessions and sign-in log
Sign out of other devices with one button; changing the password closes the other sessions. The account shows sign-ins and failed attempts with address and time.
Launch approval
A separate permission to approve campaigns and journeys: an edit after approval sends them back for review.
Project isolation
A project is a separate unit of data: profiles, events, segments, journeys, integrations and keys.
Two layers in the database
Project data in PostgreSQL is separated twice: by an application filter and by Row Level Security policies in the database itself. Without a project context the policy returns no rows.
Fair queues and quotas
A heavy campaign or import in one project does not take resources from the others.
A brand on its own domain
A white-label brand has its own verified domain with an automatic certificate; sessions are bound to the brand’s domain.
Personal data
The tools to show customer data only to those entitled to it and to fulfil export and erasure requests.
Masking by permission
Email, phone and fields marked as personal are shown in full only to members with a dedicated permission — in lists, the customer card, segments, chat, the API and for AI assistants.
Consent before every send
Consents per channel and topic are stored with source and date and are checked before every marketing message — along with the suppression list, frequency caps and quiet hours.
Opt-in and opt-out
Double opt-in is on by default. Every marketing email carries an unsubscribe link and one-click unsubscribe; customers manage their topics in a preference center on your domain.
Erasure on request
A profile is erased with a button or an API request — together with its identifiers, events, chat history and membership in segments and journeys. It can also be removed from exported ad audiences.
Data export
One customer’s data as a JSON file; all project data as an export without keys or secrets.
Retention periods
Events are kept for 12 months on Starter, 24 on Growth and up to 60 on Enterprise. A deleted project can be restored for seven days, then its data is purged.
AI and customer data
AI gets exactly as much data as the task needs and does not act in place of a person.
No shopper data
Email copy and subject lines, template translation and search query parsing work without shopper data.
Own conversation, own role
The chat bot sees only the data of the customer in its conversation; the platform assistant sees only what the user’s role allows.
A person launches
The assistant drafts segments, emails and journeys; campaigns and journeys are launched by a person.
Assistants with role permissions
External AI assistants work over MCP with the permissions of the user’s role or the key; there is a read-only mode, and a connection can be revoked.
Audit log
Who changed what in the project, and when.
What is recorded
Settings, roles and invitations, keys, integrations, segments, templates, campaigns, journey publishing, profile export and erasure — with author, time and address.
API and AI assistants
Actions through the API are recorded under the key’s name; AI assistant actions carry a separate mark.
Immutable records
Log records cannot be changed or deleted; they are kept for 24 months.
Keys and the API
A key gives a system exactly the permissions you chose.
Two kinds of keys
A write key only sends events and can live in your site’s code — it can be restricted to your domains. A secret key is server-side, with selected permissions.
Secrets are not stored in the clear
A secret key is shown once and stored only as a hash. Integration credentials are stored encrypted and never sent to the browser.
Permissions checked on every request
A key’s permissions never exceed its creator’s: a member who was removed or downgraded loses access through their keys as well. Revoking a key takes effect at once.
Signed identity and webhooks
Your server signs the shopper’s identifier, so someone else’s profile cannot be injected from the browser. Outgoing webhooks are signed, and changing requests accept an idempotency key.
Application security
Measures that work regardless of project settings.
HTTPS only
The platform is served over HTTPS only; a content security policy restricts scripts on the interface pages.
Customer-supplied addresses are checked
Addresses of webhooks, feeds, data sources and custom mail servers are checked and cannot point into the platform’s internal network.
Sandboxed templates
User templates run in a sandbox with time and resource limits, and email HTML is sanitized.
Checks before every release
Tests, static analysis, a dependency check for known vulnerabilities and a secret scan; a finding stops the release.
File checks in chat
Documents customers send via the website, email and messengers, and operators’ documents, are scanned by the ClamAV antivirus; a file can be downloaded only after the check. Documents with macros, PDF scripts or embedded objects are rejected, and GIF and JPEG images in the media library and product photo copies are cleaned of extra data.
AI security audits
In November 2026 the platform code was audited for vulnerabilities three times: access and project isolation, the API, data ingestion, files, online chat, AI features and deployment. The latest audit was done with Claude Mythos 5.2. All high- and medium-severity findings were fixed.
Reliability
What happens when something goes wrong.
Backups
Databases are backed up daily, and PostgreSQL keeps a continuous log archive. Once a week a restore from backup is verified automatically.
Zero-downtime releases
A new version rolls out gradually, with no service downtime; background jobs already running finish their work.
Events are not lost
Intake goes through a queue with deduplication; the tracker and mobile SDKs keep what was not sent and retry.
Monitoring
Services and queues are checked every minute, and failures raise alerts.
Emails go out from your domain and pass mailbox providers’ checks.
Domain authentication
DKIM with automatic key rotation, SPF and a DMARC check; the domain’s records are re-verified every hour.
Reputation protection
Complaints and hard bounces unsubscribe the address automatically, and a campaign with a high complaint rate is paused.
A check before sending
An email is checked for sender authentication, content and a spam-filter score before the campaign starts.
Still have security questions?
We will go through the details in a demo and answer your security team’s questions.
Try it on your own data
14 days on the Growth plan, no card required: up to 1,000 profiles and 500 emails a day.