v1.0.0 — Meetings in one file, AI Voice companions, and automatic updates

Bedrock Panel v1.0.0 — Meetings in one file, AI Voice companions, and automatic updates

This release gathers the 1.0.0 betas into one build for Windows, macOS, and Linux. You choose how
meeting notes are written, a Batch page shows a whole run from start to finish, AI Voice pages
become companions with a face and a name they answer to, and Bedrock Panel now updates itself.

Upgrading from 0.9.9

• Install 1.0.0 by hand. 0.9.9 has no updater, so download it below. From 1.0.0 on, new
  versions install themselves (see Automatic updates).
• Your meeting notes keep their 0.9.9 look. An existing setup starts on Note style: Classic,
  so Separate Clean Transcript and Use Details Folder keep working. New installs start on
  Minutes Compatible.
• Processed Recordings is now called Completed Meetings. It is the same folder; nothing moves.
• A new Transcribed Recordings folder (default Documents\Bedrock Panel Meetings\transcribed)
  holds recordings that are transcribed but not yet analyzed. The three meeting folders must be
  separate and may not sit inside one another; Settings → Meeting says so the moment they overlap.
• Zoom key mappings moved from each Meeting page to Settings → Meeting → Zoom key mappings.
  Mappings you customised on a Meeting page are copied over on first launch.
• Create Joplin notes for analyses is off by default for new setups; existing settings are
  unchanged.

Automatic updates

About a minute after it starts, and every 12 hours while it runs, Bedrock Panel checks for a new
version. When one is out you choose: Install when I close Bedrock Panel, Install now and
restart, Skip this version, or Not now (asked again after 24 hours or at the next start).
If a recording or transcription is running, Install now asks before closing. The tray menu shows
the update's state and has Check for updates, and the editor header shows a vX available link.

Settings → Software → Updates has Automatic updates (on by default) and Include beta
versions (off by default). Check now works either way. The portable version never updates
itself; its prompt links to the release page instead.

Note style: Classic or Minutes Compatible

Settings → Meeting → Advanced settings → Note style has two choices:

• Classic writes <recording>-analysis.md exactly as 0.9.9 did, with Separate Clean
  Transcript and Use Details Folder. A small <recording>.bedrock.json beside the sources
  keeps track of your hand edits, action-item status and exports.
• Minutes Compatible writes one <recording>.md per meeting, with the meeting's facts in a
  header that Minutes and Obsidian can read (below).

Changing the style rewrites no existing file: the next analysis of a meeting uses the new style and
keeps the old file as a revision. Both styles get the protections below.

• Re-analysis you can trust. A re-analysis that would overwrite notes you edited by hand is saved
  beside the sources instead. The Analysis screen shows Review new, which opens it with
  Apply this version. Completed action items keep their status across re-analysis.
• Exports that never double up. Joplin, Obsidian and the post-batch task list are never written
  twice, and the task list never overwrites a list you already ticked.
• The analysis CLIs run with tools off, and a Claude sign-in failure shows the real reason and
  stops a batch.

One file per meeting (Minutes Compatible)

The meeting file carries YAML frontmatter in the Minutes
schema (title, date, duration, attendees, action items, decisions, open questions, speaker map) above
the usual sections. Bedrock Panel writes the frontmatter itself from the diarizer, calendar, and
capture data. Transcript lines carry start times taken from the diarizer segments; a turn that can't
be verified gets no time. The transcript is always in the file.

Three folders, each meaning one thing

A meeting moves from Unprocessed Recordings to Transcribed Recordings to Completed
Meetings. Analysis opens on the Transcribed folder: one row per waiting transcript with
Analyze and Delete, plus Select all / Analyze selected. The Completed button opens a
browser over Completed Meetings with View, Re-analyze, and Review new.

• Filed by the meeting's date, never the day it was processed: the calendar entry's start, else
  the moment recording started, else the date in the file name.
• Publishing can't half-move a meeting. Every source is copied and checked before the notes are
  written, and the originals leave last. A failure partway leaves the meeting where it was.

Batch, with a page that shows the run

Batch on the meetings page transcribes everything unprocessed, then analyzes what the
Analysis screen lists. It opens the Batch page: one status line over every file in order. The
file being worked shows its elapsed time, the next one reads Next, and finished rows say
Transcribed, Analyzed, or the full error text. Recordings the batch leaves out (an empty
recording, or one still being written) are listed as Skipped with the reason.

Cancel batch takes two taps. The file in progress finishes its step, the rest are cancelled, and
each row says where its file went. The summary stays until the next batch; tapping Batch again is
the retry.

Recording

• Recording no longer touches your speakers. The recorder used to open your speaker and play
  silence into it for the whole call, which could make incoming voices choppy on a USB speaker
  shared with Teams. It now opens no speaker at all, its audio processing runs off the app's main
  thread, and echo cancellation, noise suppression and automatic gain are off on its mic, for a
  faithful recording. Your mic is still on the left channel and everyone else on the right.
• Merge recordings: on Unprocessed Recordings, tick two or more and tap Merge selected to
  join them into one WAV. The originals are moved into .backups, not deleted.
• Quit warning: quitting while a recording or transcription runs asks first: Keep running or
  Close anyway.
• Empty recordings are discarded. A call that flaps on and off while connecting no longer leaves
  a silent recording that fails the next batch.
• The capture info travels with a renamed recording, so the meeting file gets the real start time.
• Long meetings no longer time out during analysis: the time limit grows with the transcript. A
  timeout that does happen shows the end of what the AI CLI printed.

Obsidian export

Settings → Meeting → Advanced → Copy meeting files into an Obsidian vault copies each finished
meeting file into your vault at the same path it has under Completed. A vault copy edited
elsewhere is never overwritten: a re-analysis lands beside it as <name>.conflict.md and the
Meetings page reports it. Vault sync stays with your own Obsidian setup.

AI Voice companions

An AI Voice page can now be a companion: a character with a face, a voice, a personality and a name
it answers to.

• A face that reacts. Choose Off, Side or Creature per page. The face listens,
  thinks and talks, its mouth follows the reply's real audio, and it shows the emotion the model
  picked. Face styles include a redrawn Cartoon human and a new Stone creature (Rocky), a
  faceted five-limbed figure whose teal accents brighten as it speaks.
• Five companions to start with: Mimi, Jarvis, CRT Terminal, Startup Intern and Pixie. Each has
  its own Piper voice and personality, and a companion's prompt is added to the page's profile
  rather than replacing it. Add your own under Settings → AI Profiles → Personas by paste, file
  or URL, or install packs from GitHub repositories, public or private through your GitHub
  connection. Installed companions are never silently overwritten.
• More voices. Multi-speaker Piper voices now play the speaker you choose: write it after a #,
  as in en_GB-semaine-medium#poppy. Load available voices lists each speaker separately.

AI Voice: wake word

Set Wake word: On on an AI Voice page and it answers only when you start with the companion's
name: "Hey Jarvis, what time is it?" Everything else said in the room is ignored, never shown, and
never sent to the AI.

• The mic opens by itself; the mic button is a mute that stays muted until you tap it again.
• The name alone opens a short Listening window for your request. One wake is one request.
• An optional Follow-up window lets you answer without the name, and by default opens only when
  the reply asked you something. "Thanks", "that's all", or "never mind" closes it.
• An optional Chime plays when the name is heard.
• Better hit rate: the start of what you say is no longer clipped, and near spellings of the name
  ("Hey Nord" for Norm) count after a clear "hey". Wake word → Test log writes one line per
  utterance to main.log for tuning.
• The first answer after the wake word is faster: the AI session starts when the page begins
  listening, not on the first request.
• The editor's live preview no longer interferes with the panel page's wake word.

AI Voice: model per page

Claude AI Voice pages have a Model setting in the editor: Default, Fable, Opus, or Sonnet. A
change reaches a running conversation before the next message. Haiku is no longer offered, because it
answered without using its tools; pages set to Haiku go back to the account default.

Also in this release

• AI Usage reports GitHub Copilot's new AI-credit billing, by
  @markhollingworth-worthit in
  #85. The app itself ships via app updates; this
  release carries the one-time move of your saved Copilot allowance to the new setting.

Known limits

• Batch state does not survive an app restart.
• The wake word still sends microphone audio to your speech-to-text server, which checks the text
  for the name. If that server is remote, it may keep what it hears under its own terms.
• The new companion defaults (Mimi's voice, "Hey CRT", Pixie) apply to new installs. An existing
  install keeps the personas it has saved.
• With automatic gain off, a laptop's built-in mic may record more quietly than in 0.9.9.
• On Windows, an update's signature is checked with PowerShell. Where that is a problem, turn
  Automatic updates off and install new versions from the releases page.

Downloads

• Windows: bedrock-panel-setup.exe (installer) or bedrock-panel-portable.exe
• macOS (Apple silicon): bedrock-panel-arm64.dmg, signed and notarized
• Linux: bedrock-panel_amd64.deb or bedrock-panel-x86_64.AppImage (the AppImage needs
  libfuse2)

Please include %APPDATA%\bedrock-panel\logs\main.log with any report.
Issues: https://github.com/TeeJS/bedrock-panel/issues