Skip to content

Security and data handling

For each run: the URLs discovered, the sampled pages’ HTML as fetched by each identity, the extracted text and passages with their embeddings, the entities and claims, the simulation results, findings and the analysis. Page bodies are kept for 30 days for re-analysis and then removed; derived data is kept for the life of the workspace.

Every record carries the workspace it belongs to, and every request is scoped to the caller’s workspace. Behind that, the database enforces row-level security: the application connects with a role that can only see rows of the workspace set for the current request, so a query that ever missed its filter returns nothing rather than another workspace’s data. Two workspaces cannot see each other’s data. API keys are stored as hashes and shown once.

Analysis runs on Anthropic’s API under a no-training agreement. Enterprise workspaces can route through their own account instead — Anthropic, Amazon Bedrock or Google Vertex AI — so prompts and responses stay inside their contract. Only page content from the customer’s verified domains is sent to a model.

Signed requests and a published key — the signature is the identity, not an address list; crawling only on verified domains; robots.txt honoured. See The crawler.

SOC 2 attestation and regional data residency are on the roadmap and are stated as such in contracts until they ship.