Privacy Policy

Last Updated: May 23, 2026

1. Data Ownership & Sovereignty

At AtlasLM, we treat data ownership seriously. Unlike standard cloud AI products, your uploaded source files (PDFs, Markdown, TXT) and extracted chunks remain strictly yours. We do not sell, scan, or train proprietary models on your document corpus.

2. Vector Encryption & Storage

Your source documents are parsed page-by-page, chunked, and converted into mathematical vector embeddings. These vectors are securely stored in our PostgreSQL + pgvector database instances. Database access is highly restricted and encrypted at rest.

3. External AI Processing

To execute source-grounded chat and completions, context chunks are sent to your chosen AI model provider (such as Langdock, Blackbox AI, or OpenRouter). These transactions are executed via encrypted TLS connections and adhere to standard API data privacy policies (guaranteeing that model builders do not persist or train on API payloads).

4. Security Baselines

For our cloud web app, user sessions are protected using Supabase Auth JWT credentials, enforcing granular row-level security policies on PostgreSQL tables, ensuring that no user can ever access or query workspaces belonging to another builder.