Plática Notes

A Chrome/Chromium extension that records Google Meet transcripts and chat locally — no servers, no accounts, no network calls.

Install it from the Chrome Web Store. That is the build to use; it updates itself. To run an unreleased build instead, see docs/TEAM-INSTALL.md.

How it works

Plática Notes reads Meet's own live captions straight from the meeting's WebRTC data channels (the same stream Meet uses internally), instead of scraping the on-screen caption DOM. This means:

• No caption band on screen — captions never have to be turned on visually, so the video stays full-bleed.
• Works in background tabs — data channels keep flowing when a tab is backgrounded, so parallel meetings in different tabs are all captured.
• Speaker names resolve from the meeting roster; your own name comes from Meet's user info. Chat messages carry their sender too.

Everything is decoded in the page and kept on your machine.

Features

• Live transcript + in-meeting chat, saved as a readable Markdown .md per meeting: a short header (title, meeting link, time, participants) followed by speaker-attributed turns. Each turn carries the speaker, the full local timestamp with its UTC offset, how long the turn lasted, the elapsed time since the meeting started, and the text — so a tool can place any line in absolute time and bound a speaker's run without re-deriving either.
• In-meeting transcript panel — a floating, scrollable live transcript inside the Meet window, so you can scroll back and re-read without leaving the call. Includes a search box that filters the timeline, and a footer input to add your own notes.
• Bookmarks & notes — drop a timestamped bookmark with Alt+Shift+B (⌥⇧B on macOS), or type a note in the panel; both land on the timeline in context and are written into the saved file as ### heading blocks, distinct from speaker turns.
• Hide all on-screen UI — a popup toggle and Alt+Shift+H (⌥⇧H on macOS) hide every extension element (controls, panel, toasts) for screen-sharing or demos; recording keeps running while hidden.
• Caption alternatives (Settings, on by default): each turn also records the raw versions Google streamed before the final caption, so words the final caption dropped can still be recovered later. Turn it off for a cleaner file.
• Speaker-attributed lines (others and yourself), with the closing sentence captured in full on leave. Chat is captured too — both what others send and the messages you send yourself.
• Participant join/leave markers — when someone joins or leaves during the meeting it is marked inline on the timeline with a timestamp (in the speaker's colour), not just as the final participant list in the header.
• Default caption language (Settings) — every new meeting starts in this language; the in-meeting language pill overrides just the current meeting and is not saved. A first-run welcome page lets you pick it; fresh installs default to English.
• Recording toggle (● Rec / ⏸ Rec off): pause and resume capture without leaving the call. Off stops everything new — transcript, chat, join/leave markers, and notes — while whatever was captured before you turned it off is still saved when the meeting ends. The state survives a page reload.
• Wipe recording (🗑): clears everything captured in the current meeting so far (transcript, chat, notes, presence markers). Click once to arm, once more to confirm. If nothing remains afterwards, no file is written.
• Per-meeting privacy pill (🔒 / ☁️): private meetings download to a separate folder you can keep out of cloud sync.
• Local history of the last 30 meetings, with re-download / delete.
• Parallel meetings, mid-meeting reload (resumes), same-tab rejoin and soft-nav to a new meeting are all handled.
• Optional debug log (off by default) for diagnostics.
• Zero network: nothing is uploaded anywhere.

Output

Files land in your Downloads folder:

A file is named <date>_<time>_<title>_<meeting code>.md, for example 2026-08-04_15-59_Payments_status_exb-zusa-qnc.md. The date comes first so a folder sorts chronologically, there are no spaces, and the meeting code (absent when the meeting has none) lets you pull every occurrence of a recurring meeting with one glob.

Reading the file with a tool

Every timeline line carries the same two clocks: a full local ISO 8601 instant with its UTC offset, and the time elapsed since the meeting started as +HH:MM:SS. Both are fixed width, so one pattern matches the whole file however long the meeting ran.

Join and leave markers are ### Joined · <name> · <instant> · +<elapsed> (and ### Left · …). Two rules matter when deciding who was in the room at a given moment:

• No join marker means the person was already there when the recording started — not a missed event. Only arrivals during the meeting are marked; everyone present at the start is in the participants header instead. The same goes the other way: no leave marker means they were still there at the end.
• Your own arrival and departure are never marked, and neither is anything in the first ten seconds, which is when Meet streams the roster of people already in the call.

So "listed in the header, no join marker, speaks at 12:20" means present from the start; "no marker and never speaks" means present and silent.

Cloud sync (optional)

Sync Downloads/Platica Notes/ with your own tooling (e.g. Google Drive for desktop, rsync). Keep Platica Notes private/ and Platica Logs/ out of the synced folder — the private folder holds meetings you chose not to sync, and the debug logs embed the full transcript. Meetings marked private are excluded from the debug log entirely, so the privacy flag is honored on every output path.

Settings

The toolbar popup keeps just the in-the-moment control, Hide all on-screen controls (also Alt+Shift+H, or ⌥⇧H on macOS), which hides every extension element for screen-sharing or demos while recording keeps running. The popup also links to the meeting history and to the Settings page.

The Settings page (open it from the popup) holds everything else:

• Default caption language. Must match the spoken language, or the transcript comes out empty. It seeds every new meeting; the in-meeting pill overrides one call without changing the default.
• Private by default. New meetings start private and route to the private folder.
• Folders. The public, private, and debug-log download folders, relative to Downloads.
• Caption alternatives. Adds the raw caption versions under each turn for recovering dropped words. On by default; turn off for a cleaner file.
• Merge rejoined visits. If you accidentally leave and rejoin the same meeting within ~40 minutes, the visits are folded into one .md instead of one file per visit. On by default; turn off to keep one file per visit. A daily recurring call is never merged across days, and a private visit is never folded into a public file.
• Ask language at meeting start. Shows a prompt when each meeting begins to confirm or switch the caption language. Off by default — for people who meet in several languages. It never blocks recording.
• Debug log. Writes a full .jsonl diagnostic per meeting to Downloads/Platica Logs/. Off by default. Private meetings are never logged.

Scope & limitations

• Google Meet only.
• The caption language must match what's spoken, or the transcript comes out empty. Set a default in Settings; override one call from the in-meeting pill.
• A very short meeting may finalize before your name is resolved, falling back to a generic "Speaker N" label.

Building from source

See CLAUDE.md for build, test, and architecture notes.