Privacy in TRY Bookmarks

What the application processes

• Account and sign-in information. Account identifiers, profile information provided by your sign-in provider, linked-account records and session information. Telegram sign-in uses your Telegram identifier and available profile fields. Google or Apple sign-in, where configured, uses the verified information returned by that provider.
• Your library. Saved URLs, titles, descriptions, extracted text, previews, uploaded originals, notes, tags, collections, reading state and derived summaries, transcripts, embeddings and connections between materials.
• Requests and history. Messages and responses in the built-in Ali chat, citations and conversation records. Processing jobs, assistant action receipts and stored results can contain library content, exported data or generated analysis.
• Operational information. Connection names and permissions, request times, processing status, errors, model usage and cost metadata. Normal network requests also expose information such as your IP address to the services receiving them.
• On your device. Browser storage retains your application session and preferences. Some interface state or unfinished work may remain on the device. Removing a local installation does not by itself delete the server account.

The application uses this information to sign you in, store and retrieve your materials, search and analyze them, execute authorized actions, synchronize devices and maintain the service. A saved link can contain personal or sensitive information; choose carefully what you save and share with assistants.

External services and AI

• Fly.io hosts the primary application, its PostgreSQL database and application files. A Vercel compatibility frontend also exists for older links.
• OpenAI receives the content needed for the AI features you use. This can include questions, retrieved library passages, text for embeddings, images or audio for analysis and transcription, and text for speech in application surfaces that support it. Ordinary saved-link enrichment can also invoke AI processing.
• HikerAPI is used for supported Instagram retrieval when configured and available. Tavily is used for supported web search when configured and available. Retrieving a URL or its media also contacts the source website or media host.
• Telegram processes messages, files and identity information you exchange with the application bot. Google and Apple receive the sign-in requests you initiate through their configured sign-in methods.
• Your chosen assistant provider, such as Anthropic or OpenAI, receives the tool results delivered to Claude, ChatGPT, Codex or another connected client. These may include complete library texts, files or exports if you request them and your connection permits it.

Feature availability and configuration determine which services receive a particular request. The application does not need the password or model-provider API key from your assistant account to grant MCP access. External providers apply their own policies to training, retention and deletion of copies they receive.

What a connected assistant can access

You authorize a connection to a selected TRY Bookmarks account through OAuth. The default permission is reading. Reading can expose saved materials, original files, existing history and analysis results; it is not limited to titles or previews. Changes require write permission. Supported paid or remote processing requires analyze permission and, when the operation saves results, write permission as well.

In the application menu, open Connections to inspect a connection, narrow its permissions or revoke it. Revocation prevents further authenticated access by that connection; it does not erase the library, stored action history, earlier exports or copies already received by an assistant. Human sessions and account credentials are managed separately and cannot be managed through agent credentials.

OAuth access credentials expire after up to one hour; a grant has an absolute lifetime of up to 90 days, with rotating refresh credentials when requested. Expiry and revocation are access controls, not a promise that the related records or previously returned data are deleted at that moment. Stored external bearer credentials use hashes rather than recoverable plaintext tokens.

Storage, history and deletion

The service stores library content, account data, conversations, processing records and assistant results in its application database or files. These categories do not have a single automatic expiry period. Ordinary library content, conversations, assistant action history and stored results remain separate from the temporary records and cleanup rules described below.

Automatic cleanup is configured to remove collected Telegram group messages and emotion-log records older than 7 days, and behavior-event records older than 180 days. These category-specific checks run every six hours while the service is running. Expired sign-in requests are checked for cleanup every five minutes. Service outages or failed cleanup jobs can delay removal; these rules do not delete all other records associated with the same account.

You can delete individual bookmarks and collections through the available application controls. Deleting an item does not prove that all copies have disappeared from conversation history, analysis results, exports, cached files or backups. Historical exports or snapshots can preserve content from before an item was deleted.

The backup job is configured to create daily database copies and remove matching copies older than its 14-day cleanup threshold. This is a configured cleanup rule, not a guarantee of erasure on an exact date: cleanup depends on the job running, files can contain older records, and separate operator recovery copies or hosting snapshots may have a different lifecycle. Deletion from the live library does not immediately rewrite existing backups.

Incognito mode reduces persistence of supported chats, searches and assistant receipts. It does not prevent content being sent to an external AI or source service when that feature is used, and it is not a guarantee that no operational metadata or explicitly saved library content exists. Incognito assistant result and task caches expire after one hour; incognito session state expires after 24 hours. A restart or cache eviction can make them unavailable sooner. These limits are separate from ordinary stored history and do not delete copies already received by external providers.

To request access to your data or account-wide deletion, email rasuliazamat@gmail.com. The operator reviews these requests individually and may ask you to verify that the account belongs to you. Account-wide deletion is requested through this contact channel, separately from deleting individual items in the application. The handling of live records, retained history, files and recovery copies must be considered separately; there is no single automatic deadline that removes every copy at once.

Shared TRY access

When shared access checks are enabled, TRY Bookmarks receives a verified Telegram ID, account creation time and application identifier from the TRY apps. It checks membership in @azamatrasuli and maintains one shared 24-hour trial across the TRY family.

Paid AI calls share a $1 per-person limit per UTC day. Accounting uses an attempt identifier, timestamps, reserved and final costs and available input, output, cached and audio token counts. Message content, audio recordings and model responses are not sent to this accounting ledger.

Detailed access and cost events are retained for 90 days. Compact application and attempt UUID markers remain to prevent replay. A minimal record of the used trial remains to prevent restarting it.