HometraceAI data handling
What Hometrace sends to AI providers—and why
Hometrace AI runs when an inspector starts photo analysis, submits a voice recording, or messages the Hometrace Agent. Photo Analysis AI uses OpenAI and Anthropic for different analysis tasks. Voice AI sends audio to OpenAI for transcription, then sends the transcript and report-structure context to Anthropic for parsing. The Hometrace Agent routes requests through the Vercel AI Gateway to an open-weight model hosted by Baseten, a named U.S. inference provider. The resulting suggestions remain subject to inspector review, correction, acceptance, or rejection.
Who it is for
For inspectors who need a concrete AI data-flow answer
The relevant questions are what starts processing, which inspection content is included, which company handles each step, what Hometrace retains for the workflow, and who decides whether model output reaches the report.
Processing is inspector-triggered
Photo analysis starts when the inspector runs it. Voice processing starts when the inspector finishes a recording and submits it. Hometrace does not describe either feature as automatically analyzing every inspection.
Requests use task-specific context
The designed AI request includes the in-scope photos or recording plus relevant report labels, fields, options, notes, and template language. Hometrace policy says names, account credentials, property addresses, and client contact details are not added as separate AI request fields.
Named providers have defined roles
OpenAI provides voice transcription and portions of photo analysis. Anthropic parses voice transcripts and handles portions of photo analysis. Baseten hosts the open-weight model behind the Hometrace Agent. The provider used for a particular photo-analysis task can vary.
Open-weight models stay on named U.S. hosts
When Hometrace uses an open-weight model, requests route through the Vercel AI Gateway pinned to a single named U.S. inference provider with automatic provider routing disabled. Inspection data is never sent to the model’s author or to an unnamed inference host.
How it works
The current photo and voice data flows
Hometrace sends only the content needed for the selected task, stores workflow results for review, and leaves the report decision with the inspector.
The inspector starts the feature
A photo-analysis command or submitted voice recording establishes the scope. Avoid putting sensitive information into media or notes when it is not needed for the inspection task.
Hometrace builds the request
For photos, the request combines image content with the relevant subsection, specs, finding options, notes, and template context. For voice, the audio is transcribed before the transcript is paired with report structure and template context.
A named provider returns structured output
One or both named providers process the scoped request. Hometrace validates returned data and maps it into a transcript when applicable plus structured suggestion records rather than treating free-form model text as a finished report.
The inspector reviews the result
Hometrace stores the transcript or pending suggestions needed for the review workflow. The inspector can accept, modify, or reject suggestions and remains responsible for the final report.
What to know
Policy commitments are not a provider-wide guarantee
The current Hometrace Privacy Policy states that AI sub-processors use submitted content only to fulfill the request, do not retain it after processing, and are not authorized by Hometrace to train on it. Those are Hometrace policy commitments for this service—not a claim of universal zero data retention or no-training behavior across every provider product.
- For Hometrace Agent requests, Hometrace enables the Vercel AI Gateway’s zero-data-retention and no-prompt-training controls and disables automatic provider routing. This page does not claim a regional data-processing guarantee or a fixed retention configuration for other provider products.
- Hometrace stores inspection content and AI workflow results under its own Privacy Policy; that policy does not promise one fixed retention period for every data type.
- Photos, recordings, notes, and template language can themselves contain sensitive details, so inspectors should avoid including information the AI task does not need.
- AI output can be incomplete, incorrect, or misrouted and must be checked against the property and the inspector’s professional judgment.
- Hometrace may change AI sub-processors over time and commits to updating the published policy when it does.
Product evidence
Verify the current policy and product behavior
The legal policy names the providers and commitments. The feature guides document the actual inputs, processing steps, review controls, and limitations.