Live View & ingest data

What we store, what we don't, and what you control

What gets stored

Only what your application sends through the ingest API or SDK (for example request_data, response_data, optional tool_events). If a field was never sent, it will not appear in the UI — an empty area means “not collected”, not a server bug.

What we don't assume

We do not infer end-user prompts, tool outputs, or RAG context unless your app included them in the payload you chose to send. Secret-like patterns may be masked on the server before persistence (see Security & Data Retention).

What you can turn off (SDK)

Python and Node SDKs support privacy flags and byte limits (for example omitting message bodies or truncating large JSON). When enabled, the product shows clear Privacy / Truncated hints in Live View. See your SDK README "Security / privacy" section for env names such as PLUVIANAI_LOG_USER_CONTENT.

Release Gate vs Live View

Release Gate surfaces tool evidence from the replay pipeline; Live View snapshot detail uses stored snapshots and the same tool timeline components where applicable. Provenance labels (recorded vs simulated) explain what was grounded from your traces.

Related

Developers: full field matrix and JSON examples live in the repository under docs/live-view-trust-data-collection.md and docs/live-view-ingest-field-matrix.md.