What data is sent to the AI?
This page describes information version 2026-09-28-v6 for consent to AI use. The currently selected provider is Perplexity Agent API (Model: perplexity/sonar).
Requirements and authorisation
AI use is optional and requires both administrative enablement of your account and a separate active, current consent. Enablement and consent are separate requirements. One consent covers the initial analysis and later follow-up questions for the provider and information version described here; it does not replace checking whether you or your organisation are authorised to transmit the included data.
Initial analysis
The application sends a JSON payload to the Perplexity Agent API endpoint with model perplexity/sonar, an instructions field and the structured diagnostic context as input. The context contains collected diagnostic values including selected drive storage, RAM, Linux inode snapshots, aggregate CPU measurements and system uptime, findings, check details and ping measurements. When present, a redacted raw log of at most 100,000 characters is also included; input-budget compaction may omit it. The request also contains technical references for rendering the result and an internal job identifier. The application does not request tools, web search, URL fetching, a sandbox, MCP, model fallback or server-side conversation state.
Follow-up questions
A follow-up sends the locally stored AI analysis bounded for the dialog, a bounded diagnostic context without the raw log and without explicitly excluded secret fields, up to 20 stored completed dialog messages and the new input as an unconfirmed user observation. Text is redacted and trimmed again; automatic redaction may miss content. The follow-up is sent only after a renewed server-side check of current consent and administrative enablement.
Provider, purpose and third country
The provider is the Perplexity Agent API at https://api.perplexity.ai/v1/agent. Its purpose is technical interpretation of the selected diagnosis and follow-up support. Perplexity states that it uses computing infrastructure in North America; processing outside the EEA is therefore possible. The operator cannot fully determine or permanently guarantee the concrete processing location of an individual request, further technical processing steps or changes to the provider infrastructure.
Retention and training
Perplexity publishes statements about processing, retention, model training and infrastructure. The published statements about non-retention and non-use for model training for the Chat Completions API are not presented here as a promise for this Agent API use; whether they apply to this endpoint and model requires separate verification. Billing and technical metadata may still be collected. admin-tools does not control Perplexity’s internal processes, cannot fully verify them and cannot guarantee them permanently. Perplexity’s published API terms incorporate its DPA. The DPA contains provisions on Standard Contractual Clauses and mentions the possibility of a separate agreement on request. These general provider documents do not establish final project-specific legal clearance for this standard API project.
Metadata and storage locations
admin-tools controls payload selection and preparation, server-side enablement and consent checks, and local storage. The exact JSON request text without the Authorization header and the payload snapshot are stored in the configured local database. A fully valid Format 3 provider response is stored and displayed as a locally validated analysis. For a partially valid Format 3 response, the application stores and displays only the analysis filtered to individually evidenced findings, with an incompleteness notice. A fully invalid provider response may be stored as provider text for failure diagnostics, but it is not displayed as an analysis and is not sent in follow-up questions. Dialog messages are stored locally. These AI records follow the deletion rules of the associated diagnosis and are removed through the intended database cascades when that diagnosis is successfully deleted. According to its own information, Perplexity may process request, billing, security and usage metadata. The application does not control or verify the provider’s internal storage or infrastructure processes.
Withdrawal and risks
Revocation in the profile blocks new analyses, follow-ups and queued jobs not yet sent to Perplexity before the provider call. Analyses and dialogs already stored locally remain readable and follow the documented local retention and deletion rules. Redaction is not anonymisation: diagnostic values, free text and metadata may still concern you, other people, devices or internal systems and may contain confidential contexts. Automatic redaction may miss such content. AI answers may be wrong, incomplete or outdated and must be checked before action. The regular diagnosis and evaluation remain available without AI. Provider terms may change; therefore each consent is bound to a provider and immutable information version.