Data and security
Your data, in plain words
What Wellence keeps, who can see it, how AI uses it, and how you take it out or delete it.
Last updated 23 October 2026.
What we store, and where
- Your company's teams, people (names, work emails, roles), goals, agents, tasks and their messages, files you attach, agent results, and usage and costs.
- It's kept in a Postgres database run by Supabase, in the EU (Ireland). The app's servers run on Vercel in London. Emails go out through Resend.
- If you connect your tools (such as Google, Slack or HubSpot), Composio keeps their sign-in tokens and handles the reads and saves. Composio is in the United States.
- Files you attach are stored privately and opened through short-lived links.
- We use PostHog (EU) to see how the app is used. All text, inputs and images are hidden, your content is never recorded, and nothing is stored on your device. Switch it off in Settings.
Who can see it
- Only people signed in to your company's workspace. Each company's data is kept separate from every other's.
- Inside your company, people see their own team and what's been shared with them. Admins see the whole company, except work an agent did from one person's own files (see below).
- A guest you add to one task sees that task's conversation and nothing else. Every message they write is marked as theirs.
- Wellence's team doesn't browse customer data. We'd only look to fix a problem you've asked us to help with.
What agents read
- Agents read files, email, messages and pages (such as Google Drive, Gmail, Slack or Notion) only through the own connection of the person they're working for: whoever asked, or the agent's owner for scheduled work. Never through the company's connection or anyone else's.
- Company-wide tools, such as analytics, ad platforms and your CRM, can use a team or company connection.
- When you pick the exact folder, file, sheet or channel in an agent's setup, it reads only that. Any other read is refused before it reaches the tool.
- Each read is recorded with the tool and the item, never its content. Refused reads are recorded too.
- Work an agent did from one person's own files is shown only to that person, admins included, until they share it. It's left out of the company's download.
- What agents read from your tools, websites and files is marked as outside content. Agents are told to treat it as information, never as instructions, and reading can't send, change or delete anything.
How AI is used
- Wellence's AI runs on Anthropic's Claude by default. Interviews and Get help always run on Claude.
- A company can choose OpenAI's ChatGPT (on Microsoft's Azure OpenAI Service, or OpenAI's own API when Azure isn't available) or Google's Gemini instead, or run agents in its own AI. If it leaves the choice to Wellence, some work can go to ChatGPT or Gemini when Wellence has them switched on.
- Each provider's terms for data sent through its API are below. On the plans Wellence uses, none of them train on it, and nor do we.
- Anthropic's API runs in the United States only. It has no UK or EU processing option.
- Agents work from what your company has set up in Wellence: its goals, teams, the context you've written and the task in hand.
- Trying a draft agent before it's made uses one of your own recent tasks or text you paste. It doesn't open your connected tools, sends nothing, and its result isn't saved.
- You can use your own Anthropic, OpenAI or Gemini key instead of ours. Keys are stored encrypted (AES-GCM) and never shown again, not even to admins.
- Or run agents' AI in your own cloud account: Claude on AWS Bedrock or Google Vertex AI, or your own Azure OpenAI. The work then stays in your account and region, under your own contract with that provider.
- An admin can switch on "Only use our own AI" in Settings. Then nothing falls back to Wellence's keys or another provider: if your AI doesn't answer, the work stops and says why. The table below shows which features use your key.
Connecting Claude
- When you connect your own Claude, Wellence only sees what Claude sends when it uses a Wellence tool, plus the names of the tools you've connected there.
- It never sees your other Claude chats or files, and your Claude only sees what you can see in Wellence.
- Claude asks before it sends, posts or changes anything. You can remove Wellence any time in Claude's settings.
- An agent run inside your Claude uses your Claude's model and connectors. Its owner, approvals, run history and data rules stay in Wellence, through the Wellence connector. What Claude does outside that connector isn't recorded in Wellence.
Your own AI
- Enforced in Wellence: who can run an agent, who can see a run, and that the owner approves finished work. A run from your Claude waits for review like any other.
- Only logged: which tools Claude says it used. Wellence cannot check this, and cannot limit what Claude does with its own connectors.
- Not applied in your Claude: Wellence spend caps, credits and the picked-items rule. Work there runs on your Claude plan and your own logins, so it shows as "no credits used".
- Only people who have connected their own Claude see Continue in Claude. Copy context is plain text: only paste it into an AI your company has approved.
Data other apps send in
- A workflow can start when another app (Zapier, n8n, a form) sends data to that workflow's own private link. It's off until someone sets it up.
- Signed by default: each request must carry a signature made with that workflow's secret, and is refused if it was signed more than 5 minutes ago. Each event is accepted once, so a replayed request starts nothing. The secret is stored encrypted and shown once.
- A simpler link, with the secret in the link itself, is there for tools that can't sign. It's marked as less safe, and a new link can be made at any time.
- Each link takes at most 60 requests a minute and 1,000 a day, and 256 KB per request. A workflow started 20 times in an hour pauses itself and tells its owner; one that asks first keeps at most 20 cards waiting.
- A workflow can also start from an event in a connected tool: a new Typeform response, new rows in a Google Sheet, a new email in Gmail or a new deal in Pipedrive. It listens with the connection it would read with: the person's own for Typeform, Sheets and Gmail, the company's for Pipedrive. Events come from Composio signed with our secret, are checked against the trigger and connection they were set up on, and have the same limits. Removing the trigger stops the listening.
- A workflow can save to a connected tool: add a Google Sheet row, update a HubSpot contact or deal, create a Notion page, add a Pipedrive note or an Airtable record. Never anything that deletes. It saves only to the item picked in the step, or the one record a Find step in the same run found, checked in code, with the connection of the person the run works for.
- Each change is shown first and waits for that person to approve it in their Inbox, unless the workflow's owner lets it save on its own. Every change is kept in a record of changes: who it was for, which tool, which item, and whether it was made, refused or failed. It's made once, even if a step runs again.
- What comes in is treated as data, never as instructions: agents see it marked as untrusted. By default the workflow's owner is asked before anything starts, and an admin can stop every trigger at once.
Getting in
- You sign in with a link sent to your work email. There are no Wellence passwords to leak.
- Sign-in links are signed, expire, and work once.
- New people who sign up with your company's email domain can be made to wait until an admin lets them in. New companies start this way, and admins choose in Settings. People you invite join straight away.
- The app sends security headers that stop other sites showing it inside their pages and block scripts that aren't ours.
Your data stays yours
- An admin can download everything your company has in Wellence, any time, from Settings. The download leaves out secrets (keys, sign-in tokens and connection ids) and work from someone's own files that they haven't shared.
- An admin can delete the company from Settings. That removes every record and file it has in Wellence, disconnects its connected tools and cancels its subscription. The admin gets a receipt of what was removed, by email too. It can't be undone, and it's gone for everyone.
- Invoices and payment records stay with Stripe, our payments provider, for as long as tax law requires.
- Remove a person any time, and they lose access straight away.
What AI providers do with your data
From each provider's own terms for its API, checked 22 October 2026.
| Provider | Their terms | Source |
|---|---|---|
| Anthropic (Claude) | Doesn't train on API data. Deletes it within 30 days; content flagged for misuse can be kept for up to 2 years. Runs in the United States only: there's no UK or EU option. | Anthropic privacy centre |
| Microsoft (Azure OpenAI Service) | Doesn't train on it and doesn't share it with OpenAI. Content flagged for misuse can be reviewed, by staff in the EEA for EEA deployments. Runs in the Azure region Wellence chose. Wellence asks it not to store responses. | Microsoft Learn: data, privacy and security |
| OpenAI | Doesn't train on API data unless the account opts in. Keeps it for up to 30 days to check for misuse. Wellence asks it not to store responses. | OpenAI: your data |
| Google (Gemini API) | On the paid API, which Wellence uses, doesn't use prompts or answers to improve its products, and keeps them for 55 days to check for misuse. Google's free tier can use them, so a company's own key should be on a paid plan. | Gemini API terms |
If your company runs its agents' AI in its own cloud account (Claude on AWS Bedrock or Google Vertex AI, or its own Azure OpenAI), that provider works under your own contract with it, not as Wellence's sub-processor. The work is sent from Wellence straight to your account, in the region you chose.
Whose AI key each feature uses
Get help, and what runs on our website before you sign in, are always on Wellence's key. Everything else runs on your company's own AI once you've added it. Before agent work runs on Wellence's AI, a quick check on Wellence's Claude decides what kind of work it is; with "Only use our own AI" on, that check uses your Claude key, or is skipped.
| What it does | Whose AI key |
|---|---|
| The AI diagnostic on the website | Always Wellence's |
| The AI team preview from a public website | Always Wellence's |
| The help assistant | Always Wellence's |
| The offline replay's blind judge of today's way against the gate | Always Wellence's |
| A personal landing page | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| A personal page's AI team, from what its research already found | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| A personalised demo (its reading steps on the helper model) | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| A personalised demo's pitch | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| A sample run while hiring (the agent's model, or Automatic) | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| A suggested answer, or a starter draft, for a run question | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| A task's plan from its brief | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| A team's goal drafts from a company key result | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| A team's or the company's context | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| A test run of one workflow stage | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| An agent from one sentence | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| An agent picking a Decision's path from the answers so far, with a reason and how sure it is | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| An agent's reply in a goal discussion | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| An agent's reply on a task | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| An agent's run, in one call | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Changing a draft agent in plain words | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Changing an agent from a request | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Checking a goal is SMART, inside an interview | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Checking what a cheap model delivered, before it's kept or redone | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Checking what exists before building | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Company goal drafts from a memo | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Drafting a suggested update to an agent after the company's context changes | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Drafting a workflow from a description or SOP | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Drafting one field of an agent | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Filling a Tool step's named fields from what was read in its tool | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Fitting a workflow's actions to a brief | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Gathering a run's data from connected tools | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Goal drafts under a parent key result | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Guidance on filling in a to-do's blanks | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Hiring an agent | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Keeping a task's plan up to date | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Laying out agent work (the flagship once if it comes back garbled) | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Laying out an agent's skills | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Matching a new agent's skills to ones the company has | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Matching a task to agents | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| One step of an agent's run (a skill) | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Reading what an approved draft would send (emails, invites) before the person presses Send | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Routine or judgement, for an agent on Automatic | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Setting goals | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Setup from a website or LinkedIn page | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Splitting big work with no skill plan into parts (off for now) | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Suggested answers for a to-do's blanks | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Suggested answers to an agent's questions | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Suggesting the numbers a team should track | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| The AI diagnostic, signed in | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| The client-facing version of work | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| The few facts to ask for before drafting a task | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| The instant agent card while someone types | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| The next question in the New workflow interview | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| The plan shown at the ask, before the task exists | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| The Task helper answering a question on a task | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Tidying the words around an answer put in a document | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Transcribing a client's file | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| What a run needs to be given | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| What a task's plan reads from the company's systems | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| What an agent needs from the company's systems | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Which agent or team owns each key result | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Which saved workflow a task fits, when the rules cannot tell | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
| Writing a memo | Your company's AI (your key or your cloud), when you've added it (else Wellence's) |
GDPR
Your company is the controller of the data it puts into Wellence, and Wellence is its processor under our Data Processing Agreement. Your database is stored in the EU. Our GDPR page lists every subprocessor, how transfers are covered, and how people use their rights.
How agents are kept in check
What agents can and can't do without a person, who's responsible, where AI runs and how we think about regulation: Trust and AI.
Report a vulnerability
Found a security problem? Email support@wellence.ai with what you found and how to reproduce it. We'll reply and keep you posted until it's fixed. Please don't access other people's data or disrupt the service while testing. Our security.txt has the same details.
Questions, or a security review?
Email team@wellence.me. We'll answer your questionnaire and walk your team through it. Read our Privacy Policy and Terms of Use.