Data security & compliance

Your data stays on your server

Datachat’s core principle: sensitive data never leaves your environment. In basic mode the AI only receives the table structure — never actual data.

Built-in layers of protection

In basic mode only the table structure reaches the LLM, never an actual data row.

Deep analysis uses only limited, aggregated data (max 500 rows / 50 KB per step), with explicit permission and a full audit log.

The system only reads from the data warehouse — it never writes to the source database.

Only allowed tables and functions are accessible; every generated query is validated.

User-code filtering runs through parameterized queries — it cannot be manipulated.

Max 30 seconds per query, max 10,000 rows in basic mode.

Datachat fits your corporate identity management — from a development environment to MFA-protected cloud sign-in.

Smaller deployments, development — a built-in login form with password and optional TOTP 2FA.

Corporate intranet, Active Directory — automatic sign-in based on the AD account.

Public, cloud-based access — Microsoft or another IdP, with MFA.

Administrators control exactly who can access which data — down to the row level.

Access levels: user and administrator, with distinct privileges.
Table policy: which role can access which data-warehouse table.
User-code data filtering: “only my customers”, “only my items” — via parameterized, non-manipulable SQL.
Shared dashboards also enforce the user’s table permissions.

The placement of the language model can be aligned to your data-protection policy.

Private endpoint, GDPR-compliant — recommended for most customers.

Local LLM in basic mode, cloud for deep analysis — a balance of cost and privacy.

For a zero-cloud policy, every model runs on-premises.

We’ll show you a tailored demo built on your Octopus8 data warehouse. 30 minutes, no commitment.