Effective date: September 19, 2026. Contact: beta@chie.house.
1. Purpose of AI Processing
Chie uses automated processing to help users capture, clean, organize, search, and understand purchase records. The purpose is purchase intelligence for the user's benefit: reading receipts, extracting order details, grouping similar products, detecting duplicates, supporting search, producing household analytics, and helping users compare spend or savings opportunities.
Chie is not intended to use AI to make legal, credit, employment, housing, insurance, medical, or similarly significant decisions about users.
2. Systems Chie May Use
Depending on configuration and feature availability, Chie may use:
- Local OCR or parsing tools to read receipt images or text on Chie-controlled infrastructure;
- External OCR or document-intelligence providers to extract text, totals, dates, merchants, and line items from receipts or order documents;
- Vision models to read original receipt images directly after acceptance of the receipt-processing terms;
- Language models to normalize products, learn order-email templates, classify purchase-like content, or produce bounded purchase summaries;
- Embedding models to support semantic search and retrieval over purchase records;
- Deterministic code for rule-based analytics, totals, exports, limits, deduplication checks, and account security.
3. Data That May Be Processed
AI and automated systems may process receipt images, extracted receipt text, merchant names, item names, quantities, prices, dates, totals, taxes, fees, order numbers, shipping or tracking cues, product categories, redacted snippets, search terms, and operational metadata such as provider, model, operation type, status, timing, and estimated usage cost.
Some submitted receipts or order emails may include incidental personal information such as a name, email address, phone number, partial address, loyalty number, payment label, or last digits of a card. For original receipt-image reading, the full submitted image is sent as described below. Other operations use limited fields or redacted snippets where practical.
4. Data Minimization Rules
Chie's AI boundary is designed around minimization:
- Purchase consolidation should send item/order fields and ephemeral run identifiers, not account names, email addresses, phone numbers, postal addresses, stable account IDs, or freeform personal profiles.
- Forwarded-email template learning should use extracted order-like metadata and redacted snippets, not raw full emails, sender addresses, personal names, tracking URLs, payment hints, or stable source identifiers when those details are not needed.
- Search and embeddings should be account-scoped and should avoid exposing another user's private records.
- Admin aggregate keyword reporting should use minimum cohort thresholds so a keyword or category is not shown when too few accounts contributed to it.
- Provider usage metadata should record what operation ran without exposing unnecessary purchase contents.
Original receipt images — your ongoing choice
By accepting these receipt-processing terms, you authorize Chie Pi to send your original receipt images to OpenAI to read, organize, and analyze your purchases. This includes images submitted through uploads and connected receipt channels, including Atout Privé. Images are not locally redacted before submission: names, addresses, loyalty numbers, payment details, barcodes, and other information visible on the receipt may be included.
Your acceptance is recorded once and covers future receipts under this notice until you turn receipt processing off in Chie Pi Settings. You do not need to approve each receipt. If we materially change this processing scope, we will request acceptance of the updated terms before using that scope.
Turning processing off blocks new original-image submissions to the provider, including queued receipts not yet sent. Requests already sent may finish and cannot be recalled. Turning it off does not delete saved receipts or cancel billing. You can use the separate deletion controls or contact us about deleting your data.
We send receipt images for extraction with response storage disabled. This does not guarantee zero provider retention. OpenAI may retain abuse-monitoring data for up to 30 days by default, with exceptions for legal or safety reasons. Its API data is not used for model training by default unless sharing is explicitly enabled. See OpenAI's data controls for its handling and exceptions.
The extracted record can contain mistakes. We check the output for structural and arithmetic problems and keep uncertain receipts available for review. Keep your originals for important financial records.
5. Processing controls
Receipt images may be sent to an external vision or document-processing provider only when the configured operation allows external processing and the feature needs image analysis. Chie may record the provider, model, operation, policy version, timing, and usage metrics for accountability, cost control, debugging, and security.
The direct original-image reader does not require local OCR or redaction. If consent is missing or withdrawn, it does not send the image. If extraction is unavailable or uncertain, the receipt may remain pending or require review.
6. Provider Handling
Chie may be configured with providers such as OCR services, document-intelligence services, language-model providers, embedding providers, cloud infrastructure, or local models. Provider availability and retention settings depend on the configured deployment and provider contracts.
Where provider settings allow it, Chie is to choose options that limit retention, training, and secondary use of submitted data. We do not authorize providers to use user purchase contents for advertising profiles. Provider-specific legal terms and technical controls may still apply.
7. Model Output Limits
AI outputs can be wrong. OCR may misread a total, line item, date, or merchant. Language models may infer the wrong category, merge products incorrectly, or produce an incomplete explanation. Chie uses deterministic checks where practical, but users should review important records before relying on them.
AI output is not professional advice and should not be used as the sole basis for legal, tax, accounting, financial, insurance, medical, or similarly important decisions.
8. Human and Admin Review
Chie is designed so administrators can support accounts through masked contacts, opaque account identifiers, usage counts, limits, aggregate provider usage, job status, and operational metadata. Admin screens are not intended to show user receipt, order, item, or purchase contents.
If a future support process requires reviewing user-submitted content, it should happen only with the user's explicit direction or another lawful and documented reason, and should be limited to the content necessary to resolve the issue.
9. Logs, Audit Metadata, and Retention
Chie may store provider operation metadata, confidence values, extraction status, timing, usage counts, cost estimates, and structural audit events. These records help debug processing failures, enforce limits, investigate abuse, and improve reliability. They should not contain more user purchase content than necessary for the operation.
When a receipt or order is permanently deleted, Chie is intended to remove derived search documents, embeddings, provenance, and content-bearing rows that could reconstruct that record, while preserving limited non-content audit metadata where needed for security or legal reasons.
10. Security Controls
AI processing is part of Chie's broader security model. Safeguards may include password hashing, optional passkeys, session and CSRF protections, rate limits, access controls, masked admin views, privacy-boundary prompt builders, redaction/tokenization utilities, account-scoped retrieval, and logging of provider operations.
Current local production storage and operational snapshots are protected through access restrictions, file permissions, service isolation, and limited operator access. Those controls are not the same as encryption at rest, and local snapshots are not off-site disaster-recovery backups. Customer-facing network transfers under Chie's control use HTTPS/TLS.
No AI or security system can guarantee perfect protection. Chie's commitment is to reduce exposure, avoid unnecessary data movement, and treat purchase records as sensitive user data.
11. User Choices
You can choose what records to submit, whether supported records are private or shared, and whether to delete records where the app supports deletion. You may request account or privacy support by contacting beta@chie.house.
Some AI processing is necessary for Chie's core receipt and order features. If you do not want a record processed, do not submit that record.
12. Changes to This Notice
We may update this notice as Chie's AI providers, model settings, processing flows, or privacy controls change. Material changes may require renewed acceptance or additional notice.