The Chocolate Guide in ChatGPT

Information the Guide receives

OpenAI sends the structured tool inputs needed to answer a request. These may include product search terms, budget, cacao percentage, maker, cacao origin, flavor or dietary preference, gift occasion or recipient category, stock preference, a public product or variant ID, and tasting focus.

OpenAI protocol metadata may accompany a tool request. Depending on the ChatGPT client and account context, it can include locale, user agent, coarse location such as country, region, city, or approximate coordinates, anonymized organization metadata, and pseudonymous technical identifiers for a user or conversation. This metadata transits the hosting service as part of request processing even when the Guide ignores it. The Guide does not request or use these categories for its features, and its application code does not persist them.

The Guide does not require a Bar & Cocoa account and does not request or use Bar & Cocoa customer accounts, carts, orders, payment details, saved addresses, or ChatGPT account information.

How information is used

Inputs are used to validate the request, search and compare public catalog facts, provide reviewed gift and tasting guidance, and check selected public products for current price and availability.

Search terms, dietary preferences, gift details, and full prompts are not sent to Shopify. For selected results, the service calls only public Bar & Cocoa product and cart endpoints using a public product handle or variant identifier.

Please do not send sensitive information

Do not include names, contact details, payment information, credentials, precise location, medical information, or other sensitive data. Free-text dietary information can reveal sensitive information. The Guide is not medical advice and cannot determine allergy safety.

Storage, logs, and retention

• The application does not store request bodies, search terms, conversation history, or user profiles in a filesystem or database.
• Persistent storage contains a public catalog cache. Short-lived caches contain only public product and commerce data keyed by public handle or variant ID.
• For completed Guide tool calls, the application stores daily aggregate counts by one of eight fixed tool names, success or error outcome, and four coarse latency ranges. The schema-versioned aggregate file on the existing application volume retains at most 90 UTC days. Recent daily snapshots are also written to application logs for operational recovery.
• A private Bar & Cocoa operations dashboard in Coda receives copies of the daily aggregate snapshots. It refreshes daily and automatically removes active dashboard rows older than the rolling 90-UTC-day window. Coda may temporarily retain deleted content in its normal revision history, trash, backup, or recovery systems under its own policies; removal from the active dashboard does not promise immediate erasure from provider recovery systems.
• The aggregate counters do not contain prompts, arguments, request bodies, results, exception text, product selections, IP addresses, cookies, user or session identifiers, request IDs, or profiles. Health checks, tool discovery, and website requests are not counted.
• Usage counters include tests and reviewer activity. They are not counts of unique people, orders, or sales. While the service runs, a privacy-safe hourly heartbeat advances the covered-through time without changing tool counts. A continuously observed idle UTC day can therefore report zero calls. The first day, restarts, and detected coverage gaps are marked partial and are never backfilled as observed zero activity. A missing day means monitoring coverage was not recorded. Replaying an earlier daily snapshot for recovery does not advance its displayed covered-through time.
• Application logs also record startup and catalog refresh success or failure. The application does not log MCP request bodies or tool arguments.
• Railway, the hosting provider, automatically records operational HTTP metadata such as request ID, timestamp, method, path, host, status, user agent, source IP, edge region, byte counts, and timing. Railway also provides DNS logs. This metadata supports operating, troubleshooting, and protecting the hosted service.
• The hosting workspace is on Railway Pro. Railway documents a 30-day default log retention period for Pro workspaces. Bar & Cocoa has not independently verified a service-specific retention override, so this is the documented platform default rather than a guarantee about a particular deletion event.

Your choices

You can stop future Guide tool requests by not invoking the Guide or by disconnecting it in ChatGPT. For applicable requests to access, correct, or delete information, email info@barandcocoa.com with the subject “Privacy Request.” Because the application does not create a user profile or store request bodies and search terms, Bar & Cocoa may not be able to locate or associate anonymous application activity with you. Requests remain subject to applicable law and available records.

Providers and other sites

Railway hosts the Guide, Coda hosts the private Bar & Cocoa operations dashboard, and public product data comes from Bar & Cocoa's Shopify storefront. The application code does not use request data for behavioral advertising and does not set cookies, load advertising trackers, or use third-party analytics SDKs. Its only custom usage measurement is the privacy-light daily aggregate described above.

If you follow a product link, you leave the Guide and visit barandcocoa.com, where the store's privacy policy applies. OpenAI separately processes your ChatGPT conversation under its own terms and privacy notice.

Contact

For questions or a privacy request, email info@barandcocoa.com with the subject “Privacy Request.”