Updated 13 October 2026. How Tenure handles your customers' data, your billing, and your AI agents.
Customers can always leave. Every screen of the cancel flow has a full-size way out, and the widget repairs any exit a host page hides or covers, and reports it. A one-step “Cancel now and skip the questions” button shows for every customer by default. With the widget or SDK, if Tenure is unreachable or a decision takes too long, cancellation is handed back to your app's own cancel path. A hosted cancel page is a Tenure page, so it has no path of its own to fall back to: when it can't load the customer's options it offers a request to cancel instead, which is recorded and you are told, and if Tenure itself is down the page doesn't load.
Every record is scoped to an organization id and every query is filtered by the signed-in user's organization or the API key's organization. Public widget keys can only create and resolve cancel sessions.
Passwords are bcrypt-hashed. API keys are shown once and stored as SHA-256 hashes; they carry scopes, an optional expiry and a test-mode flag. Sessions are encrypted, HttpOnly, SameSite cookies that are invalidated on password change, on password reset and by "Sign out everywhere".
The credentials you connect (Stripe, Chargebee, Resend and PostHog keys, live and test billing keys, and webhook and widget signing secrets) are encrypted at rest with AES-256-GCM under a key (SECRETS_KEY) that lives only in the hosting platform's secret store, never in the database; Tenure uses no external key-management service. Keys can be rotated with a documented script; the dashboard shows only the last four characters.
| Can change | owner | admin | developer | member | viewer |
|---|---|---|---|---|---|
| Customer journeys | Can change | Can change | ViewView only | ViewView only | ViewView only |
| Customers | Can change | Can change | ViewView only | ViewView only | ViewView only |
| Talk to us requests | Yes | Yes | ViewView only | Yes | ViewView only |
| Billing and integrations | Can change | Can change | ViewView only | ViewView only | ViewView only |
| Team | Can change, incl. roles | Can change | ViewView only | ViewView only | ViewView only |
| API keys | Can change | Can change | Can change | ViewView only | ViewView only |
| Exports and audit log | Can change | Can change | No | No | No |
| Settings | Can change | Can change | ViewView only | ViewView only | ViewView only |
Five roles gate every configuration action server-side: owner (everything, including roles and deleting the workspace), admin (configuration, integrations and invites), developer (reads everything and manages API keys and webhooks, but not billing execution settings or the team), member (works cancel attempts, settings read-only) and viewer (read-only). Owners and admins can restrict the dashboard to an IP allowlist (CIDR). Changing someone's role signs their existing sessions out. Settings › Team shows what each role can do.
The app has no staff login, admin console or impersonation: no one at Tenure can open your workspace in the app unless you invite them. Stored data can otherwise be reached only through Tenure's own hosting and database accounts (Vercel and Supabase), to run and repair the service.
Agents can propose; only people approve. API keys, delegated tokens and MCP connections carry the Read, Runtime and Draft scopes (ids read, decide and draft_write), as Settings › Developer names them. There is no publish scope: no key or MCP connection can approve, confirm or switch on anything.
Offers execute only from your published plan, only after an explicit accept, only server-side. Agents receive guardrails (max discount, no improvisation) and never reach your billing provider's API themselves: a key with the Runtime scope can record a customer's acceptance or issue a refund, and Tenure applies it through the same checks, counted against your Spending and safety budget. On a live page, Tenure changes billing only for a customer whose id your server signed with your widget secret; unsigned pages get no billing changes, and cancelling uses your own path. With the widget or SDK, your own cancel path still proceeds if Tenure is down.
Signed with HMAC-SHA256 over timestamp + body (x-tenure-signature, x-tenure-timestamp) so your backend can verify origin and reject replays. Every delivery is recorded with status, attempts and response code; failed sends retry with exponential backoff (up to 8 attempts) and can be replayed from the Events sent list in Settings › Developer. Stripe webhooks are verified with your signing secret.
| From | What | Kept |
|---|---|---|
| From your billing | Email address and name, plan, price, seats, status, trial and invoices | Until you erase the customer or delete the workspace |
| From your app | Customer id, languages, the events you send, page views of identified customers (can be switched off) | Until you erase the customer or delete the workspace |
| With each cancel attempt | Region, the reason picked, what the customer wrote (minimised), what they were offered and chose | Typed words 180 days; the rest until erased |
Tenure keeps what Settings › Customer data and privacy lists. From your billing: who each customer is (email address and name), their plan, price, seats, status, trial and invoices. From your app: their id, the language they were shown and the one their browser asks for, the events you send, and, for identified customers, page views with the page path and title (set data-pageviews="off" on the widget script to stop them). With each cancel attempt: the region it came from (country, and state in the US, from the hosting platform's location lookup) to apply cancellation rules, the reason picked and what the customer wrote (minimised, see below), and what they were offered and chose. Customers can be erased with all their sessions from the dashboard ("Erase customer" on the customer page); the API exports a customer's data but does not erase it. Owners can export the whole workspace as JSON or delete it outright (all child records cascade).
Personal details are removed before any model sees a customer's words. Email addresses, links, phone numbers, card-like numbers and the customer's name are removed in code, and at most 1,200 characters are sent. A hardship answer is never stored. Customers' typed words are kept for 180 days.
When you connect your own database (Postgres, Supabase or MySQL; Settings › Your product data), Tenure first lists table and column names through a read-only login over verified TLS. If you choose to read examples, it reads up to five rows and thirty columns from that selected table or view to help identify customer IDs; these examples are not stored. After setup, it reads only the view or table and columns you pick. It doesn't copy your table: when a decision needs a customer (a cancel click, an offer, an email about to send, their page in Tenure) it looks that one customer up by ID, and scans for quiet customers or usage limits ask your database for the matching IDs only. It keeps just each matched customer's latest looked-up values, overwritten on the next lookup, to reuse for 10 minutes and to fall back on when your database doesn't answer. Rows that match no billing customer are never stored. Disconnecting deletes the credentials and those values.
When you connect PostHog (Settings › Your product data), Tenure keeps a personal API key with only the Query Read scope, encrypted like your other credentials, and asks PostHog only about the events you pick: a customer's daily counts when a decision needs them (reused for 15 minutes, and as a fallback when PostHog doesn't answer), and about once an hour which of your customers went quiet. Segment sends your calls to a webhook for your workspace, signed with a shared secret; Tenure counts only the events you pick for customers Tenure knows, per customer and day for 62 days, and drops everything else at the door, event properties included. Message ids are kept as hashes for 62 days to prevent retries from counting again while their usage remains in the reporting window, and a user id learned from an identify call is kept only as a hash. Disconnecting deletes the key or secret and the counts.
The churn report reads no names or emails: reading your Stripe or Chargebee history never asks for customer records. History rows are deleted 25 months after each subscription ended, and the comments customers typed when cancelling in Stripe are kept for at most 180 days.
Setup reads only your public site and the documents you add, never anything about your customers. A file you upload is read to draft your plan and is never stored.
In short: typing an address on trytenure.co makes Tenure read that site's public home page once; nothing is linked to you, and the count of reads from your IP address is kept one hour. In detail: on trytenure.co, the site field can show a sample cancel screen before making an account. Preview cancel screen keeps you on the page; Draft my plan opens free account setup. The preview uses the site's look when we can read it, or Tenure's default look when we cannot. Tenure's servers then read that site's public home page (and a sign-in page it links to, for its look) with their stylesheets, under the same rules as setup's read, and send back only the look, the site's name and the address of its logo. No Firecrawl, no AI model and no account are involved. The read is kept in server memory and by the hosting CDN for up to 6 hours (10 minutes when the site couldn't be read), so the same site isn't read again; if you then sign up with that site, setup may reuse it. Your IP address (for IPv6, its /64 network) is used only to count how many sites you have us read, up to 6 an hour: the count is kept for that hour and isn't linked to an account or to the site you typed. Your browser loads the site's logo directly from that site.
Every decision, execution, email and admin change is written to a hash-chained, verifiable audit log per organization: each row's SHA-256 hash covers the previous row, so deletions or edits are detectable with the Verify integrity check. Sign-ins and failed sign-ins are logged with their IP address. The log is exportable as CSV or JSON by owners and admins, and retention is configurable per workspace (default 365 days; 0 keeps it forever). Rows past that age are purged nightly, and the purge itself is written to the log. Requests carry an x-request-id that is echoed in responses and structured logs for correlation.
Results reports what was observed: customer moments, offers accepted, billing changes confirmed and money collected, each number labelled with its kind and the count it rests on (a rate is withheld below 20). Each accepted offer's forecast is shown beside what billing actually collected once its period has ended. No one is held out as a comparison group: Tenure doesn’t run one yet at any size (one would need at least 1,000 eligible accounts and 40 eligible customers a month), so Results never reports how much Tenure caused. Test-mode and simulated activity is never counted in live figures.
Tenure offers only what you published: each moment follows your plan's lines, their rows and your approved limits, and code applies every eligibility rule, cooldown and budget. The decision model answers typed questions about the moment (see Subprocessors); it never invents an offer. Tenure doesn't train models on your customers' data. TypeSafe and OpenAI process what they receive under their API terms; whether those terms allow training, and how long they keep inputs, isn't yet confirmed for Tenure's accounts, so ask us before sending customer text. You can disable “What they write” in Settings › Customer data and privacy; this stops customer-text processing, not all model processing.
Each subprocessor above is on Tenure's own account. The services you connect (Stripe or Chargebee for billing, and your own Resend for customer email) are yours: Tenure sends them what billing changes and email delivery need, with your credentials; with your own billing, that billing data goes, signed, to your own billing service or backend instead. Tenure doesn't train models on your customers' data. TypeSafe and OpenAI process what they receive under their API terms; whether those terms allow training, and how long they keep inputs, isn't yet confirmed for Tenure's accounts, so ask us before sending customer text. No analytics or error-tracking service receives your data.
TypeSafe (typesafe.ai)
OpenAI
Firecrawl (firecrawl.dev)
Resend (Tenure's own account)
Vercel
Supabase
The hosted service runs on Vercel in the United States (Portland, Oregon) and stores its data in a Supabase Postgres database in US West (Oregon). Workspaces share that service; every record is scoped to its workspace (see Tenant isolation).
None is claimed on this page. Ask security@trytenure.co for the current status of certifications (such as SOC 2 or ISO 27001) and penetration tests.
Found a vulnerability? Email security@trytenure.co (see /.well-known/security.txt). Please do not test against other tenants' data.