MeetOnTrack 0.1.14 for Mac

MeetOnTrack 0.1.14

Technical beta for Apple Silicon with macOS 15+. Windows is a separate package.

• Check AI is reliable for live analysis. Its test file used to sit in a
  temporary folder that the AI tool is not allowed to read, so the check often
  failed although AI worked.
• Speech models have human names everywhere, in the language of the page, for
  example "Parakeet v3 (25 languages)" or "Parakeet v3 (языков: 25)". Engines
  are shown as "FluidAudio" and "OpenAI", also in the recorder's setup window.

Everything from 0.1.13 is included: the automatic speech model choice, the
live transcript that starts mid-call, the red banner, plain error messages
and call audio that recovers by itself.

Cloud AI receives the selected text context of the task it runs. Cloud speech
recognition receives the audio only when you choose a cloud engine.
Google/SMTP integrations remain optional and off for new installations. No
usage-statistics feature is included.

Update

Finish the call and quit Capture, then run:

curl -fsSL https://meetontrack.com/install.sh | bash

Versions 0.1.10 and 0.1.11 do not show this update in the app: the
project's GitHub address changed, and those versions reject release pages at
the new address. Run the command above once. Versions 0.1.12 and 0.1.13 show
this update in the app.

The installer verifies the pinned package checksum and preserves recordings and
settings. Automatic installation is deferred; version notifications are enabled
by default and can be disabled in Settings.

Verification and limits

Source: 5aaaec8ec3e5bfad28929e4f0dc9262ec706a375.
ZIP SHA-256: e6e95a901a9e96d9d7bef14fad184641fcf7f4479aadc17a1e4398ebabb5bb09.
382 Swift tests in 50 suites: 379 passed; 3 retranscription tests need a
LocalVQE development library that is absent from a source checkout (the package
includes it). 362 web tests: 359 passed, 2 optional skips, 1 needs Pillow.
Exact archive: 2,823 file hashes, fresh install, repeat install preserving
synthetic settings and recordings, native ad-hoc signature, embedded
Python/server import passed.

This package is ad-hoc signed and not Apple-notarized. Retain normal macOS
security protections. Real AI and cloud speech calls were not exercised by the
release checks; run the Getting started checks after choosing. Full
clean-machine permission/audio acceptance and the device/reconnect matrix
remain follow-up checks. Speaker separation may need manual correction. Local
folder retrieval supports UTF-8 text, Markdown, CSV and source code, not general
PDF/Office ingestion.

MeetOnTrack 0.1.13 for Mac

MeetOnTrack 0.1.13

Technical beta for Apple Silicon with macOS 15+. Windows is a separate package.

• When a speech model is already downloaded, MeetOnTrack picks it for you.
  With several downloaded models it takes the newest one. You can still change
  the choice in Getting started.
• The live transcript starts in the middle of a call as soon as speech
  recognition becomes available. You no longer have to start a new recording.
• A red banner shows anything that blocks recording, the transcript or AI,
  with a button that fixes it or opens the right setting. When everything
  works, there is no banner.
• Error messages are plain language instead of technical text.
• "Start new meeting" works whenever audio is recording.
• Call audio recovers by itself within about 20 seconds when the capture of
  the other side goes quiet.
• Removed: the All/Me/Others filter, "Analyse now", the automatic analysis
  checkbox and the Questions/Risks/Decisions tabs.

Cloud AI receives the selected text context of the task it runs. Cloud speech
recognition receives the audio only when you choose a cloud engine.
Google/SMTP integrations remain optional and off for new installations. No
usage-statistics feature is included.

Update

Finish the call and quit Capture, then run:

curl -fsSL https://meetontrack.com/install.sh | bash

Versions 0.1.10 and 0.1.11 do not show this update in the app: the
project's GitHub address changed, and those versions reject release pages at
the new address. Run the command above once. Version 0.1.12 shows this update
in the app.

The installer verifies the pinned package checksum and preserves recordings and
settings. Automatic installation is deferred; version notifications are enabled
by default and can be disabled in Settings.

Verification and limits

Source: 6e809812c3fa30c5a01344a807c4f4b13c1cef4d.
ZIP SHA-256: 664895732cf40c0d28651d97334ddad5180b793bd31b766001684daa2c885868.
379 Swift tests in 49 suites: 375 passed; 3 retranscription tests need a
LocalVQE development library that is absent from a source checkout (the package
includes it), and 1 interface-language test reads the speech model downloaded
on the build Mac, whose engine name is the same in both languages.
361 web tests: 358 passed, 2 optional skips, 1 needs Pillow.
Exact archive: 2,823 file hashes, fresh install, repeat install preserving
synthetic settings and recordings, native ad-hoc signature, embedded
Python/server import passed.

This package is ad-hoc signed and not Apple-notarized. Retain normal macOS
security protections. Real AI and cloud speech calls were not exercised by the
release checks; run the Getting started checks after choosing. Full
clean-machine permission/audio acceptance and the device/reconnect matrix
remain follow-up checks. Speaker separation may need manual correction. Local
folder retrieval supports UTF-8 text, Markdown, CSV and source code, not general
PDF/Office ingestion.

MeetOnTrack 0.1.12 for Mac

MeetOnTrack 0.1.12

Technical beta for Apple Silicon with macOS 15+. Windows is a separate package.

• Choose the AI once. In Getting started, pick the tool, model, effort and
  speed one time, and every AI task uses that choice: live analysis and chat,
  live translation, screenshot relevance, summary rewrite, meeting summary and
  speaker names. If 0.1.11 had the same choice for every task, it is kept. A
  task the chosen model cannot do, for example screenshots with a model that
  does not accept images, is shown as unavailable with the reason, and the
  other tasks keep working.
• One speech recognition choice. Getting started has one speech card for the
  engine, the model and the language. The same model writes the live and the
  final transcript. The options come from what this Mac and your accounts
  offer, each model shows its download size, and nothing is preselected.
• If the installer already downloaded exactly one speech model, MeetOnTrack
  uses it with no question. With several models or none, choose one in
  Getting started.
• A Parakeet model that is already on this Mac, also one downloaded by another
  app such as FluidVoice, is reused instead of downloaded again. The other
  app's files are not changed.
• Speakers are separated live and again after the call. The speaker model is
  downloaded together with the local speech models. If it cannot be
  downloaded, transcripts still work and show the two sides of the call.
• Call audio is now labelled "Others" ("Остальные" in Russian). Your own
  microphone stays "You".
• A recording made before the speech choice is ready waits for it. No
  transcription attempt is used up, and the recording is transcribed as soon
  as the choice is ready.
• Fixed: live translations of a finished meeting are kept when the service
  restarts. The installer no longer stops with "Bootstrap failed: 5:
  Input/output error" when it restarts the service during an update.

Cloud AI receives the selected text context of the task it runs. Cloud speech
recognition receives the audio only when you choose a cloud engine.
Google/SMTP integrations remain optional and off for new installations. No
usage-statistics feature is included.

Update

Finish the call and quit Capture, then run:

curl -fsSL https://meetontrack.com/install.sh | bash

Versions 0.1.10 and 0.1.11 do not show this update in the app: the
project's GitHub address changed, and those versions reject release pages at
the new address. Run the command above once. Version 0.1.12 checks the new
address, so later versions are announced in the app again.

The installer verifies the pinned package checksum and preserves recordings and
settings. Automatic installation is deferred; version notifications are enabled
by default and can be disabled in Settings.

Verification and limits

Source: 84d5aaa1679bbeb75ba09c5f3f071e7e5e5622d2.
ZIP SHA-256: 1a6bf0a3f25734376c2a87db2430be3f36144d15e2bac31db74828ab515a0fea.
377 Swift tests in 49 suites: 374 passed; 3 retranscription tests need a
LocalVQE development library that is absent from a source checkout (the package
includes it). 343 web tests: 340 passed, 2 optional skips, 1 needs Pillow.
Exact archive: 2,820 file hashes, fresh install, repeat install preserving
synthetic settings and recordings, native ad-hoc signature, embedded
Python/server import passed.

This package is ad-hoc signed and not Apple-notarized. Retain normal macOS
security protections. Real AI and cloud speech calls were not exercised by the
release checks; run the Getting started checks after choosing. Full
clean-machine permission/audio acceptance and the device/reconnect matrix
remain follow-up checks. Speaker separation may need manual correction. Local
folder retrieval supports UTF-8 text, Markdown, CSV and source code, not general
PDF/Office ingestion.

MeetOnTrack 0.1.14 for Windows

MeetOnTrack 0.1.14 for Windows

Technical beta for Windows 11 24H2+, x64. Capture, .NET, Python and the web UI
are included; no compiler or SDK is required. Initial setup downloads local
speech models. AI answers are optional and use an AI tool you already have.

• Check AI is reliable for live analysis. Its test file used to sit in a
  temporary folder that the AI tool is not allowed to read, so the check could
  fail although AI worked.

Everything from 0.1.13 is included: no more "Access is denied", Codex in its
own Windows sandbox, the reliable image check, upgrades that never cut a
recording, the red banner and plain error messages.

The speech recognition choice from the Mac release is not available on Windows
yet (MEETONTRACK-69). Getting started says so, and the Windows recorder keeps
transcribing with its local speech models as before.

Finish the call and close Capture before installing or updating:

irm https://meetontrack.com/install.ps1 | iex

Version 0.1.10 and 0.1.11 do not show this update in the app: the project's
GitHub address changed, and those versions reject release pages at the new
address. Run the command above once. Version 0.1.12 and later check the new
address.

Recordings, model data and settings live outside the versioned app directory.
The installer verifies the package checksum. No automatic installer is enabled.

Build identity

Application source: 5aaaec8ec3e5bfad28929e4f0dc9262ec706a375.
Native source: 50e7448c880538ab96b57e093b4f7214a7a51de0
(recorder run 39649348946).
ZIP SHA-256: 8f49ad1eb30172b04e12e9f3e8ac447bdba6f6860797e77d2681b5fa9212e2bb.
The integrated package version is 0.1.14; its independently versioned native
component retains 0.6.6. The web update checker compares integrated versions.

Limits

Unsigned technical beta; keep SmartScreen and antivirus enabled. Real-device
audio, permissions, headphone reconnects and installed tray acceptance are not
claimed for this release. Test a short call and inspect both sides before
relying on a long recording. Local PDF/Office extraction and optional screen
OCR dependencies are not included. Speaker labels may need correction.

MeetOnTrack 0.1.13 for Windows

MeetOnTrack 0.1.13 for Windows

Technical beta for Windows 11 24H2+, x64. Capture, .NET, Python and the web UI
are included; no compiler or SDK is required. Initial setup downloads local
speech models. AI answers are optional and use an AI tool you already have.

• Fixed: AI help, the meeting archive and reports failed with "Access is
  denied" for some people. A folder locked that way is moved aside to
  archive.inaccessible-<time> or reports.inaccessible-<time>, nothing in it
  is read or deleted, and a new folder is made. No administrator rights are
  needed.
• Codex works for a normal user. MeetOnTrack uses the Windows sandbox you set
  up in Codex. If none is set up, Check AI says why Codex refused and what to
  do.
• The AI image check is reliable: it no longer fails because a model named
  the test colour "salmon" or "pink".
• Updating replaces the old version's web server. An old recorder that is
  recording keeps recording; it hands over to the new version once it is
  idle, so an update never cuts a recording. A recorder too old to hand over
  is replaced at the next sign-in. The old versioned sign-in entry is removed;
  the Startup shortcut starts MeetOnTrack.
• A red banner shows anything that blocks recording, the transcript or AI,
  with a button that fixes it or opens the right setting.
• Error messages are plain language instead of technical text.
• "Start new meeting" works whenever audio is recording.
• Removed: the All/Me/Others filter, "Analyse now", the automatic analysis
  checkbox and the Questions/Risks/Decisions tabs.
• The recorder keeps other programs' keys when it saves config.json.

The speech recognition choice from the Mac release is not available on Windows
yet (MEETONTRACK-69). Getting started says so, and the Windows recorder keeps
transcribing with its local speech models as before.

Finish the call and close Capture before installing or updating:

irm https://meetontrack.com/install.ps1 | iex

Version 0.1.10 and 0.1.11 do not show this update in the app: the project's
GitHub address changed, and those versions reject release pages at the new
address. Run the command above once. Version 0.1.12 and later check the new
address.

Recordings, model data and settings live outside the versioned app directory.
The installer verifies the package checksum. No automatic installer is enabled.

Build identity

Application source: 6e809812c3fa30c5a01344a807c4f4b13c1cef4d.
Native source: 50e7448c880538ab96b57e093b4f7214a7a51de0
(recorder run 39649348946).
ZIP SHA-256: ea182811b10137060cb8c8b331976a318e023efa2491c52b505c25b2cec6e7c9.
The integrated package version is 0.1.13; its independently versioned native
component retains 0.6.6. The web update checker compares integrated versions.

Limits

Unsigned technical beta; keep SmartScreen and antivirus enabled. Real-device
audio, permissions, headphone reconnects and installed tray acceptance are not
claimed for this release. Test a short call and inspect both sides before
relying on a long recording. Local PDF/Office extraction and optional screen
OCR dependencies are not included. Speaker labels may need correction.

MeetOnTrack 0.1.12 for Windows

MeetOnTrack 0.1.12 for Windows

Technical beta for Windows 11 24H2+, x64. Capture, .NET, Python and the web UI
are included; no compiler or SDK is required. Initial setup downloads local
speech models. AI answers are optional and use an AI tool you already have.

• Choose the AI once. In Getting started, pick the tool, model, effort and
  speed one time, and every AI task in the web assistant uses that choice: live
  analysis and chat, live translation, screenshot relevance and summary
  rewrite. If 0.1.11 had the same choice for every task, it is kept. A task
  the chosen model cannot do is shown as unavailable with the reason, and the
  other tasks keep working.
• Call audio is labelled "Others" ("Остальные" in Russian) in the web
  transcript. Your own microphone stays "You".
• Fixed: live translations of a finished meeting are kept when the service
  restarts.

The speech recognition choice from the Mac release is not available on Windows
yet. Getting started says so, and the Windows recorder keeps transcribing with
its local speech models as before. The native recorder component is unchanged.

Finish the call and close Capture before installing or updating:

irm https://meetontrack.com/install.ps1 | iex

Version 0.1.10 and 0.1.11 do not show this update in the app: the project's
GitHub address changed, and those versions reject release pages at the new
address. Run the command above once. Version 0.1.12 checks the new address.

Recordings, model data and settings live outside the versioned app directory.
The installer verifies the package checksum. No automatic installer is enabled.

Build identity

Application source: 775ac309bca9d116a7d51b5f8f52e2538716eb38.
Native source: e733138e2e61d74aac1d03b050624e558e30d155 (its windows/ tree is
identical to 0.1.10's b36ee3005e383c5c87ed4bceaf179ee8c95f8bb0).
ZIP SHA-256: 6618f0239155a6e5d46fb2e6a29c28ee48a809e9aa9edce6ad943ec90c648e0b.
The integrated package version is 0.1.12; its independently versioned native
component retains 0.6.6. The web update checker compares integrated versions.

The Mac 0.1.12 package was built from 84d5aaa. The Windows package adds one
Windows-only change on top: the web server no longer starts the recorder to
list speech options, because the Windows recorder does not have that command
yet and would open its window instead.

Exact package acceptance passed on a Windows runner: 1,740 manifest hashes,
fresh/repeat installation, recording preservation, native metadata, embedded
server self-test, PowerShell syntax and five installer fixtures. Hardware audio
is not tested on the hosted runner.

Limits

Unsigned technical beta; keep SmartScreen and antivirus enabled. Real-device
audio, permissions, headphone reconnects and installed tray acceptance are not
claimed for this release. Real AI calls through each provider were not
exercised by the release checks; run the Getting started check after choosing.
Test a short call and inspect both sides before relying on a long recording.
Local PDF/Office extraction and optional screen OCR dependencies are not
included. Speaker labels may need correction.

MeetOnTrack 0.1.11 for Mac

MeetOnTrack 0.1.11

Technical beta for Apple Silicon with macOS 15+. Windows is a separate package.

• AI tool, model, effort and speed are chosen per task: live analysis and chat,
  live translation, screenshot relevance, summary rewrite, meeting summary and
  speaker names. Options are discovered from the AI tools installed and signed
  in on this Mac (Codex, Claude Code, local Ollama) and from what your plan
  offers. Nothing is hardcoded or preselected. Antigravity is detected but not
  yet selectable, because it cannot be held to read-only, no-tools calls.
• After updating, open Getting started and choose a tool and model for each
  task, then run the check. Until a task has a choice, its AI step is skipped
  with a clear message. Recording and local transcription work without AI.
• Native summaries and speaker names now use the same chosen AI and the shared
  settings file; a legacy "no summary backend" setting migrates once to
  summaries off.
• Fixed: an old "Install Codex" error saved with a meeting no longer reappears
  after restart once the AI setup has been repaired.

Speech-to-text engine selection is unchanged in this release. Cloud AI receives
the selected text context of the task it runs. Google/SMTP integrations remain
optional and off for new installations. No usage-statistics feature is included.

Update

Finish the call and quit Capture, then run:

curl -fsSL https://meetontrack.com/install.sh | bash

The installer verifies the pinned package checksum and preserves recordings and
settings. Automatic installation is deferred; version notifications are enabled
by default and can be disabled in Settings.

Verification and limits

Source: effb7a2b65be3cd3e02cc1bfd2be0864340d5ac2.
ZIP SHA-256: 59b5b1c843fc6da5075deb6fe8c7891e5aa826bfaba167a8888e5fe9f62b80db.
398 Swift tests in 48 suites: 395 passed; 3 retranscription tests need a
LocalVQE development library that is absent from a source checkout (the package
includes it). 314 web tests: 311 passed, 2 optional skips, 1 needs Pillow.
Exact archive: 2,814 file hashes, fresh install, repeat install preserving
synthetic settings and recordings, native ad-hoc signature, embedded
Python/server import passed.

This package is ad-hoc signed and not Apple-notarized. Retain normal macOS
security protections. Real AI calls through each provider were not exercised by
the release checks; run the Getting started check after choosing. Full
clean-machine permission/audio acceptance and the device/reconnect matrix
remain follow-up checks. Speaker separation may need manual correction. Local
folder retrieval supports UTF-8 text, Markdown, CSV and source code, not general
PDF/Office ingestion.

MeetOnTrack 0.1.11 for Windows

MeetOnTrack 0.1.11 for Windows

Technical beta for Windows 11 24H2+, x64. Capture, .NET, Python and the web UI
are included; no compiler or SDK is required. Initial setup downloads local
speech models. AI answers are optional and use an AI tool you already have.

• AI tool, model, effort and speed are chosen per task in the web assistant
  (live analysis and chat, live translation, screenshot relevance, summary
  rewrite). The native recorder component is unchanged. Options are discovered from the AI tools installed and signed in on this computer (Codex,
  Claude Code, local Ollama) and from what your plan offers. Nothing is
  hardcoded or preselected. Antigravity is detected but not yet selectable,
  because it cannot be held to read-only, no-tools calls.
• After updating, open Getting started and choose a tool and model for each
  task, then run the check. Until a task has a choice, its AI step is skipped
  with a clear message. Recording and local transcription work without AI.
• Fixed: an old "Install Codex" error saved with a meeting no longer reappears
  after restart once the AI setup has been repaired.

Speech-to-text engine selection is unchanged in this release.

Finish the call and close Capture before installing or updating:

irm https://meetontrack.com/install.ps1 | iex

Recordings, model data and settings live outside the versioned app directory.
The installer verifies the package checksum. No automatic installer is enabled.

Build identity

Application source: effb7a2b65be3cd3e02cc1bfd2be0864340d5ac2.
Native source: e733138e2e61d74aac1d03b050624e558e30d155 (its windows/ tree is
identical to 0.1.10's b36ee3005e383c5c87ed4bceaf179ee8c95f8bb0).
ZIP SHA-256: bc60f21e3c5089781fb1f439fb9cf1a3ce89694e6db0b4acdac1c5e96909fdf8.
The integrated package version is 0.1.11; its independently versioned native
component retains 0.6.6. The web update checker compares integrated versions.

The native Windows build and core/live regressions passed in GitHub Actions.
Exact package acceptance passed on a Windows runner: 1,734 manifest hashes,
fresh/repeat installation, recording preservation, native metadata, embedded
server self-test, PowerShell syntax and five installer fixtures. Hardware audio is
not tested on the hosted runner.

Limits

Unsigned technical beta; keep SmartScreen and antivirus enabled. Real-device
audio, permissions, headphone reconnects and installed tray acceptance are not
claimed for this release. Real AI calls through each provider were not
exercised by the release checks; run the Getting started check after choosing.
Test a short call and inspect both sides before relying on a long recording.
Local PDF/Office extraction and optional screen OCR dependencies are not
included. Speaker labels may need correction.

MeetOnTrack 0.1.10 for Mac

MeetOnTrack 0.1.10

Technical beta for Apple Silicon with macOS 15+. Windows is a separate package.

• MeetOnTrack identity and M icon throughout native setup, Dock and menu bar.
• First-run language selection, clearer recording/model setup and optional AI.
• Meeting library, timed notes, meaningful questions and a separate meeting
  boundary during a continuous recording, with confirmation.
• Connected local folders without Git; optional read-only Google context.
• Optional report delivery: SMTP or a user's own Google OAuth JSON, independent
  email/Drive consent, visible status and explicit retry controls.
• New-version notification, release notes and a copyable update command.
• Retained recording recovery and reduced repeated background reads/work.

The Mac package includes native Capture, Python 3.12.13, web UI and compiled
helpers. No Python, Swift compiler or Apple SDK is required on the user's Mac.
Codex is optional for AI answers; capture and local transcription work without it.
Google Chrome is needed for PDF export. Cloud AI receives the selected text
context. Google/SMTP integrations are optional and off for new installations.
No usage-statistics feature is included.

Update

Finish the call and quit Capture, then run:

curl -fsSL https://meetontrack.com/install.sh | bash

The installer verifies the pinned package checksum and preserves recordings and
settings. Automatic installation is deferred; version notifications are enabled
by default and can be disabled in Settings.

Verification and limits

Source: ae27f2b8a323a1b7d20db3a6ef23e1345cb4249b.
ZIP SHA-256: 2df0822063ccbfd2ba14d8463664a3d5f2ae83ebfd31877cff47d89558c32593.
369 Swift tests passed. 260 web tests passed with 2 optional skips. Exact archive:
2,808 file hashes, fresh install, repeat install preserving synthetic settings
and recordings, native ad-hoc signature, embedded Python/server import passed.
The current icon was installed and visually confirmed on the owner's Mac.

This package is ad-hoc signed and not Apple-notarized. Retain normal macOS
security protections. Full clean-machine permission/audio acceptance and the
device/reconnect matrix remain follow-up checks. Speaker separation may need
manual correction. Local folder retrieval supports UTF-8 text, Markdown, CSV and
source code, not general PDF/Office ingestion. Google account consent and real
SMTP acceptance depend on the user's chosen provider configuration.

MeetOnTrack 0.1.10 for Windows

MeetOnTrack 0.1.10 for Windows

Technical beta for Windows 11 24H2+, x64. Capture, .NET, Python and the web UI
are included; no compiler or SDK is required. Initial setup downloads local
speech models. AI answers optionally use your own Codex account and its limits.

• MeetOnTrack native identity and M icon, including tray and installed executable.
• Clear first-run setup, optional AI readiness check and connected local folders.
• Meeting history, timed notes and meeting analysis in the local assistant.
• Optional SMTP or your own Google Desktop OAuth JSON for report delivery.
• Optional Calendar/Gmail/Drive context, controlled separately from delivery.
• New-version indicator, release notes and a copyable manual update command.
• No usage-statistics control or telemetry implementation.

Finish the call and close Capture before installing or updating:

irm https://meetontrack.com/install.ps1 | iex

Recordings, model data and settings live outside the versioned app directory.
The installer verifies the package checksum. No automatic installer is enabled.

Build identity

Application source: ae27f2b8a323a1b7d20db3a6ef23e1345cb4249b.
Native source: b36ee3005e383c5c87ed4bceaf179ee8c95f8bb0.
ZIP SHA-256: 3126a3fc819bb80d1ddd762ce14310d42bf99ae3b974983f0872fdcff39026c9.
The integrated package version is 0.1.10; its independently versioned native
component retains 0.6.6. The web update checker compares integrated versions.

The native Windows build and core/live regressions passed in GitHub Actions.
Exact package acceptance passed on a Windows runner: 1,728 manifest hashes,
fresh/repeat installation, recording preservation, native metadata, embedded
server self-test, PowerShell syntax and five installer fixtures. Hardware audio
is not tested on the hosted runner.

Limits

Unsigned technical beta; keep SmartScreen and antivirus enabled. The owner's
physical Windows host was offline during this release, so real-device audio,
permissions, headphone reconnects and installed tray acceptance are not claimed.
Test a short call and inspect both sides before relying on a long recording.
Local PDF/Office extraction and optional screen OCR dependencies are not included.
Speaker labels may need correction. Provider consent and SMTP configuration
must be checked using the user's chosen account.