Security

Authentication & access

• JWT-based authentication with secure token refresh
• Role-based access control (RBAC) for user and admin roles
• Per-user dataset isolation users can only access their own data unless they share it
• Rate limiting on authentication and public share endpoints

Data protection

• Account passwords hashed with bcrypt
• Traffic encrypted in transit over HTTPS
• Uploaded files and reports stored privately and served only to authorized users
• Application secrets kept in environment configuration, never in code

Database connections & imports

• Database passwords and import links are encrypted at rest and never returned by the API
• Connections are tested before saving; we recommend a read-only database user
• Imports run in a read-only transaction with a statement timeout and row limit; only single SELECT queries are allowed
• Only public hosts can be reached private and internal network addresses are blocked

AI chat sandbox

When chat runs a query, it executes in an isolated in-memory sandbox that contains only the data from that analysis. Only read-only SELECT queries are allowed; file, network, and system access are disabled, and each query is limited in time, memory, and rows returned.

Sharing controls

• Nothing is shared unless you create a link or invite someone
• Share links are read-only, can expire, and can be revoked at any time
• Report downloads through a link are off unless you allow them
• Invited teammates must sign in with the invited, verified email address viewers can't comment, and only you can chat, re-run, or download cleaned data

Audit logging

Sign-ins, account changes, dataset uploads and imports, analyses, connection changes, sharing, invites, comments, billing, and admin actions are recorded in an audit log reviewed by administrators.

Infrastructure

Datana runs behind an Nginx reverse proxy with PostgreSQL, Redis, and isolated background workers for analyses and refreshes. Errors and performance are monitored, and failed jobs alert our team. Organizations with strict data-residency or network requirements can contact us about private deployments.

AI processing

Analysis insights and chat answers are generated with third-party AI (LLM) providers. If a provider is unavailable, Datana falls back to rule-based analysis your results still complete. Context sent for inference is limited to what the task needs column summaries, a few example values, computed results, and chat query results (capped at 200 rows) not your full dataset.

Reporting security issues

If you discover a vulnerability, please report it responsibly to [email protected]. Do not disclose publicly until we have had a reasonable opportunity to investigate and remediate.