Security and data
Your work stays in your AI tool. Only a result you approve leaves it, and for coaching a record of the method offered.
The test and coaching in your AI tool run under your account and on your subscription. Only the test result you approve leaves it, and for coaching a record of the method offered. Your chats, files and code never leave your tool.
Data flow during the test
- 01ConnectYou connect the add-on with your own account. You give us no password or API key.
- 02ReadYour AI tool reads your chats from the last 7, 14 or 28 days. We cannot see into them.
- 03EvaluateYour model writes the report, not ours. We run no AI of our own over your chats.
- 04ApproveYour tool shows you the whole report and asks whether to save it. Nothing is sent without your “Yes”.
- 05SaveOnly the result reaches us: level, score and generalised observations. We store it in Frankfurt.
- Connection. You sign in with your own account. You give us no password and no API key, and on the consent screen you see what the plugin may and may not do.
- Reading. Your AI tool reads your chats from the last 7, 14 or 28 days. We cannot see into them.
- Evaluation. Your own model writes the report according to our rules. We run no AI of our own over your chats.
- Approval. The tool shows you the whole report and asks whether to save it. Silence or “fine, go on” does not count as consent, and you can correct the report or not save it at all.
- Saving. What reaches us is a report with the level and score, and 3 to 8 generalized insights it is based on. It is stored in a database in Frankfurt. If the report contained a password or a card number, the server would reject the whole thing.
Data sent and not sent
From the test, after your “Yes”
- Level and score
- The report text you saw
- 3 to 8 generalised observations the report is based on
- The name of the AI model that judged the test
From coaching
- Which method coaching offered you
- The area, mode and time of the offer
Nothing else. It only keeps coaching from offering you the same thing next time.
Never, it stays in your tool
- Your chats
- Files and documents
- Code
- Prompts and coaching answers
We do not store passwords or card numbers; the server rejects such a result in full.
After your “Yes”, the test sends the level and score, the text of the report you saw, 3 to 8 generalized insights and the name of the AI model that evaluated the test. From coaching, only this is sent: which method was offered to you, in which area, in which mode and when, so coaching does not offer you the same thing next time. Your chats, files and documents, code, and coaching prompts and answers are never sent.
The plugin deletes nothing, sends no emails or messages, and cannot see your colleagues' accounts or company overviews. Your tool is told to read the history only as source material, never as instructions, and every save needs your explicit “Yes”.
Where data is kept and how it is protected
We store data in a data center in Frankfurt, in the EU. In transit we use TLS 1.3, and stored data is encrypted with AES-256. Providers of language models are contractually bound not to train on it. The plugin's sign-in lasts 10 minutes and renews itself for up to 30 days, and we do not store sign-in tokens in readable form.
We keep data for as long as you have an account. When you delete the account, the account, tests, reports, coaching and attachments are deleted immediately. Backups expire within 7 days and error records in Sentry within 90 days. You can export your data yourself in the app, in your profile under Data & privacy. If you pay for your subscription yourself, ask support to delete the account at podpora@aibility.cz, because of billing. We send the DPA and the list of sub-processors on request, and the public list is at aibility.org/legal/entity-list (in Czech).
Disconnect in one click
In the app, under Settings → Connections, you see which AI tools you have connected. After you disconnect, the plugin loses access immediately.
When you do not take the test in an AI tool
On these two routes your data is handled differently. With the test through a ready-made prompt, your own AI tool evaluates, and we store only the report you paste into the app. In a chat with the AI guide in the app, Claude models run through the OpenRouter service, and the model processes data in the USA. What goes to the model is the content of the conversation and the necessary context, such as your name, frequency or tools, and never your email, sign-in tokens or payment details.
Next: for IT and legal there is a page on the web with enablement steps for each tool, the DPA and the list of processors (in Czech).