What waits for you, and what does not
An agent that can send email, pay and post has to be wrong safely. These are the gates, channel by channel, and nothing more is implied.
Waits for a person by default
- From Kaer Mail or your connected Gmail, held with a preview of what will go out.
- Slack messages
- Held for a person before they post.
- Social posts
- Held for a person before they publish.
- Payments in its browser
- A click the agent's browser recognises as a payment is held until you approve it.
- Anything its safety check flags
- Shown to you as an Approve or Deny card instead of being quietly skipped.
The agent finishes itself
- Reading and drafting
- Mail, documents and pages it has access to.
- Work on its own computer
- Browsing, running commands, building.
- Forms you asked for
- In the browser, a click that would spend money waits for you; other steps you asked for, like filling in and sending a form, the agent finishes itself.
- Routine replies, if you choose
- You can choose to let routine email replies or posts go out automatically.
- Everything but paying, if you turn permissions off
- The account owner can switch from Auto to Permissions off under Approvals in Settings, behind a warning. Paying and booking still wait, and every hard limit stays.
Its own computer, not yours
The agent works on its own isolated computer, with its own disk and browser, walled off from other customers' machines. When a task ends, its machine is released.
Yours to revoke, and written down
You grant each connection yourself and can take any one back. Each task keeps the steps it took, and your account keeps its security events.
Granted by you, revocable at any time from the connectors screen without touching the others. Tokens and credentials are encrypted at rest.
Each task keeps the steps the agent took and the tools it used, so you can see how a result was reached.
A timestamped record of sign-ins, sessions and other security events on your account.
Your data, your switch
Your correspondence, documents, customer records and code are the context the agent works from. While “Help improve our services” is on — it is on by default, and you can turn it off any time in Settings — Kaer uses how you work with it, including prompts, messages and coding actions with secrets removed, to fix problems, improve its services and train its own models. Nothing is used to train third-party models.
- Improvement is your switch
From 22 October 2026, “Help improve our services” is on by default: usage, prompts, messages and coding actions, linked to your account with secrets removed. Turn it off in Settings and collection stops and the samples already kept are deleted. Private chats are never collected.
How it works - Where it lives
Kaer's servers and primary storage are in the EU, and Kaer processes your queries in Germany. The AI providers that process prompts are listed in the Privacy Policy, each with its country.
See the providers - A DPA for every business
If you use Kaer for business, our Data Processing Addendum applies automatically, on every plan. Nothing to request or sign.
Read the DPA - Zero data retention for businesses
In the Kaer app, on Enterprise, zero data retention (ZDR) means prompts, files and outputs are processed, then not stored once the work is done. For the API, see API data retention below.
Ask about ZDR
API data retention
The API at api.kaer.ai has two data modes, chosen per request or per account: standard, the default, and zero data retention (ZDR). In both, Kaer never keeps the model's output and never uses API prompts or completions to train any model.
Standard, the default
- Prompt
- Until 13 December 2026, not kept. From 14 December 2026, kept for up to 30 days for abuse and safety review, then deleted.
- Completion
- Never kept.
- Client IP
- A keyed hash, never the IP itself, cleared after 90 days.
- Request metadata
- Time, endpoint and model, status, token counts, cost, the key used and the country of the call, for billing.
- Training
- Never used to train any model.
- Price
- The standard rate.
Zero data retention (ZDR)
- Prompt
- Never kept.
- Completion
- Never kept.
- Client IP
- No hash kept.
- Request metadata
- Time, endpoint and model, status, token counts, cost and the key used, for billing. No country.
- Training
- Never used to train any model.
- Price
- A premium over the standard rate, shown on API pricing and in GET /v1/models.
How to choose zero data retention
Zero data retention on the API is for Enterprise accounts. A ZDR request from any other account is refused.
- ZDR base URL
Send requests to
https://api.kaer.ai/v1/zdr, for examplehttps://api.kaer.ai/v1/zdr/chat/completions. - ZDR header
Add
X-Kaer-Data-Mode: zdrto any request. - Account setting
An account set to zero retention uses ZDR for every request it sends.
- Model providers
In both modes, the AI model provider that serves a request processes the prompt to produce the completion. Each one is listed, with its country, in the Privacy Policy.
See the providers
The prompt is the request's messages and tool definitions; the completion is the model's output. The standard-mode prompt rule takes effect on 14 December 2026. Until then, the API keeps no prompt and no completion in either mode. The Kaer app has its own rules, including “Help improve our services”, set out in the Privacy Policy. Read the full API data policy
Questions
Are you SOC 2 or ISO 27001 certified?
Can the agent send email as me without approval?
Can I turn the approvals off?
What happens to the machine a job ran on?
Where is my data stored, and who processes it?
Do you offer zero data retention (ZDR)?
Does Kaer train on API prompts or completions?
Do you sign a DPA?
How do I revoke access?
How do I report a vulnerability?
Bring your security questions.
We would rather answer the hard questions before you buy than after. Book thirty minutes and ask them.
