Using Takku

A meeting

1. Start from the menu bar or with ⌃⌥⌘R, from any app. While Takku records, its icon is a red dot and the menu shows the elapsed time. with Suggest recording on, in Settings, General, a notification offers to start when a call begins
2. The meeting note opens in your editor (Obsidian if installed, otherwise your Markdown editor). Write your own notes under Personal notes: Takku never changes them, and uses them to steer the summary.
3. Press ⌃⌥⌘M when something matters. That passage gets a ⭐ in the transcript, and the summary gives it priority.
4. Press ⌃⌥⌘S to keep what’s on screen, such as a slide or a diagram. It marks the moment too, and embeds the picture under the matching point of the summary and in the transcript. the model never sees the picture: Takku places it by the times the summary cites
5. Stop from the menu or with ⌃⌥⌘R again. Takku finishes the transcript, writes the summary at the top of the note with the time of each point, links the transcript and gives the note a short title. A notification tells you when it’s done.

If Takku hears nothing from the microphone or from the call for 2 minutes, the menu says so. If it has heard nothing since the start, a notification tells you which permission to check.

A recording made elsewhere

A voice memo from your iPhone, a lecture, a meeting in person, the recording of a call from Zoom or Teams: choose Import audio… in the Takku menu and pick one or more files, audio or video (m4a, mp3, wav, mp4, mov…). Each becomes a meeting like the others, with transcript, speakers and summary.

The note takes the date and time the file was recorded and, when the file’s metadata says so, the calendar event of that moment. Takku reads the file and leaves your original where it is.

While a live call is transcribed as you talk, an imported file is transcribed all at once: allow about four minutes per hour of audio.

Another summary

You choose the template and the summary profile in the Takku menu. In Recent meetings, each meeting offers two ways to run it again with the template and profile currently selected:

• Regenerate Summary writes a new summary from the current transcript in your notes folder, keeping any edits you made to the text.
• Reprocess Recording… rebuilds the transcript from the audio (while the audio is kept), applying your latest vocabulary and speaker names, then writes a new summary.

Choose Transcript as the profile to get only the transcript, with no summary.

Templates

A template is a Markdown file in ~/.takku/templates/: its text tells the model what to write and how, in your own words. Create one from Settings, Notes, New Template, then open it from its ⋯ menu with Edit With, in any text editor. The same menu makes it the default for new meetings.

Takku writes the summary in the language of the meeting. To fix its language, add summary_language: en (or it, fr…) to the template’s frontmatter.

Add full_text: true to get the whole meeting written out instead of summarised: cleaned up, with everything people said, for example the full text of a lecture to study from. Takku writes it ten minutes at a time and joins the parts, so the end of a long recording is as detailed as the beginning.

Two templates for lectures come with Takku, next to Notes: Lecture, study notes topic by topic, and Full text, the full text of the lecture. If your templates folder is older than them, copy them into ~/.takku/templates/.

Vocabulary

Names, acronyms and technical words that speech recognition gets wrong go in ~/.takku/vocabulary.md, which you open from Settings, Notes, Edit Vocabulary With. One term per line: the right spelling, then after = what is heard instead, then after | a short description. The last two are optional.

- Kubernetes = cubernetes, kubernetis
- dbt = di bi ti | data transformation tool
- Holacracy

Takku replaces the variants in the transcript as whole words, so don’t list one that is also an ordinary word. Every term goes to the model, which spells it right in the summary.

Speaker names

The transcript tells voices apart as Speaker 1, Speaker 2 and so on, and the summary names them when the call makes it clear who they are. When it can’t, open Recent meetings and choose Name speakers…: press ▶ to hear a few seconds of each voice, pick a name from the calendar participants or type one, and press Save and Regenerate.

With Recognise voices turned on in Settings, Transcription (off by default), Takku remembers the voices you name on your Mac and suggests their names in future meetings. A voice is biometric data: turn it on only for people who agreed.

You can also write the names in the note’s frontmatter and choose Regenerate Summary:

speakers:
  - Speaker 1 = Mario Rossi
  - Speaker 3 = Anna Bianchi

Calendar

Turn on Use the calendar in Settings, General, and allow access. When a meeting starts during an event, the note takes the event’s name and lists who was invited, which helps the summary name the speakers. If the list is wrong, correct it in the note: it counts the next time you regenerate the summary.

Hook

A hook takes each meeting somewhere else once Takku is done with it: Notion, Confluence, a git commit of your notes folder, anything a script can do. Put an executable at ~/.takku/hooks/after-processing: when a meeting is ready, and again after every regenerated summary or reprocessed recording, Takku runs it with the paths of the meeting note and of the transcript.

How to write one, and a ready-made hook that copies every meeting to Notion: Hooks.

Your meetings in an AI assistant

Ask Claude, or another assistant that supports MCP, about your meetings: what did we decide about the release with Mario last week? It searches your notes and transcripts and answers from them. MCP is the standard way assistants read from other apps. Takku doesn’t need to be open.

In Claude Code, run once:

claude mcp add --scope user takku -- /Applications/Takku.app/Contents/MacOS/Takku mcp

In Claude Desktop, open Settings, Developer, Edit Config, add takku to mcpServers and restart Claude:

{ "mcpServers": { "takku": { "command": "/Applications/Takku.app/Contents/MacOS/Takku", "args": ["mcp"] } } }

What the assistant reads goes to its model, as anything else you give it: a meeting with a customer is the customer’s data.