Privacy policy

1. About RajOS

RajOS is Raj Grover’s private personal knowledge and task system. F.R.I.D.A.Y. is its conversational assistant. This policy describes the public information website and the private application’s Google data integration. The assistant is for the owner’s personal use; this website does not offer registration or access to the assistant.

2. Google data accessed and changed

With the account owner’s authorization, the knowledge reader searches the configured RajOS folder in Google Drive and its subfolders. It reads file identifiers, names, paths, types, modification times and available sizes. It retrieves Google Docs as text and supported Markdown and text files, subject to a 500 KB per-file limit. An optional manually initiated import can retrieve a document selected by link. Authorization also reads the account’s display name and email address.

The reader uses Google Drive read-only permission (drive.readonly). This Google permission is broader than the folder restriction enforced by the application.

A separate writer uses drive.file permission and an application allowlist of three personal records. It inserts or deletes document text only after the owner approves the proposed change with a passkey. A separate backup connection uses drive.file to store encrypted recovery backups in the owner’s Google Drive.

3. Purpose and local processing

Document content and metadata support personal search, contextual answers, projects, tasks and approved record updates. The integration is intended for personal information, separate from employer data.

The application stores local document copies, metadata, searchable excerpts and indexes. F.R.I.D.A.Y.’s primary chat model and search-embedding model run locally through Ollama. Local processing does not mean that every copy or use of the information stays on the device.

4. External processing and sharing

Google: Google provides authorized document access and stores encrypted recovery backups. Backups include knowledge data, chat-related state, logs, configuration and authentication material needed for recovery; restic encrypts this material before upload.

Anthropic / Claude Code: Connected Claude Code sessions can receive retrieved Google document excerpts through the knowledge tools. Tool-level retrieval excludes restricted partitions, but a coding session granted direct filesystem access may also read local knowledge files. Delegated Claude tasks receive their request text; Google-derived information included in a request can therefore be sent to Anthropic. The delegation workspace excludes the main knowledge and data directories.

Other owner-directed services: The owner may separately supply documents to NotebookLM or copy information into other AI tools. Those transfers are governed by the selected service and its account settings. The reviewed configuration did not establish active automatic Gemini delegation.

Notifications: Apple Web Push transports encrypted notifications using generic text rather than document content. Delivery still involves the push service and technical delivery metadata.

Security processing: The inspected configuration selects local scanning and also configures Cisco AI Defense credentials. Whether any cloud security path receives inspected content or related metadata has not been verified. This policy therefore does not guarantee exclusively local security processing; it also does not assert that a transfer to Cisco has occurred.

External providers apply their own applicable terms, privacy policies and account settings. This policy does not promise that external providers have zero retention or that all connected AI accounts disable training.

5. Retention and deletion

Current knowledge: Synchronization is scheduled every 15 minutes. When a document leaves the configured Drive crawl, its current cached copy is removed on the next successful sync and the active indexes are rebuilt. Synchronization failures can delay this. Separately retained baseline indexes are not automatically pruned.

Chat: The dashboard stores conversation metadata separately from the underlying transcripts. Saved-chat metadata is retained until deletion or an optional 7-, 30-, 90- or 365-day retention setting applies. Temporary-chat metadata is scheduled for expiry after 24 hours; cleanup runs periodically. Conversation events are retained for 90 days. These periods do not establish deletion of transcript content: OpenClaw stores transcripts separately, and deleting a chat leaves a compressed transcript archive. No verified automatic deletion deadline is stated for those archives or OpenClaw’s underlying session storage.

Other records: Chat attachments are eligible for deletion after 14 days unused. Imports require manual deletion. Logs can contain Drive names and paths; logs, change proposals and baseline indexes have no verified automatic pruning. Security audit storage has no verified deletion deadline. These records may remain until explicitly removed.

Encrypted backups: The daily backup job applies a retention policy selecting snapshots from the most recent 14 days, eight weeks and 12 months that contain backups, then prunes unreferenced backup data. The rules are combined, and snapshots may overlap. This is not a guaranteed 12-month erasure deadline: retention depends on available snapshots, snapshot grouping and successful maintenance. Deleted local information can remain in a retained backup.

Deletion requests: Contact the owner using the address below to request a review and removal of retained copies. Removing a source file, deleting a chat or revoking Google access does not automatically erase all archives, logs, backups or information already supplied to another provider. Complete removal requires addressing those copies separately. No instant or universal erasure guarantee is made.

6. Revoking Google access

The account owner can review and remove application connections through Google Account third-party connections. Review the reader, writer and backup authorizations as applicable. Revocation stops future access through the revoked authorization; it does not delete existing local or backup copies. Reconnecting requires authorization again.

7. Public website privacy

These static pages contain no analytics scripts, advertising, embedded third-party content, forms, cookies set by site code or browser storage. Visiting them does not connect to the private assistant or request Google authorization.

When hosted on Cloudflare Pages, Cloudflare processes network requests and technical information to deliver and protect the website. See Cloudflare’s privacy policy. Links to other websites are subject to those websites’ policies.

8. Contact and policy changes

Owner: Raj Grover.
Privacy contact: was-wives-23@icloud.com.

Changes in data handling should be reflected here before new uses begin, with renewed authorization where required.