FAQ

For most sites, yes: we set no cookies, mint no persistent identifier and store no personal data. The final call belongs to your counsel; we document the technical facts.

Not unless you enable one. No model is on by default. If you turn one on you have two options: Ollama on your own server (nothing leaves the box) or your own API key. In both cases what reaches the model is an aggregated evidence bundle, not raw visitor data.

For most sites, yes. Not for ad-platform attribution (Google/Meta Ads ROAS) — we deliberately do not build that.

Yes, since September 2026 — and it is off by default. For a long time we said we would never build it: recording a visitor's screen means recording the ID number they typed into your form, and if masking misses one field you created your customer's compliance problem. So it ships the other way round from most tools: every text node and every input value is masked unless you unmask an element on purpose, password and card fields are never captured at all, Do Not Track is always honoured, and a site has to opt in twice. The aggregate signals — field-level form abandonment, rage clicks, scroll depth — still answer most of the same questions without recording anyone. See session replay.

You are not cut off immediately: data keeps flowing inside a 20% grace band with a dashboard warning. Once that is exhausted new events are rejected — and the rejected count is shown, so you can see what you lost.

Self-hosted, the database file is yours — query it directly. In the cloud, /api/stats returns the evidence bundle as JSON and the MCP server exposes the same data to an agent. Bulk CSV export does not exist yet.

Yes, over MCP. It is read-only, and every answer includes the query that produced it so the agent can cite evidence rather than assert a number.