You choose the evidence. NeedPace minimises what it keeps.
No whole-mailbox access, explicit retention for raw documents and human control over every prediction.
A proactive product should reduce what you remember without creating a permanent archive of everything you do. NeedPace separates temporary raw evidence, confirmed structured state, predictions and audit metadata so each category can have the right access and retention rule.
Manual quick add and selected receipt or PDF upload work without connecting Gmail or Outlook. A private forwarding address can process only messages you actively choose to send when the inbound provider is enabled. It is not a licence to read the rest of your correspondence.
No Gmail OAuth in the MVP
No Outlook OAuth in the MVP
No retailer or bank integration required
02
Raw source evidence remains personal
An uploaded receipt, forwarded message body, attachment and raw extracted text belong to the importing user. Household membership, including owner or administrator status, does not automatically grant access. A confirmed derived product or Need may be shared at household scope while the underlying raw source stays personal.
Importer-only raw source
Household-scoped derived state only when selected
Every write preserves actor and audit context
03
Retention is explicit
Raw email bodies, attachments, receipts and documents have a configured maximum retention of 30 days. Structured provenance can be retained longer to explain corrections and predictions without keeping the full source. Deletion workflows use tombstones so an expired object is not silently re-exposed after a backup restore.
04
Secure processing with honest boundaries
Data is encrypted during transmission and stored in private, encrypted object storage. Files are validated and scanned by ClamAV before OCR. Selected documents may be sent to configured external AI providers through stateless endpoints under no-training, no-hosted-file and zero-data-retention controls. This is not end-to-end encryption, and we do not describe it as such.
05
AI proposes; structured application rules decide
AI extraction must match a strict schema and is validated again by the application before candidate data can be confirmed. Important state is stored as database records rather than free-form model text. Low-confidence output goes to user confirmation, and uncertain product aliases are never silently merged.
06
You stay in control
Every prediction can be corrected, dismissed, snoozed or overridden. You can change quantities, correct products, mark a Need as no longer relevant and request account or household deletion. NeedPace does not use customer data to train proprietary models and has no implemented consent path that would permit such training in this MVP.
Start with the smallest useful signal
Manual quick add creates a structured Need without uploading a document.