Recording starts even when Chrome won't share the tab's audio

Version 1.11.5. One fix for something new users ran into right after signing up.

Fixes

• No more "Extension has not been invoked for the current page". Chrome refused the recording when you pressed Record on a page the extension may not listen to: Call Copilot's own settings or sign-up page, a new tab, chrome://, the Chrome Web Store. Now such a recording starts on the microphone alone, and the side panel says "Recording your voice only…". The audio source label reads "Mic only".
• The call's audio joins with one button. Go to the tab with the call, click the Call Copilot icon in the browser toolbar and choose "Add This Tab's Sound": the tab's audio joins the recording that is already running, no restart. The extension confirms that the audio was added.
• Clicking the extension icon no longer changes the recording by itself. It only grants access to the tab, for a screenshot for example. Adding the audio or switching tabs is always an explicit button.

A whole call in one file, subtitles, a new AI model — and a direct line to me

Version 1.11.4. The theme is what happens after the call: taking the conversation with you, sharing it, and telling me when something does not work.

What's new

• The whole call as one file. The download menu of a call now starts with "Whole call (.zip)": transcript, subtitles, summary, notes, the chat about the call, the audio and the screenshots, in one archive. Below it, the separate files as before.
• Import a call back. "Import call" on the dashboard takes such a file and puts the call back into your archive, audio included — on another computer, in another browser, or after you deleted it. If the call is already there, you choose: replace it or keep both. Moving the whole database at once still goes through Settings → Backup (guide).
• Subtitles. A call downloads as .srt or .vtt, timed by the words themselves and split at sentence ends and speaker changes. Edits you made to the transcript are in them.
• Screenshots as one archive instead of one download per image.
• Feedback from inside the extension. A "Feedback" button in the side panel, on the dashboard, on an opened call and in Settings, and "Report" next to an error message. I read every message myself and reply by email. It carries the extension version, the installation ID and, when sent from an error, that error's text — nothing from your transcripts.
• A new cloud AI model. Summaries, chat and titles now run on OpenAI GPT-6 Luna, with Google Gemini 2.5 Flash Lite as a fallback. Requests go only to providers with zero data retention — no logging, no training on your data — and if one provider is down, another takes over instead of the AI stopping altogether. Gemini Nano on your own machine stays available in Settings.
• Titles from the whole call. Regenerating a title now reads the beginning, the middle and the end of a long call, not only its first minutes.

Fixes

• The chat answered in the language of the transcript, not in the one you picked for the AI. It now answers in the chosen language, like the summary.
• Dictation into the chat could return nothing when the AI language differed from the one you speak. Dictation now always recognises the spoken language by itself.
• A summary with many screenshots (30 or more) stopped updating, and regenerating it in another language silently kept the old one. Fixed.
• "Download audio" failed for a recording that had been paused and resumed. It downloads now.
• Notifications on the dashboard never appeared. They do now.

A recording that cannot connect now says so — and offers a way around

Versions 1.10.2 to 1.10.4, released over the last few days. The theme is one situation: the network between you and the transcription service does not let the connection through.

What's new

• A clear answer instead of an endless "Connecting to Deepgram…". If the transcription service cannot be reached, the recording gives up after five seconds and the side panel says what happened: the network, a VPN or an antivirus is blocking it, and no minutes were used. Before, the panel either waited forever or, since 1.10.2, for 45 seconds — and then showed an internal message that helped nobody.
• "Try via Call Copilot server". A button under that message. Press it and the audio goes to the transcription service through my server in Germany instead of connecting directly — which usually gets through where the direct connection does not. The server passes the stream on as it is and keeps nothing. The choice is remembered for the installation and can be switched off in Settings → Setup → "Connection"; it is never switched on for you. This covers transcription only — the AI features still connect directly. Guide.
• The message stays on screen until you close it, instead of vanishing after a few seconds while you are still reading.

Fixes

• The first recording right after a purchase could refuse to start (1.10.2). The start is now acknowledged immediately and its outcome reported separately, so a busy moment in the background no longer fails a recording that was actually coming up.
• The Chrome Web Store listing was trimmed to what the extension does (1.10.3).

System-audio recording — the whole computer, not one tab

The extension has a second audio source. It used to listen to a browser tab only; now it can record everything your computer plays: desktop Zoom and Teams, a messenger, a media player, a second browser.

What's new

• System audio as a recording source. Chosen in the new "Audio sources" panel, available both in the side panel and in the extension popup. The source is fixed for the duration of a recording and switched before the next one.
• The "Audio sources" panel brings the source and the microphone picker together in one place, with an explanation on each option.
• Precise messages when a start fails. If recording did not begin, the extension names the reason: the screen picker was closed, "Share system audio" was left off, or macOS screen-recording permission was not granted.
• Requires Chrome 141 or newer and macOS 14.2 or newer (or Windows, or ChromeOS). Step-by-step guide.

Fixes

• A microphone next to system audio is no longer quiet: input processing is now resolved per device, and a built-in mic gets echo cancellation back.
• The marketing-consent checkbox at sign-up is saved again.
• The audio-source hint no longer sticks around after a device has been picked.
• Popup: the window grows with the Quick Actions card, and a Record click on a page that cannot be recorded now gives a clear answer.

Recordings no longer stop halfway through the minutes you paid for

A metering bug hard-stopped recordings at roughly half the balance they were entitled to. Fixed.

Fixes

• A recording now runs to the end of the minutes you paid for: already-billed minutes were being counted twice.
• The sidebar countdown no longer loses its start anchor when the panel opens at the same moment a recording begins.

Demo opens the sidebar on the first click

Two releases over the weekend: the main one on Sunday evening and a follow-up with fixes on Monday morning.

What's New

• The demo now opens the sidebar on the first click
• The main release shipped Sunday evening after live testing
• A follow-up release on Monday morning fixed a couple more bugs

Fully simulated demo and fixed onboarding

The demo no longer depends on the transcription service — you can try the product before registering.

What's New

• The demo mode is fully simulated: it shows how the product works without spending transcription minutes
• The demo is available before registration — see the product first, decide later
• Fixed onboarding: the first run walks you through the steps to your first recording again

Mandatory registration and abuse protection

Two releases shipped in a single night: a large bug-fix pass and new rules for anonymous users.

What's New

• Registration is now required: free tokens are no longer handed out to anonymous users
• Large bug-fix pass across the whole extension, based on a full review
• Added in-product feedback collection

Transcription token protection and Install ID

A security release: closed a hole that let someone burn forty hours of transcription in a single day.

What's New

• Closed a hole in how the extension obtains its Deepgram token (the service that powers transcription): previously anyone could request one
• Token requests are now tied to an Install ID — an identifier of the extension install
• Known limitation: versions before 1.7.11 do not send an Install ID, so updating the extension is required to keep transcribing

Public launch in the Chrome Web Store

Call Copilot is officially live in the public Chrome Web Store.

What's New

• The extension is available to everyone in the Chrome Web Store
• Live call transcript in the sidebar — no bot joins the call
• Three-page onboarding and a demo session for the first run
• Data export and import
• Fixed a bug with tab switching