What
changed.

New

• Remember to stop your meeting recording: When Tyll suspects your meeting has ended, “Meeting over?” appears in the recording controls. You can choose to keep recording or stop. Recording continues until you decide.

Improved

• More room to assign speakers: The speaker list makes better use of the available space. You can scroll through longer lists while keeping the controls to finish speaker assignment within reach.

Fixed

• Dictate in full while reviewing speakers: While a meeting’s speaker review was open, dictation could insert no text or only part of what you said. Your spoken text is now inserted in full in this view too.

• Start dictation more promptly: Once preparation completes, the dictation controls now appear directly, without an extra fade-in delay.

New

• Read your transcript without waiting for the summary: Once your conversation has been transcribed, you can open and search the text and listen to the recording. Your summary continues in the meantime. Even if you cancel it, your transcript stays available.
• Get notified when speakers are ready to confirm: Tyll can now notify you when you can check who spoke in a meeting, even while you’re working in another app. One click takes you straight to the meeting.

Improved

• Download large updates less often: Even if you’ve skipped several versions, Tyll can now more often download just what has changed. This saves download time and data.
• Find meetings and parts of a conversation more easily: The person filter now finds all meetings with the person you’re looking for. Within a meeting, you can search the text and filter by individual speakers.
• Recognise familiar voices better: Tyll is now better at recognising voices you’ve already assigned to a person.
• Correct mistaken speaker assignments: When you remove or change an assignment, Tyll also removes the voice sample saved with it. The mistaken assignment is no longer used to help recognise voices.
• See whether your agent is ready: The home screen and settings make it easier to see whether you still need to set something up.

New

• Get started with guidance: The new introduction walks you through your dictation key, required permissions, a test dictation, and agent setup in four steps. It also appears once for existing installations and can be reopened later through “Help › Show Introduction”.

Improved

• Choose your agent more clearly: Settings now show the available agents and their status in one selection list, including “No agent”. The relevant dictation option appears directly below; meeting access is configured separately and independently.
• Cleaner version history: An entry for a version that was never published has been removed from the changelog.

Improved

• See all changes since your version: The update window now includes changes from every newer published version up to the offered update, grouped by version. This keeps you informed even when you skip several releases.

Fixed

• Download app updates over wired connections: A wired internet connection is no longer blocked by the previous Wi-Fi requirement. Tyll downloads updates over available, unmetered connections without Low Data Mode. The update settings now describe these requirements consistently.

New

• Use meeting data in Claude Desktop chats: The Claude switch under Settings → Agents sets up meeting access for Claude Code and, if installed, Claude Desktop together. It shows “on” when both configurations point to this Tyll installation.
• Quit Claude Desktop before changing access: If Claude Desktop is still running, Tyll changes nothing and shows a notice. Quit Claude Desktop, turn meeting access on or off, then start Claude Desktop again. This prevents Claude from overwriting the changed configuration while running.

What runs better

• Meeting recording without a custom audio driver: Tyll captures microphone and system audio natively and cleans its own microphone track with FluidAudio on your Mac. Teams, Meet, and other calling apps keep their own audio processing. Start, stop, pause, resume, dictation during a meeting recording, and background transcription through MCP remain available.
• Switch microphone and speaker independently: The new “Tyll Microphone” and “Tyll Speaker” devices follow their respective system defaults independently, even when Tyll is closed. The bundled device service starts at login, and Tyll checks it whenever you open the app. Settings show its version and operating status.
• Recording prompts follow the Tyll devices: When both Tyll microphone and speaker are actively in use, Tyll shows a recording prompt, regardless of which app uses them. Recording still starts only after you click.

When updating

• The old audio driver and Audio Host are removed: The package update removes the former Tyll audio devices and background service. No Mac restart or logout is required; audio may briefly be interrupted during the change. The new devices are available after Tyll starts. Select them again in your calling app if needed. Previous recordings are preserved.
• Meeting recordings include all system audio: This includes other apps and notifications. Manual recording remains available regardless of your calling app’s device selection. Automatic prompts require active use of both new Tyll devices.

Fixed

• Device changes and recording completion: The new capture path preserves audio section boundaries when switching devices, pausing, and resuming. Missing audio data is reported as an error, while captured original data remains available.
• Use Codex after the app update: Tyll finds the Codex CLI bundled with the ChatGPT app in its new location. An unavailable saved source no longer prevents selection of an available source, and contextual dictation supports the updated Codex connection.

New

• Ask questions while the meeting is still running: A connected agent can read the preliminary live transcript through MCP during recording and until the meeting has finished processing. Tyll transcribes every meeting recording locally in the background. The transcript trails by a few seconds and may contain recognition errors. “Lokal” denotes the microphone track and “Remote” the system audio track. Speaker names arrive with the finished transcript. There is no live transcript view in the app.
• Data flow when asking questions: Recording and live transcription stay on your Mac. When you ask a connected agent, it can read the preliminary transcript and process it through its provider. The same boundaries of your agent account apply as when accessing finished meetings.

What runs better

• Completed live sections are reused after recording stops: The first processing run reuses matching, fully computed live sections. Changed inputs and subsequent processing attempts cause the relevant sections to be computed again.
• Meetings are always detected automatically: The meeting detection switch and app exclusion list have been removed. After updating, Tyll shows recording prompts even if detection was previously disabled or an app excluded. You can still dismiss a prompt for the individual meeting.
• Shorter labels and relevant participant context: Meeting labels stay more concise. Confirmed participants and their roles are passed directly to the summary without an additional participant lookup.
• Library search is ready to type: ⌘F opens the clearly visible search field and places keyboard focus there immediately.
• Recording sources remain until explicitly deleted: Tyll no longer automatically rewrites or deletes the original recording sections after processing. This may use more storage than the previous compressed archive.

Fixed

• Reprocessing preserves every recording section: After a processing failure and retry, the complete track list remains available, including earlier sections.
• A metadata failure does not discard a finished transcript: If summary, title, or label generation fails, the successfully generated transcript is still saved. Existing metadata and notes remain, and Tyll displays the error with the existing recovery actions.
• Speaker prompts require a suitable voice sample: Tyll hides the prompt when no sufficiently continuous sample is available for confirmation.
• The former live draft path has been removed: The new transcript uses the shared transcription path without retaining the superseded draft path alongside it.

New

• Agents can read, extend, and clean up the glossary: When a technical term comes up while working with an agent, it can be added as a preferred spelling or removed again right there. The app picks up changes the next time it becomes active.
• Preferred spellings feed into meeting summaries: The active glossary entries are available to the summary as a reference, so company and technical terms appear correctly spelled. Facts are still established by the transcript alone.
• Labels can be created in Settings without a meeting: The labels page creates a new label directly instead of only allowing it to be assigned on an existing meeting.

What runs better

• Automatic meeting labels stay limited to distinct main topics: Tyll no longer assigns extra labels for side aspects that a broader topic in the same conversation already covers.
• Settings for agent features are easier to scan: The section presents the available features more compactly, and the library label filter is labeled more clearly.
• The glossary is kept in one place: Preferred spellings live in the database rather than in settings data, so the app and agents see the same state. Existing entries are preserved through the migration.

Fixed

• An open glossary edit is no longer lost: Touching an entry without changing its value now keeps the edit instead of reverting it.
• Every route into a settings section leads to the same place: Selecting a settings page runs through one shared path and stays consistent between the list, the menu, and app startup.

New

• Meeting labels organize conversations by customer, project, and topic: Labels can be created and assigned directly on a meeting, used to filter the entire library across pages, and managed in Settings.
• New meetings receive suitable labels automatically after summarization: Tyll uses up to three existing or new labels and marks automatic assignments with a dashed border.
• Meeting labels are also available to agents: The meeting tools can list, add, remove, and use labels for search.

What runs better

• Meeting summaries are easier to scan: Discussed topics appear as short, self-contained facts. Long meetings get more processing time for summaries.
• Meeting topic search computes full-text matches more efficiently: Results, filters, sorting, and snippets remain unchanged.
• Speaker groups are consolidated from more reliable voice evidence: Tyll uses speech-active regions, guards against inflating mixed groups, and favors the cluster that actually carries speech. Unassigned groups are consistently named “Unknown” after review.

Fixed

• Local meeting recordings remain active until stopped: An unused audio path no longer ends a healthy app recording after a few seconds.
• All visible start controls show the same recording readiness: The app menu, toolbar, and status bar menu are now enabled or disabled together for the same state.
• Agents receive only fully finalized meetings: Preliminary transcripts stay unavailable through the meeting tools until speaker assignment and corrections are complete.
• Dictations without recognized speech end quietly: No error or empty history entry appears, and the dictation HUD stays fully hidden while idle.
• Meeting labels remain current after edits and reloads: Filters, chips, renames, and deletions show the latest stored state.
• Stale waveforms no longer overwrite a newer selection: After switching sources, only the result for the current audio file is displayed.

What runs better

• Audio device changes are applied reliably: Tyll detects changes to input and output devices while running and re-establishes capture on the physical audio route that is actually active.
• Agents can be signed in again directly where a summary failed: When Codex or Claude is signed out, Tyll shows the cause and the relevant sign-in action both in the app and on the affected meeting.
• Speakers are identified more completely in longer meetings: Audio windows are read from the correct segment file. This preserves more usable speaker samples in recordings longer than five minutes.
• Short duplicate transcript fragments are removed more conservatively: A local single-word segment is removed only when its complete time window is already covered continuously by the remote track.

Fixed

• A corrected contact choice in speaker review is preserved: When previously merged speaker groups are assigned to different contacts, Tyll separates them again and saves both assignments.

New

• Meeting knowledge is clearly structured for connected agents: Codex and Claude agents can search meetings, read transcripts directly, retrieve the summary stored in Tyll, and update its Markdown. Transcripts remain read-only.

What runs better

• Glossary terms only replace acoustically similar recognitions: The glossary still helps with specialist terms while interfering less with clearly dissimilar spoken words.
• Speaker identification only presents verifiable voices: Candidates without a playable sample are no longer offered for naming or contact assignment. They remain anonymous speakers in the final transcript.

Fixed

• New folders can once again be created and used in the save dialog.

What runs better

• Automatic titles follow the app language: Meetings, dictations, and imported recordings are named in the selected language across the library, quick search, status bar, and detail views. Custom titles remain unchanged.
• Default speaker labels are translated consistently: “You”, “Speaker”, and numbered speakers now use English in every view. Linked contacts and names you entered yourself remain intact.
• License dates use the right format: Transfer-related dates now use the same language and formatting in the license window and settings.

Fixed

• Automatically generated English names could remain visible after switching languages.
• Saving an unchanged speaker row could persist a translated default label as a real name.

What runs better

• Formatting is preserved in Apple Notes: Headings, paragraphs and lists from a dictation are no longer flattened into plain text when inserted.
• Fewer lost words with offset audio tracks: Tyll now checks speech activity in the correct section of the source file. Utterances and their speakers are no longer removed because the wrong audio window was measured.
• More complete sentence beginnings: When a recognized word begins at the artificial separator between two speech sections, it is assigned to the following real audio section and retained in the transcript.
• More precise glossary corrections: Similar terms are replaced more conservatively. For example, “Message” remains intact instead of being incorrectly changed to “Merge”.

Fixed

• Individual opening words disappeared between separate speech sections even though they had already been recognized correctly.
• Wrong audio windows removed speakers and text. Activity is now checked in the correct section of the source file.
• Apple Notes lost the original text formatting during dictation.

What’s new

• Codex and Claude as local agents: Tyll detects installed agent runtimes and connects to them. Meeting summaries run through them on your Mac.
• Open meetings to agents: Access to meetings and transcripts can be turned on and off per agent. The access is read only.
• Context dictation: On request, dictation takes the focused window into account so terms and names land better.
• Automatic meeting detection: Tyll detects running meetings independently of the app you use and asks whether to record. Individual apps can be excluded.
• One pill for recording and dictation: Both now share a single indicator that you can move freely.
• Summary and transcript separated: Meeting detail now has two tabs.
• Automatic export: Finalized meetings can land as Markdown in a folder of your choice.
• Diagnostic report: A report can be saved and sent to support. It includes the most recent event history.
• Audio samples in speaker review: Instead of a text preview there are now assembled audio samples, even for very short fragments.

What is different

• Summaries run through local agents: The direct connection to OpenAI, Anthropic, Ollama and LM Studio is gone. If you had one configured, no provider will be selected after the update and you pick Codex or Claude instead. Stored keys remain untouched in the keychain.
• Settings consolidated: Export, agents and summaries now live together under Integrations.
• Transcript enhancement removed: The switch between the original text and an enhanced version is gone.

What runs better

• Tyll speaks your language: The app starts in your system language. Texts that used to stay English now follow the language you set, among them license notices, audio status and the messages you see when a recording is finalized.
• Fewer lost words: Noticeably fewer words are lost at segment and batch boundaries, duplicates and language switches occur less often, and sentence boundaries land more cleanly.
• More accurate speaker detection: Clusters are merged more reliably, speaker changes are placed more precisely, and short interjections are no longer lost.
• Finished sooner: Transcription runs alongside speaker detection, and spare compute is used better.
• Calmer library and playback: The library loads more as you go instead of fetching everything at once. Playing long recordings needs far less memory.
• Smaller updates: An update now downloads only what changed instead of the whole application.

Fixed

• The transcript was lost when no speakers were detected. Such recordings now keep their transcript, with segments shown as unassigned.
• Empty transcript for very short recordings and immediate speech.
• Audio disappeared silently at segment boundaries when the recording rotated.
• Meeting detection spoke up during dictation and video playback when no meeting was running.
• The audio driver reported ready without ever being checked. If it is missing, Tyll now says so and names the way back.
• Merging a copied recording could lose the audio of the source recording.
• Transcript text and contact names appeared in the macOS system log.

What’s new

• People filtering for meetings: The meeting library can now be filtered by linked contacts.
• Participants in meeting rows: Linked people are shown directly in the library so relevant meetings are easier to spot.
• Contact links are repaired: When a speaker name exactly matches one unique contact, Tyll restores the missing link.
• Contact assignments survive finalization: Queued meeting finalization now keeps speaker-contact assignments intact.
• More robust demo start: An unreadable local demo marker no longer blocks starting the 14-day demo.

What’s new

• LM Studio as a local enhancement provider: Tyll can now run transcript enhancement through a local LM Studio model.
• Improved provider management: Settings now handle providers, models, base URLs, and credentials more clearly.
• Stricter validation before saving: Incomplete provider URLs, model names, or credentials are caught earlier.
• Structured results are checked: Tyll validates enhancement responses before summaries or transcript updates are saved.
• Apple Foundation removed: The Apple Foundation provider is no longer part of the enhancement provider list.