Last updated October 22, 2026
Privacy & data use
This is an independent conversation archive and product research project. It is not an OpenAI service and is not affiliated with or endorsed by OpenAI or Anthropic.
What you choose to share
The current early-access form asks only for your email. By selecting Notify me, you agree to contact about the pilot and product research. Earlier versions also offered optional workflow questions and a separate feedback form; existing responses retain their original consent choices. We also store the submission time, consent choices, a record identifier, the concept version you responded to, and a hashed withdrawal code.
Please do not send API keys, source code, secrets or private conversations through the email or research forms. The archive tool reads only files you explicitly select; it does not connect to your AI accounts.
Local conversation archives
The /archive tool reads the files you choose entirely in your browser. Conversation contents are not uploaded, sent to an AI provider or included in website analytics. Original UTF-8 source files and their SHA-256 integrity hashes are saved in this browser’s IndexedDB when available. Browser storage is unencrypted and can be cleared or lost; it is not a backup for a broken computer. On shared devices, other people with access to the same browser profile may be able to read it.
Download a backup and keep it on another device or in a private cloud folder. Optional password-protected JSON backups use AES-256-GCM with PBKDF2-SHA256 and 310,000 iterations; passwords are not stored and cannot be reset. Raw JSON, Markdown, offline HTML and local-helper snapshots are unencrypted. The Remove local copy action removes the archive stored by this tool, not downloaded files or original exports. Unsupported metadata and attachment references remain in the source files; external attachments are not copied.
The optional Codex backup helper operates locally without network requests. It reads JSONL text sessions from sessions and archived_sessions, excluding credentials and project files. Users choose the destination and configure their own cloud-folder sync; This public archive tool does not provide hosted conversation storage.
Invitation-only CodexAnywhere Vault pilot
A separate macOS pilot can capture ordinary and compressed Codex session files from a chosen local source. It excludes authentication files, project files and external attachments. Conversation files, filenames and readable content are encrypted on the Mac before upload. Cloudflare R2 stores ciphertext; a separate D1 database stores account and device identifiers, ciphertext hashes and sizes, quota use and version times. The server does not receive the recovery code or encryption key.
The recovery code grants independent account ownership and decrypts backups without the original AI account. The Mac can save it in its system Keychain. Keep a separate private copy: we cannot reset it by email or restore a lost code. Revoking a device blocks its device token, but anyone holding the recovery code can reconnect. Offline HTML, Markdown and restored original files are readable copies; protect the destination folder.
Automatic capture runs every five minutes while the app is running and the Mac is awake. Offline versions stay queued. The app reports local capture, confirmed cloud storage and verified recovery separately. Pilot limits stop new uploads without deleting earlier versions. There is no automatic expiry or self-service cloud deletion in this pilot; uninstalling the app keeps backups and its recovery code. Private conversation content is not sent to website analytics or session recording. Infrastructure providers may process operational metadata under their own policies.
This pilot requires an invitation and is separate from the public browser archive. Public automatic-backup distribution remains pending Apple signing and notarization and human pilot acceptance. It does not restore an AI account or import conversations into the original AI application.
How it is used
The project owner can read submissions to understand demand and problems with Codex. If you supply an email and agree to contact, the owner may contact you about this research or a possible pilot. Your response is not posted publicly. We do not use it for unrelated marketing.
Where it is stored
Responses are stored in this project’s dedicated Cloudflare D1 database through a private server connection. Vercel hosts this page. The project owner can access responses through the authenticated Cloudflare D1 console. The separate website inbox requires configured GitHub administrator sign-in and a verified owner identity before showing feedback. Email sign-up is available when server storage is configured; it does not require administrator sign-in. No response is saved by an unavailable form. Hosting and authentication providers process requests as needed to operate the site and may retain operational logs under their own policies.
Simple interest counts
Activating the Notify me button attempts to record an aggregate count for that UTC day, button location, and concept version. Repeated activations can count again, including when the email form is incomplete. The count contains no email address, conversation content or withdrawal code. Opening the local archive does not record an interest click. These counts are clicks, not unique visitors or successful sign-ups. Your email is submitted separately when the form is valid; completed interest forms and other feedback are counted separately. The current independent-archive and hosted-backup interest campaign is kept separate from earlier cross-device, continuous-execution and offline-history concepts. No advertising trackers are added to this page.
Website analytics
Public pages use Vercel Web Analytics to measure page views, referrers, browser and device categories, and approximate country-level traffic. Vercel describes this service as cookie-free. We remove URL query parameters and fragments before sending page-view events, and exclude archive, administrator, review, withdrawal and API routes. We do not send form contents, email addresses or Codex conversations as analytics events. See Vercel’s Web Analytics privacy information for its processing and visitor identification methods.
Microsoft Clarity on public pages
Public introduction and guide pages use Microsoft Clarity to record clicks, scrolling and page sessions, together with device and browser information, to improve usability. Clarity runs without analytics cookies by default. Public page text and email forms are masked. Archive, administrator, review, withdrawal and API routes, URLs with query parameters, and unrecognized URL fragments are not recorded. Recording stops before an email submission so its private removal link is not captured. No Codex account, private conversation, response inbox or source code is connected to Clarity.
The Analytics preferences link lets you turn recording off or explicitly allow analytics cookies to connect page visits. Advertising storage stays denied. Your choice is saved locally for 180 days; Do Not Track and Global Privacy Control disable Clarity. Turning recording off clears this site’s Clarity cookies and reloads the page. Without cookies, visits cannot reliably be linked across pages. See Microsoft’s privacy statement and Clarity’s consent information.
Withdraw your response
After a successful submission, you receive a private withdrawal link and code. Save it. Anyone with that link can remove that response, so keep it private. Use the withdrawal page to delete the saved response and its email. It does not remove hosting providers’ operational logs or messages already sent.
Responses have no automatic expiry at present. You can withdraw at any time. We do not send a submission receipt by email, and cannot recover a lost withdrawal code through this page. Anonymous feedback cannot be matched back to you without that code.