Help and About — Peter Joseph Reply Helper
Version 1.3.10
Peter Joseph Reply Helper captures a user-selected YouTube, X, Reddit or Facebook post, prefers a complete English transcript, records where separately attached images belong and builds one clipboard package containing the source material, reply context and project-authored master prompt.
The extension is designed for a manually controlled workflow. It does not automatically post replies or submit anything to ChatGPT.
Desktop Chrome 150 or later for this release.
A compatible device and Chrome-managed on-device model for direct-media audio transcription.
A one-time browser-managed model download may be required.
Caption capture and manual transcript entry remain available when the local audio model is unavailable.
Basic workflow
Open an individual supported post, video or Reel.
Open the Peter Joseph Reply Helper side panel.
Accept the first-run privacy notice.
Press “Capture current social-media tab”.
Grant the optional supported-site/media-host access requested for that platform.
Let the extension finish its caption, translation and direct-media checks.
Review the captured source and transcript.
Paste the exact reply target and surrounding thread when relevant.
Mark where separately attached images belong.
Press “Prepare complete package”.
Copy the package, attach relevant images and manually submit it to the selected Custom GPT.
Use a complete English caption or transcript exposed by the source platform.
On YouTube, check exact signed timed-text requests already loaded by the current player.
Ask YouTube's own player interface for an alternate directly available media resource when required.
If only a complete non-English caption track is available, try Chrome's local Translator API.
If no usable caption transcript exists, process the selected video's directly available media/audio with Chrome's built-in on-device multimodal AI.
If captions, translation and direct-media transcription fail, stop and allow a manual transcript.
The extension never silently treats a non-English transcript as English.
The local media method processes downloaded media data rather than waiting for the page video to play from start to finish. Decoded audio is split into 20-second core windows with 2-second overlap. A result is marked complete only after every planned window has covered the full decoded duration.
If a page contains several loaded videos, the extension rejects candidates whose decoded duration does not match the selected player and continues through a bounded candidate set.
YouTube first tries token-free media access and can then use the already-open page's current session for a same-origin player request concerning the selected video. Authentication-cookie values and derived session headers are not stored, copied into diagnostics or sent to the extension developer.
Facebook may briefly activate the selected player muted for no more than 900 milliseconds solely to expose caption or direct-media requests. It restores the previous pause, time and mute state and does not transcribe that playback.
There is no full-duration normal-speed page-playback transcription, tab-audio recording or browser speech-recognition fallback.
Follow-up actions
After preparing a package, the extension can copy instructions for:
More blunt
No holds barred
More factual
More technical
Different angle
Another 12 replies
Concise summary
Claims plus individual replies
Integral focus
Brutal/rude
Absolute conviction
Savage sarcasm
George Carlin-inspired satire
Stafford Beer-informed cybernetics
These controls change the requested tone or task. The prompts retain factual-grounding, source-safety and protected-trait safeguards. Some modes can request strong language.
Chrome asks for supported-site access when a platform is captured for the first time. Related media-host access is used only for the selected source, such as Googlevideo media belonging to a selected YouTube video or Facebook CDN media belonging to a selected Reel.
Settings and drafts stay in local Chrome storage. Raw media/audio is held temporarily in browser memory and is not intentionally retained. The extension has no developer-controlled server, account system, analytics, advertising or automatic posting.
Open Settings to:
read the Privacy Policy;
open Help and About;
copy a privacy-sanitised diagnostic report;
clear the current task; or
delete all saved settings/drafts and revoke optional site permissions.
The transcript model is unavailable
Confirm Chrome is version 150 or later and that the device supports Chrome's on-device audio-capable Prompt API. The browser may need to download its model. If unavailable, paste a complete transcript manually.
A platform caption cannot be found
Press Capture once and allow the automatic workflow to finish. Platform players change regularly. If captions and direct media are both unavailable, the extension will stop rather than invent missing speech.
The wrong Facebook, X or Reddit video was found
Keep the intended post or Reel selected and visible, clear the task and capture again. The extension compares decoded media duration with the selected player and rejects obvious mismatches.
YouTube direct-media fallback fails
Reload the selected YouTube page once and capture again. The workflow is deliberately finite and will stop after bounded direct candidates rather than enter a retry loop. A manual transcript remains available.
Permissions need to be reset
Open Settings, expand Privacy reset and press “Delete all saved data and site permissions”. The next capture will request only the platform access needed at that time.
Support needs diagnostics
After a failed capture, open Settings and press “Copy capture diagnostics”. Review the report before sending it. Signed media URL values are removed, and the report is not transmitted automatically.
The extension detects image material in captured X, Reddit and Facebook originals. Images in a manually pasted reply target or thread must be marked manually. When there are several images, label them in attachment order, for example:
Image 1 — original post
Image 2 — earlier thread comment
Image 3 — exact reply target
Automatic captions, translation and local AI transcription can contain errors.
Verify important quotations against the original source.
The extension does not bypass logins, private groups, platform restrictions or unavailable content.
It does not automatically download or upload separately attached images.
Transcript availability and speed depend on the platform, media resource, Chrome model and device hardware.
If every automatic method fails, use a manually supplied complete transcript.
This is an independent, unofficial project. It is not produced, approved, sponsored or endorsed by Peter Joseph. The name “Peter Joseph” is used descriptively to identify the intended analytical perspective and optional Custom GPT destination. The built-in prompt is a project-authored synthesis and must not be represented as Peter Joseph's own wording or approval.
Source work: “Peter Joseph.jpg”
Photographer: iTomath
Source: Wikimedia Commons, File:Peter Joseph.jpg
Licence: Creative Commons Attribution-ShareAlike 3.0 Unported (CC BY-SA 3.0)
Changes: cropped, resized and converted into extension icon assets
Attribution does not imply endorsement by the photographer, Wikimedia Commons or Peter Joseph.
Questions can be sent to:
peterjosephsangertranslator@gmail.com
Do not email private post content, account details, authentication information or material you are not authorised to share.