Encrypted everywhere
AES-256 at rest across Postgres and Storage. TLS 1.2+ for everything in transit. HSTS enforced site-wide with a two-year max-age.
Security
Built for confidential
strategic work.
MediaDatak handles broadcaster studies, internal strategy memos, and competitive intelligence. Below is what we actually do — written to be specific and verifiable, not vague reassurance.
At a glance
Four guarantees.
Tenant isolation in the database
Postgres Row Level Security on every tenant-scoped table — not just the application. A workspace can only ever see its own rows.
Per-workspace vector namespaces
Embeddings are isolated by namespace, per workspace. No cross-tenant similarity search is possible.
Real deletion
Delete a document and its rows, chunks, storage object, parsed cache, and vectors are all removed in the same operation.
The detail
What we store, and how it's protected.
What we store and where
Account data (users, workspaces, memberships), workspace content (uploaded documents, extracted text, embeddings metadata), and operational logs. All of it lives in Supabase (Postgres + Storage) in the EU (Frankfurt), except embeddings vectors, which live in Pinecone.
Encryption
At rest: AES-256 across Postgres and Storage, managed by Supabase. In transit: TLS 1.2+ everywhere, HSTS site-wide. Application-layer: the MediaDatak integration token is additionally encrypted with pgcrypto.
Tenant isolation
Every tenant-scoped table enforces Postgres Row Level Security. Policies check workspace membership on every query. Document chunks cannot be read directly via the database API at all — retrieval happens through server-side RPCs scoped to the caller.
No training on your data
OpenAI and Anthropic API contracts contractually exclude API content from model training. We don't enable any data-sharing or store flags. The full sub-processor list is public, below.
Sub-processors
A small, specific list.
Each is used for one purpose. We do not sell data, and none of these may use your content to train models.
- Supabase — Postgres, Auth, Storage (EU/Frankfurt)
- Pinecone — vector embeddings storage
- OpenAI — embeddings + inference
- Anthropic — LLM inference
- Netlify — hosting + serverless functions
- Resend — transactional email
Security & privacy FAQ
Straight answers.
Are my uploaded files secure?
Yes. Files are stored in private Supabase Storage buckets in EU/Frankfurt, encrypted at rest with AES-256, and accessible only via short-lived signed URLs. Access is gated by Row Level Security on the database side — even a developer mistake in application code cannot leak files between workspaces.
Is my data encrypted at rest and in transit?
Yes. Both Postgres and Storage are encrypted at rest with AES-256, managed by Supabase. TLS 1.2+ is enforced everywhere — browser to MediaDatak, MediaDatak to Supabase, MediaDatak to AI providers. HSTS with a two-year max-age is enabled site-wide.
Can LLMs train on my content?
No. MediaDatak uses the OpenAI and Anthropic APIs, both of which contractually exclude API content from model training. We do not enable any data-sharing or store flags.
Is my data shared with third parties?
Only with the sub-processors listed above, each only for the specific purpose described (hosting, parsing, embeddings, AI inference, email). MediaDatak ships no third-party analytics or trackers on the application.
Can other users access my documents?
No. Workspace membership is the access boundary. A user with no membership in your workspace cannot read your documents, conversations, or embeddings — enforced at the database via Row Level Security. Document chunks cannot be read directly via the database API; retrieval happens through server-side RPCs scoped to the caller.
What happens when I delete a file?
In the same operation we remove: the Postgres row, every chunk row (cascade), the raw file in Storage, the parsed markdown cache, and every vector for that document. In-flight parsing jobs are cancelled. The deletion is written to the activity log.
Are embeddings and extracted text also protected?
Yes — they are treated as the same sensitivity tier as the source file. Extracted text lives in a private Storage bucket with the same RLS rules. Embeddings live in a per-workspace namespace; queries are always namespace-scoped, so no cross-tenant similarity match is possible.
Does the Simulate engine use my audience data?
No. Simulate never ingests microdata. It works exclusively from aggregated statistics — census distributions, survey marginals, market-level behavioral data. No individual record ever enters the pipeline. GDPR and CCPA compliant by architecture, not by policy.
Is the platform suitable for confidential business documents?
Yes — that is the design center. MediaDatak is built for broadcaster studies, audience research, strategy memos, and competitive intelligence, where a confidentiality incident is existential. Customers with stricter requirements (e.g. zero third-party processors for parsing) should ask about enterprise options.
Questions about security?
Email security@mediadatak.com — or see how the platform works on a live walkthrough.