Encrypted at rest
- Chat messages
- AI-written handover and ticket summaries
- AI memory for each chat
- LLM API keys
- Channel and CRM credentials
- Database connector passwords and SSH secrets
- Ticket titles, text and comments
- Customer names, emails and phone numbers
- Knowledge base content
- Some other integration settings
Protect these with database and disk encryption on your host.
Sign-in and keys
Stored as bcrypt hashes, with a strength check.
Short-lived access tokens, with refresh tokens that last 7 days.
Stored as bcrypt hashes, not as plain text.
A production install refuses to start with default secrets.
Every request is signature-checked, with a 5-minute replay window.
You can turn on a daily limit per visitor for each agent.
If you self-host
Chats and knowledge live in your own database. Embeddings run locally.
The shipped web server listens on port 80. You add HTTPS in front of it.
Message text goes to the model provider you configure, so choose one you trust.
What leaves your servers
This is the self-hosted picture. On our cloud, we run the database and the servers for you.
| Data | Where it goes |
|---|---|
| Message text | The AI model provider you configure |
| Embeddings for your knowledge | Created locally on your install |
| AI framework telemetry | Off by default |
| Your database | Stays on your infrastructure |
Connecting a database safely
AI ticketing can read Postgres or MySQL, so the rules there are strict. It runs SELECT queries only, on tables you allow, with a row limit. You can mask columns and scope rows to the customer on the ticket. Use a read-only database user as well, so the database enforces it too.
What we do not have yet
We are not SOC 2, ISO 27001 or HIPAA certified. If you need that today, choose a larger vendor.
Knowledge, users, channels and integrations can be deleted. There is no one-click delete for a single customer yet.
Advisories and reports
We publish security advisories on GitHub, including the ones that were our fault. If you find a problem, report it there privately.
Security questions
Is it end-to-end encrypted?
No. Messages are encrypted at rest in the database, and the app has to read them to answer.
Who sees my chat text?
Message text goes to the AI model provider in use, so its terms apply. On your own key or your own servers, you choose that provider.
Where can I read the code?
On GitHub. The core is Apache-2.0.