Amazon Web Services
Compute, storage, mail (Amazon SES), file scanning. The site and the cabinet run in the EU (Frankfurt); the agent's environment runs in the region you choose or in your own AWS account.
Trust
Trust: what happens to your data
Only what is true today: where the data lives and for how long, who takes part in processing it, how it is protected and what we do if something goes wrong. We hold no certifications yet — so we show what can be checked.
How long it is kept
Retention is a property of the kind of data, not a decision made at the moment of writing.
| What | How long |
|---|---|
| Full text of conversations with the agent | 30 days after the last message |
| Web chat history in the cabinet | 90 days |
| A conversation with the consultant on the site | up to 90 days |
| The agent's notes about the work | while you use the agent |
| Backups of the agent's environment | 35 days; encrypted and verified by restore every day |
| Logs of the functions and the agent's environment | 30 days |
| Site CDN logs (address, browser, page) | 90 days; only numbers go into analytics |
| A file from the request form before scanning | up to 14 days; a file with a threat is deleted at once |
Who takes part in processing
Three kinds of participants — and you choose the model and the channel.
The model provider
The one whose subscription or key you connected — Anthropic, OpenAI and others — or Amazon Bedrock in your AWS. A request is processed on that provider's terms.
The channel
Telegram, if you connected a bot. The cabinet's web chat works without intermediaries.
The site does not track you
Check it yourself: developer tools → Network.
No cookies on the pages
Only the cabinet sets one (the session after sign-in) and the consultant chat (to continue the conversation).
Not a single third-party request
No counters, no external fonts, no widgets. A network audit checks this on every release.
Statistics without scripts
Counted from the CDN logs as daily numbers — nothing runs in your browser for it.
Protection
A separate environment per customer
Its own machine, secrets, logs and backups. Data of different customers does not mix.
No inbound ports
The agent's environment accepts no incoming connections: management goes through AWS Systems Manager, questions through a queue.
Keys and tokens are write-only
Kept in AWS Secrets Manager. After saving nobody sees them, and they never reach the agent's memory.
Sign-in without passwords
A one-time code by email. Roles: owner and observer.
HTTPS and strict policies
HTTPS only, a strict content security policy, and the page cannot be framed by another site.
Files are scanned
Every file from the form is checked by Amazon GuardDuty; only a clean one goes to work.
The agent keeps its boundaries
Says it is an AI, does not follow instructions found in documents, does not remember special categories of data. Built-in tools are off — only the ones you grant.
Monitoring
CloudWatch alarms and a morning digest on the agents and the site.
If something goes wrong
- We record it and assess the level — from a leak to a single failed turn.
- First we stop the damage: the agent, a line or the chat — with one button.
- We tell you about a breach of your data's security without undue delay; the deadline is set in the data processing agreement.
- We review it within five working days: the cause, a new rule and the check that proves it.
When you leave
- We hand over an export of the agent's data.
- We delete the environment: the machine, secrets, logs and cabinet records.
- Backups expire on their own — within 35 days; the encryption key is destroyed after 30 days.
Honestly, about what we do not have yet
We hold no SOC 2 or ISO 27001 certification. The privacy policy, terms and data processing agreement are with our lawyer; we do not publish text that has not been agreed. Questions left? Ask in a request — we will answer to the point.