CollectorBucket
YOUR COLLECTION, YOUR SPACE

Privacy & your photos.

CollectorBucket is an early version of a collection notebook with optional discovery. Here’s how this version works.

Your items start private.

Personal collection records and photos are associated with your signed-in account. New uploads are private. After its photos pass screening or a moderator approves it, and you review the details, you can choose “Show this find in Discover.” Shared items expose only their photos, name, category, material, dimensions, condition, public story, city, and country. Purchase cost, storage location, private notes, account identifier, and email are never included in discovery. CollectorBucket is public: shared finds can be viewed without signing in, indexed by search engines and previewed when their link is shared. Turn sharing off to remove an item and its photos from discovery and its find page. People who already copied or downloaded a shared photo may retain their copy. Example finds contain invented objects and locations, clearly labeled as illustrative.

Places and browsing.

City and country are entered by you; image recognition does not infer a location from a photo. “My area” is an optional preference stored on your device for browsing and prefilling your own uploads. No GPS or exact address is requested. Choosing a different country or worldwide browsing is always available.

Saved finds.

Your shortlist stores only find identifiers on this device, without sign-in or a copy of anyone’s private records. It does not sync between devices. Shared finds are checked again when you open Saved; a find made private or held for review is omitted. Illustrative examples remain labeled. Clearing browser storage removes the shortlist.

Sign-in and storage.

CollectorBucket offers native passkey registration and sign-in without a password or email address. Your device or credential provider holds the private key; the app stores an opaque account identifier, optional public display name, public-key credential data and short-lived authentication challenges. Biometric data is not sent to CollectorBucket. Session cookies are secure and HttpOnly in production; only token hashes are stored. Sessions expire after seven idle days or 30 days total. Account settings let you add, rename and remove passkeys, revoke other sessions, download one-time recovery codes and delete your account. Successful recovery replaces old passkeys and revokes all sessions. Store a backup independently; losing every credential and recovery code can make access impossible. Short-lived, hashed network identifiers help limit authentication abuse and expire within two hours. Collection details are stored in Cloudflare D1; uploaded images are in Cloudflare R2. Existing collectors can still connect their original ChatGPT account and attach a passkey without changing ownership. The hosting platform supplies their identifier, email and optional display name. That verified legacy email is retained for account continuity and administrator authorization; it is not public. New passkey-only accounts have no email address. Display names and matching email text never establish ownership.

Uploaded photos.

This version accepts JPG, PNG, and WebP. Your browser resizes photos to a maximum dimension of 1,400 pixels and saves a JPEG copy for upload. This removes embedded metadata such as camera location, but anything visible in the image remains visible. Keep your original full-resolution photos elsewhere.

Background cleanup.

AI cleanup is enabled for new photos by default. A self-hosted U²-Net-p model runs on your device to separate the object from its background. Preview the result before saving and turn cleanup off if it removes glass, fine edges or useful details. Cleanup does not generate a new object or retouch its marks and wear, but segmentation can remove parts by mistake. The resized original and any cleaned copy are both stored; use Photo & community preferences to restore originals. Existing uploads are not automatically changed. If cleanup cannot run, the original is kept. Both original and cleaned photos are screened before sharing. Removing a photo removes both copies.

Automatic screening and suggestions.

Every uploaded photo, including additional views, is submitted for automatic screening. When the same photo or text was checked for your account within the last 30 days, its cached screening summary may be reused. Otherwise it is sent to a Cloudflare Workers AI model to check for inappropriate content, relevance to collectible objects, and recognition confidence. The system suggests a name and category and may request an additional view. There is no switch to skip screening. Potentially inappropriate photos are blocked before storage. Uncertain, unrecognized, unrelated, or unavailable screening results keep a record private for moderator review. Checks can miss content or flag acceptable photos; they are not a guarantee of safety or authenticity.

Moderator access and comments.

The site administrator can review flagged private photos and public item details, approve or reject finds and comments, ask for better photos, and resolve reports. The review desk does not expose purchase costs, private storage locations, or private collection notes. A decision log records review actions and blocked attempts; images from blocked new uploads are not retained.

Comments require sign-in, automatic text screening, and moderator approval before others see them. Your chosen display name appears publicly, but your email and account identifier do not. Pending comments are visible to you and the administrator. You can remove your comment. Reports and daily limits help reduce abuse. Public text is screened before a find is shared. Comments disappear from public view if a find is unshared or held for review.

Cloudflare’s handling of inference data is described in its Workers AI data-usage documentation.

Votes and private contact.

Collectors can disable comments, votes, or private contact requests for each find. Voting requires sign-in; one vote is allowed per account per find, and collectors cannot vote for their own finds. Only approved, shared finds accept new activity.

Contact requests and replies stay in the platform inbox. Your chosen display name is visible to the other participant; your email and account identifier are not. A collector must accept a new contact request before replies can be sent. Messages are screened, links and personal contact details are refused, and uncertain messages wait for administrator review. Participants can block and report messages. Administrators can see flagged or reported message text, and review actions are logged. Limits allow three new contact requests and 30 messages per account per day; a short-term limit also applies.

Email copies are optional and off by default. When a verified sending service is connected and you opt in, approved messages can be sent to an existing verified account email, up to ten copies a day. You can opt out in Inbox or through an email unsubscribe link. Bounces and spam complaints turn copies off. Email delivery and inbox placement depend on the recipient provider and are not guaranteed. Message threads remain after a find is removed so participants can retain their conversation; the removed find no longer accepts new contacts.

Account removal.

Account settings include permanent account deletion after recent passkey verification and a typed confirmation. This removes the collection, stored photos, credentials, recovery codes, preferences and private conversations, including those conversations for the other participant. Minimal moderation records and a deleted-account identifier may remain to prevent old logins restoring deleted ownership. Previously downloaded copies cannot be recalled. Export important records and images first.

Export and removal.

Export records to download your item details as JSON, or choose CSV for a spreadsheet. CSV leaves out private costs, storage locations, buckets and notes unless you explicitly include them. Open a find’s saved record to print or save a PDF with its photos; private details are off by default. JSON includes private record details. Images are not embedded in JSON or CSV; download them individually from each item. Unsaved photo previews, editor drafts and inbox reply drafts stay in this tab’s memory. We warn before leaving with unfinished work; browser crashes can still lose it. Open My collection and choose a find. “Remove photo” deletes the selected image when the find has more than one photo; the remaining find becomes private until you choose to share it again. “Remove find” deletes the stored record and all of its uploaded images. A confirmation appears before deletion. To delete the last photo, remove the find. If storage removal is interrupted, “Retry photo removal” completes it from My collection. Derived screening summaries and content fingerprints in the screening cache expire after 30 days; cached summaries do not contain the uploaded image or submitted text itself. Review decision logs may remain. You can remove all your items this way. You can also permanently delete your account from Account settings.

Early-version limits.

Up to eight photos can be added at a time, with up to 500 saved photos per account. If connected, photo screening is limited to 30 submissions per account per day. The site also has a shared daily automatic-checking allowance of 100 units: a new photo check uses four, and a new text check uses one. Cached summaries use no units. The administrator can pause new automatic checks. This is a usage guardrail, not a promise of unlimited free service; hosting, storage, and provider allowances still apply. Rechecking an existing find is limited to three attempts per item; additional photo submissions use the daily allowance. New comments and reports are each limited to 20 per account per day. If screening is unavailable or its limit is reached, photos may be saved privately for moderator review and cannot be shared until cleared. Keep your own backups of important records and original images.

Back to your collection