Your AI Identity stays in this browser

Browser-local AI Identity

Your profession, industry, selected goals and tasks, work-style choices, model and tool environment, subscriptions, platforms, languages, budget, and privacy requirement are stored in browser storage on the device where you save them. OpenGPT does not upload this profile, inspect the named services, or silently sync it to an account.

Bounded decision memory

Memory keeps only limited references needed to revisit a decision or strategy: canonical IDs, candidate model IDs, a task slug, decision state, strategy path, outcome, timeline, and timestamps. It does not store prompts, pasted answers, review notes, files, URLs, arbitrary payloads, or a free-form memory body. Lists are capped and older references are removed when the limit is reached.

Migration, import, export, and reset

A legacy AI Profile v1 is shown for review and is migrated only after you explicitly confirm. A current-profile JSON file is size-limited and fully parsed and validated before a confirmation can replace the saved profile; invalid input cannot partially overwrite it. Export creates a local file only when requested. Reset removes AI Identity keys from this browser but does not delete separately exported files or the model-decision workspace.

Separate account and workspace data

AI Identity is not included in the existing portable model-workspace backup, and that backup does not import or overwrite AI Identity. Signing in for submissions does not grant access to AI Identity, ChatGPT conversations, files, memory, API keys, or billing. Account-linked submission identity remains governed by the separate export and anonymization controls below.

Account identity and submissions

When you use an account-linked feature, OpenGPT can receive a site-specific user ID plus the name and email supplied by the authentication platform. OpenGPT cannot use that sign-in to read your chats or other private account content.

A valid account email is required for a model or Agent submission. OpenGPT uses it only to send the submission receipt and the final review result; it is not used for marketing.

Public-data submissions are tied to the submitter so reviewers can audit abuse and evidence. Submitter identity is retained for at most 365 days. The account workspace provides export and anonymization controls; anonymization closes pending submissions while public evidence may remain without identity. Anonymizing or deleting submission identity also cancels unsent notification jobs and removes their queued recipient address.

Account-linked submissions →

Analytics and network boundary

Public pages may load limited product analytics, but profile fields, work-style choices, environment IDs, memory references, prompts, answers, reports, email addresses, file contents, credentials, and API keys are prohibited event data. AI Identity and Builder adapters are deterministic local code and do not call an AI service.

• Do not enter confidential project details into the profile.
• Clearing site data or changing browsers can remove or hide local information.
• Where consent is required, analytics loads only after you allow it.
• Links to providers and downloaded files follow the destination or device rules you choose.

Questions and requests

For a privacy question or a request concerning account-linked submission identity, contact contact@opengpt.com. Use the in-product profile export and reset controls for browser-local AI Identity.

Your cloud account

If this Site enables free cloud accounts, full records are stored only after you select each record and explicitly consent. Existing local content is not migrated automatically.

Selected cloud content is retained until you actively delete it. Saving, uploading, downloading and deleting each require their own explicit action. Local content is not synchronized automatically.

The platform’s backup retention and deletion propagation times have not been verified. Removing online content does not prove that every backup or copy has been erased immediately.

Sign-in identifies the uploader, not the original author or historical confirmer. Hashes check integrity, not identity or model quality. Cloud records do not grant execution, acceptance or confirmation authority; no model is called.

A minimal deletion marker retains the account identifier and content hash to prevent a late upload from restoring this record. The marker is not a copy of its content. This exact record cannot be uploaded again in this version.

Close this cloud profile

Delete every active cloud record individually before closing the profile. No record is deleted automatically.

After deleting every active cloud record individually, I confirm removal of this profile’s display name and language preference. Minimal closure and deletion markers remain, and this version cannot reopen the profile. My ChatGPT account and local copies are unaffected; backup deletion is not verified.

Task packs and manual acceptance saved on this device

Save a snapshot only after explicit confirmation. Unsaved drafts stay temporary; earlier saved snapshots remain in history. Nothing is uploaded and no model runs.

Saved content uses localStorage for this site, browser and profile only. It is not cloud storage or on-device model inference. Clearing browser data, storage limits or device loss can remove it. Export a backup explicitly when needed.

Read-only history — not a new acceptance confirmation

Hashes check file integrity, not truthful claims, provider or model identity, execution or result quality. Imported confirmation times and confirmed-by fields are unauthenticated claims. Import transfers no approval, identity or authority.

The existing workspace backup does not include these new full-content records. Use the separate task-pack history export to back them up.

Manual retrial summary

Review recorded differences between the previous and new acceptance results, and mark what is not directly comparable. These are content differences, not an automatic verdict on better or worse, truth or model quality. Earlier history stays unchanged. Nothing is uploaded and no model runs.

I explicitly agree to save or download the full summary, including selected task content, pass conditions, pasted outputs, reasons, evidence and my notes. Keep downloaded files private.

Saving uses this site's localStorage in this browser and profile only. It is not cloud sync, encryption or guaranteed durable storage. Browser-data clearing, storage limits or device loss can remove saved summaries.

Metadata omits task and result text and your notes, but includes identifiers, hashes and counts. It is not anonymous or encrypted.

Summaries are saved separately from task-pack history and are not included in its backup. Use the dedicated summary importer to restore a full summary JSON export as read-only history; the task-pack history restore does not support this format.

Choose a full summary JSON export. Metadata, Markdown and task-pack history backups cannot be restored here. Validation and preview do not save anything. Nothing is uploaded and no model runs.

Restoration preserves the earlier records and items that are not directly comparable. It adds read-only history, not an editable task or execution permission. Local storage is not cloud sync, encryption or guaranteed durable storage.

The imported confirmation, name and time remain historical claims, not your new acceptance or authenticated identity. Hashes check integrity, not truth, authorship, rights or result quality. Restoration does not judge which result is better.