What Datis can reach, and what it cannot
Datis reads your databases, repositories and cloud accounts, and its agents write code against them. This page says where that access starts, where it stops, and how what we hold is protected. It is written to be checked, not admired.
The short version
The five sentences that matter — the detailed version arrives with general availability.
One connection, opened outward from your network. Datis never asks you to open a port, and there is no inbound path from us to you.
Every connection is created read-only. Only a workspace Admin can change that, only after re-authenticating, and the change lands in the audit log with actor, time and IP.
Every statement runs with a 30 second timeout and the row cap you set on the connection. A runaway question fails; it does not page you.
Nothing at all goes to a model provider for training. Inference only, zero retention.
Connection credentials are sealed before they touch disk. There is no code path that hands one back.
What the agent may never do
These are not settings you have to go and find. They are the floor, and the ones that come from policy can only be made stricter.
Push to a protected branch
Touch a path outside the allowlist
Run a migration, or any statement that is not a SELECT
Read the value of a secret
Apply infrastructure without a named approver
Exceed the diff-size cap or the spend cap
Open a pull request on a failing test suite
Send your code or your data to a model provider for training
Sub-processors
The companies that process customer data on our behalf. We announce a change 30 days before it takes effect, and you can object in writing.
Which provider runs the model behind Ask and the agents is your choice, set per workspace — Google Cloud is today’s default, with AWS Bedrock among the options.
The list with effective dates will ship as an appendix to the DPA. Write to contact@datis.app to be told about changes before they happen.