Prompta.cam aims to process only the minimum data required for the teleprompter, AI features, library, payments, support, bot communication, and additional Managed Bot products. We do not sell user data.
1. Telegram profile
When the Mini App is opened, Telegram provides signed session data. Prompta.cam verifies the signature on the server and may store the Telegram ID, username, name, language, and technical account update dates.
2. Video, camera, and microphone
The main video recording in the Mini App is created on the user’s device. In the normal recording flow, Prompta.cam does not upload or store the full recorded video file on its server. A user’s separate submission of voice, audio, or video to Personal AI Coach is described in section 4.
When AI Live is enabled, individual compressed frames from the current camera are sent to the Prompta.cam server and then to the AI provider for analysis. Prompta.cam does not store these frames in D1 or R2.
When Speech Context is enabled, local VAD identifies short speech chunks. Only those chunks are sent to the server for recognition. Recognized text is kept in Mini App memory in a short rolling window and is not written by Prompta.cam to D1, R2, or localStorage.
3. Scripts and AI
Script text is sent to the AI provider only when a feature needs it, including the AI editor, Take Director, contextual explanations, and related features. In Generate mode, the user may also explicitly submit their own text instruction to AI. Prompta.cam does not store ordinary Mini App AI request text, frames, or model responses in technical telemetry unless a separate feature is explicitly described otherwise below.
Data sent to a third-party AI provider is processed to fulfil the request under that provider’s terms and privacy policy.
4. Managed Bots and AI Coach
When Prompta.cam is opened from a Coach exercise, the server verifies the task owner using the signed Telegram session. Recording by itself does not send video. Only an explicit press of Send to Coach uploads the recording through the Prompta.cam server to the user’s private Coach chat in Telegram and starts the existing review. The Telegram copy is subject to Telegram’s retention and deletion rules. Separate download or sharing of the recording remains available.
To link a recording to an exercise and prevent duplicate submissions, Prompta.cam may store the task instructions, current focus, owner and bot IDs, links to the source and resulting sessions, link expiry, upload or queue status, and minimal Telegram file metadata. Link expiry does not delete training history. Recording bytes are held in temporary memory during upload and processing but are not stored by Prompta.cam in D1 or R2. Exercise instructions may be sent with recording materials to produce the review. Bot tokens and signed session data are not included in task links or public task API responses.
If a user creates a Managed Bot through Prompta.cam, we may store the owner and bot Telegram IDs, bot username or name, selected product, access status, webhook state, technical errors, and aggregated monthly AI usage. The user remains the owner of the bot in Telegram.
Prompta.cam receives the technical Managed Bot token through Telegram’s manager API only to manage the bot. The token is not stored in D1 in plain text and may be present briefly in Worker memory. A hash of a random secret is stored to verify the webhook.
During Coach onboarding, Prompta.cam may store in D1 the user’s selected training goal, planned sessions per week, onboarding stage, link to the baseline training session, current focus, and counts of completed takes. This is needed to continue the program between Telegram sessions. The same media processing and retention rules described below apply to the baseline session.
Coach v101 may additionally store the user’s selected experience level—Beginner, Intermediate, or Experienced—and the status of an optional video diagnostic: offered, skipped, processing, or completed; a link to the diagnostic session; a structured starting-program recommendation; and a short diagnostic summary. The optional diagnostic uses a video or video note of about 30–60 seconds and the same temporary media processing rules described below. The transcript and structured diagnostic review may be stored as a Coach training session; Prompta.cam does not store the source video or extracted frames. The diagnostic is used to recommend a starting program and focus, not to calculate a percentage score or infer the user’s character, emotions, or health.
Only one Coach program is active at a time. For each goal, Prompta.cam may store a separate structured track containing the saved current focus and its key, training rhythm, exact number of completed sessions, link to the baseline session, and links to the current exercise plan and 28-day cycle. When programs are switched, the previous track is paused rather than erased. On return, Prompta.cam restores saved progress and, if the saved plan or cycle remains valid, continues it; otherwise it creates the next suitable exercise or cycle without deleting completed history. Changing the experience level also preserves program history while allowing later exercises to use a different difficulty.
Coach v102 may store a separate categorical skill profile for each saved program. It covers confidence and observable on-camera control, speech structure, natural delivery, camera connection, pace and pauses, expressiveness, hook, and call to action. After each successfully analyzed training take, Prompta.cam may save one structured signal and short explanation for each skill, the current priority, and a link to the source session. The user sees only the states not enough data yet, currently inconsistent, already consistent, and next priority; Prompta.cam does not calculate an overall or percentage score. Already consistent requires repeated comparable observations. A paused program does not update the active program’s profile. Profile updates are part of the same atomic operation as training-session completion and do not create another AI request or resend stored transcripts.
For Today’s training, Coach may create and store a deterministic block of three to five exercises based on the selected goal, current focus, weekly rhythm, and selected experience level. Exercises may include real camera-speaking practice: one take without restarting, a short explanation of one idea, a hook, a response to a comment or objection, reducing one idea from 60 to 30 seconds, speaking without a full script, and a final real-world Story, Reel, expert, or sales clip. Prompta.cam stores exercise text and order, difficulty metadata, assignment and completion state, and linked recording-task and training-session IDs so an unfinished exercise can be resumed and credited once. Stored transcripts are not sent to the AI model when this plan is created, and percentage scores are not calculated.
Coach v103 may turn one planned exercise into a stored three-stage lesson: a short explanation and concrete example, a first take, an immediate repeat with one correction, and transfer of the same skill into a new real-world context. Prompta.cam may store lesson stage and status, explanation, example, main and transfer tasks, one length-limited correction, and exact links to assigned recording tasks and completed training sessions. Each submitted take uses the existing Coach analysis request; v103 does not create another AI request or store another transcript copy. The first take and repeat update take-level focus and skill data but do not complete the planned exercise or increase the 28-day cycle count. Successful transfer completes and credits the exercise once. If existing limited progression rules switch or restart the focus earlier, the old lesson is closed as replaced and the next lesson uses the authoritative current focus.
Coach v104 uses an adaptive exercise library and combines the selected program, self-selected level, and current categorical skill priority. For every new plan, Prompta.cam uses stored assignment history and selects the least-used suitable option instead of choosing randomly. The system may store the program, level, target skill, main and transfer exercise identifiers, deterministic learning copy, exact plan and step linkage, and assignment time. This history is isolated by user and program, contains no recording, selected frames, or transcript, creates no additional AI request, does not change the authoritative focus, and does not itself add plan or 28-day cycle credit.
Coach may also maintain a rolling 28-day training cycle, storing a snapshot of the selected goal and rhythm, start and end times, planned and exact completed session counts, and links between the cycle, assigned exercise, recording task, and completed session. Only a successfully analyzed Prompta.cam exercise assigned through this chain is credited. Opening a task, recording without submitting, sending media directly to Telegram, reprocessing, and an expired or replaced cycle are not credited. Cycle calculation is deterministic, does not read stored transcripts, does not call AI again, and shows exact numbers rather than percentages.
Coach v106 does not automatically start a new 28-day cycle after the current cycle’s goal is reached or its term ends. Prompta.cam may store a link to the completed cycle and the completion reason, previous rhythm and level, and the user’s explicitly selected next step: continue, change rhythm, raise level, or switch to another program, together with the resulting rhythm, level, or target program. This decision record contains no media, selected frames, transcript, or new AI response and does not itself add training credit. If a saved program has been paused for at least 60 days, Coach may offer a short 30-second re-baseline; choosing it preserves the original baseline and completed-session history, and the new take is processed under the existing Coach media and analysis rules.
Coach v107 personalizes only messages that bring the user back to an existing program. If the user enabled reminders, Coach considers the active program’s weekly rhythm, the exact number of credited sessions, and the time of the last training session or reminder. No reminder is sent when the weekly goal is complete or while a choice after a completed 28-day cycle remains unfinished; without a new session, another reminder may be sent no more than once every seven days. After every second confirmed transition to a new focus, Prompta.cam may queue one deduplicated milestone message linked to the completed session. /apply creates a deterministic task for applying the current skill in a real Story, Reel, expert answer, or sales clip. Opening the task does not call AI, upload media, publish content, or add training credit; publication or submission always requires a separate explicit user action.
Coach v108 lets the user explicitly record and submit that real-audience task for review through Prompta.cam. The recording goes through the existing protected Coach media pipeline, speech recognition, delivery measurements, and—when video is present—analysis of selected frames. The request uses the ordinary Coach AI limit. The result is stored with the Coach session and recording task. This review does not complete a lesson or exercise, increase the 28-day cycle count, change the current focus or skill profile, create a milestone, or reset reminder timing.
When the user explicitly opens Share to Story, Prompta.cam creates an owner-bound progress snapshot from structured cycle counts, current focus, and the latest categorical signal. The Mini App creates a 1080×1920 JPEG card on the user’s device. Only that generated card is uploaded to Prompta.cam and made available to Telegram through a random temporary HTTPS link for the native Story editor or explicit download. The user presses the send button and decides whether to publish the Story, cancel, or download the card. The card and link remain available for no more than 24 hours and are removed by scheduled cleanup. This flow does not read a stored transcript, upload source media, call AI again, calculate percentages, or add a public profile link.
Coach may create weekly and 30-day reports and a shareable progress card from stored structured session observations, categorical focus decisions, and measurable speech values. Stored transcripts are not sent to the AI model again to create a report. Automatic reports are enabled for the active program and can be disabled with /reports. Training-rhythm reminders remain off until the user enables them with /reminders. To prevent duplicates and retry temporarily undelivered messages, Prompta.cam may store these settings, next delivery time, and minimal delivery events for reports, reminders, and milestone messages, including links to the source session, attempt count, status, and technical error.
Personal AI Coach accepts text, voice messages, audio, video, and video notes. When media is submitted, Coach temporarily obtains the file from Telegram servers using Telegram file_id and uses it for speech recognition and review. The original media file is not stored by Prompta.cam in D1 or R2.
For voice messages, audio, video, and video notes, Coach may temporarily send downloaded media bytes to an internal Cloudflare media-processing container with FFmpeg to measure speech activity and silence intervals. For video, the same temporary processing may extract a small number of JPEG frames from different moments. Original media and extracted images are not stored by Prompta.cam in D1 or R2. Selected frames may be sent to the AI provider for visual review.
The sequence of frames is a sparse sample rather than continuous video. Coach may compare only visible facts across the supplied moments: framing and person position, visible lighting, composition and background, visible changes in posture or the position of hands and body, and the general direction of the face toward the camera in selected frames. This is not precise eye tracking and does not prove what occurred between samples or throughout the clip.
If extraction of multiple frames is unavailable, Coach may fall back to the Telegram thumbnail and evaluate only static properties of that single frame. One thumbnail is not used as evidence of behavior over time.
To continue training, Prompta.cam stores in D1 a limited recent text context for Coach and, for training media, the transcript, measurable speech metrics, derived speech-activity and pause intervals, structured review, structured visual categories or observations and selected-frame timestamps when available, the next focus, and training status. Intervals are determined by a fixed loudness threshold and are not assessments of emotions, intent, pitch, or semantic sentence boundaries. Original media, thumbnail, and extracted JPEG frames are not retained in training history. /reset clears the recent conversation context but does not delete training metrics history or stored program tracks.
Coach must not infer a user’s identity, age, health, character, emotions, or intentions from an image. The full user bot token is not exposed to the browser or administrator dashboard.
5. Library
Script templates and favorite AI recommendations are saved in D1 only after the user explicitly selects Save or adds an item to favorites. The user may delete these records through the library interface.
6. Payments and AI tokens
For payments, Prompta.cam stores order identifiers, Telegram payment charge ID, the plan, package, or Managed Bot product, Stars amount, status, and dates. Prompta.cam does not receive bank-card data.
For AI tokens, Prompta.cam stores the balance, purchases, charges or refunds, and technical operation references. AI telemetry stores technical fields such as mode, status, and latency, without the camera frame, script, or model response.
7. Messages, support, and broadcasts
When the user writes to the main Prompta.cam bot, text messages and technical records about the message type may be stored in D1 so the service owner can view the conversation history and reply through the internal dashboard. For photos, videos, voice messages, and files sent to the main bot, only the fact or type of the message or its caption is stored, not the file itself. Media handling inside Personal AI Coach is described in section 4.
When the user contacts /support or /paysupport, the request text and technical ticket number are also stored and may appear in the same conversation history.
For selected broadcasts, Prompta.cam stores campaign text, the selected Telegram ID list, delivery status, and a technical error if Telegram did not accept the message. Sent means the Telegram Bot API accepted the message; Prompta.cam does not receive a separate read receipt.
The user may disable selected broadcasts with /unsubscribe or /stop and enable them again with /subscribe.
8. Consents
Prompta.cam stores the version of the accepted Terms and published Privacy Policy and the consent time. A Privacy Policy update alone does not block purchases with another checkbox; purchase consent remains Terms-only.
9. Retention, deletion, and security
User templates and favorites are kept until deleted by the user or removed through administrative cleanup. Generated Coach Story cards, owner-bound snapshots, and random media links remain available for no more than 24 hours.
Message history, Coach context, selected experience level, optional diagnostic status and structured recommendation, saved, paused, or completed program tracks, plans and exercises, lesson stages and limited corrections, 28-day cycle counts, post-cycle decisions and their task or session links, transcripts, speech metrics, derived activity and pause intervals, structured visual training metrics or observations, settings and delivery events for reports, reminders, and milestone messages, support history, and broadcast logs may be kept to continue the conversation, resume progress, track improvement, provide support, and maintain an operational audit until user-requested or administrative cleanup. Payment records may be retained longer for refunds, disputes, and accounting.
A request to delete or anonymize Prompta.cam data does not delete the user’s Telegram account and does not necessarily delete a Managed Bot owned by the user in Telegram. Prompta.cam may delete or disable its own context, training-history records, configuration, and webhook within technical limits; minimum payment records may be retained for refunds, disputes, and accounting.
A deletion area is available at the end of the Privacy Policy inside the Mini App, where the user may submit a request to delete or anonymize Prompta.cam data. A request can also be made through the public Delete account page or by email.
Prompta.cam uses server-side Telegram initData verification, a webhook secret, a production-origin gate, a protected administrator session, and server-side storage of AI keys. Administrator keys, Telegram initData, and bot tokens must not be included in conversation logs.
10. Support
For privacy and payment questions, use /support or /paysupport in the main Prompta.cam bot or email support@promtcam.com. To opt out of selected broadcasts, use /unsubscribe.