What's new in Audicap 0.12.0

Audicap 0.12.0 (October 2026) brings several summaries in one go, access for AI assistants such as Claude, three new languages and cutting for finished recordings.

Several templates in one go

A meeting often needs more than one view: minutes for the archive, a task list for the team, perhaps a sentiment analysis. Now you choose all templates at once.

How it works

1. Open a recording and expand AI Summary on the right.
2. Click Template and tick every template you need. The search at the top of the list filters by name.
3. Click Start Summarization.

Each template creates its own summary in its own tab. You see the tab whose text arrives first right away. The others appear next to it without pulling you away.

At the same time or one after the other?

Provider	Order
Cloud (Anthropic, OpenAI, Gemini, OpenRouter, your own endpoint on the network)	all templates at the same time
Local (built-in model, Ollama, OpenAI-compatible server on this computer such as LM Studio)	one after the other, so the templates do not share memory

Waiting templates show “Waits for the previous template to finish” in the status bar. Cancel stops all open runs; whatever is finished stays. If a template fails, the message names it and the others keep running.

Automatic too: If the automatic summary is switched on under Settings → AI Summary, Audicap creates all templates selected there after each processing run.

Good to know

The included template “Action Items Only” is now called Action Items. Your own template that has the name of an included one is marked “(own)”.

Summary language per recording

The summary usually follows the interface language. For a conversation with international customers you can now choose a different language per recording, such as English minutes while Audicap runs in German.

How it works

1. Open the recording and expand AI Summary.
2. Choose the language under Summary language. Same as interface stores no fixed choice and follows a later language change.
3. Start the summary.

The chat about this recording answers in the chosen language too.

Why some templates keep their language: The included templates exist in all five languages. Templates you created or edited exist only in the language you wrote them in. A yellow note below the field names them.

New: French, Spanish and Chinese

Audicap now comes in five languages: German, English, French, Spanish and Chinese (Simplified). The interface, menus, dialogs, notifications, summaries and all included templates are translated. On first start Audicap uses the language of your system. You can change it under Settings → General → App and results language.

Tip

The new languages are machine-translated and reviewed. If a wording looks off, tell us via Support in the app.

Access for AI assistants (MCP)

With the Model Context Protocol (MCP), AI assistants on your computer, such as Claude Desktop or Claude Code, can read and search your recordings, transcripts and summaries. So you can ask: “What did we decide about pricing in the last planning meeting?”

Setup

1. Open Settings → MCP and switch on Allow access for other programs. This needs an active subscription or the trial.
2. For Claude Desktop: copy the configuration shown, add it to claude_desktop_config.json under mcpServers and restart Claude Desktop. Audicap must be running.
3. For Claude Code: copy the command shown and run it in the terminal.

Point to exactly what you mean: Copy for MCP hands the assistant an exact reference. You find it in the menu of a recording, as a button in the bar when several recordings are selected, and as Copy recordings for MCP in the menu of a space. Paste the reference into the chat with the assistant.

What the assistant may read: recordings (title, date, duration, participants, space, tags), transcripts with speakers and times, all summaries and a search across them. It also finds recordings by space, tags and date range. Notes and voice prints stay in Audicap. The assistant cannot change or delete anything.

Security

• Off by default.
• Reachable only from this computer, never from the network.
• Every access needs a key. Create new key locks out programs that use the old one. Claude Desktop gets the current key through Audicap itself.

Caution

Whatever a connected assistant reads goes to its provider. Only switch access on if that is fine for your recordings.

Cut a range out of a recording

Small talk at the start, a break in the middle, tidying up at the end: you can now remove such parts from a finished recording.

How it works

1. Open the recording and click the scissors in the player (Cut out a range).
2. Drag across the part of the waveform you want gone. Dragging again replaces the selection.
3. The line below shows where the cut snaps to sentence boundaries, what is removed and how long the recording will be.
4. Click Delete and confirm.

For several parts, repeat the cut. The transcript is adjusted and your corrections stay. Audicap creates the summaries anew; if one cannot be created again, a message names it. The original file goes to the trash, nothing is deleted for good.

Question when a call ends

If you started a recording by hand and the detected call (Teams, Zoom, Meet and others) ends, Audicap asks after 30 seconds: End the recording too? with End and Keep recording.

• Without a click, recording continues. Audicap never stops on its own.
• If the same call comes back within the 30 seconds, the question is dropped.
• Recordings from the calendar and from automatic detection behave as before.

Organizing: tags and spaces

• Tags without case variants: Tags no longer distinguish upper and lower case. They are stored in lower case; the update merged variants such as “Customer” and “customer”.
• Child spaces when moving: Move to… shows child spaces as a second level.

Further improvements

• Local by default: New installations summarize with the built-in model. Setup offers the model download right away; without a model a link leads to the right setting.
• Smaller Windows installer: Audicap now takes about 880 MB instead of 1.5 GB. NVIDIA graphics cards run via Vulkan.
• One-time downloads after the update: 11 MB for Parakeet German on the Mac (downloads by itself) and the speaker detection rebuilt with security updates (about 235 MB, under Settings → Transcription with Download components).
• Clean characters: Whisper transcripts no longer contain “�” when a character was split between two word pieces. This mainly affected Chinese.
• No profile pictures from Gravatar: Audicap no longer fetches profile pictures from Gravatar, and the switch for it is gone. You set profile pictures yourself; existing ones stay.
• Model catalog: Qwen3 4B is gone; it scored lowest in all five languages. An existing download can be deleted in model management.
• Tidying up: Deleting a recording now also moves its metadata to the trash.