-
Thursday 22 October
Recorded from the first word.
Teams, Zoom, Meet, FaceTime or a meeting in your browser: turn on auto-start, and pepito records the call without you thinking about it, welcome and agenda included. A notification sound starts nothing.
No bot joins the call, no browser extension, no screen recording.
-
Recording
Notes, if you want them.
Jot down what matters during the call. Your notes go into the minutes, with the calendar event you pick: its title, participants and location.
-
After the call
Transcribed on the Mac. Written up by your AI.
When the meeting ends, the transcript and the minutes follow, in your language, written by the AI you choose and shaped by your own instructions.
-
Wednesday 21 October, in the room
Meeting in person? It starts on its own too.
Workshops, one-to-ones, a meeting at a client's office: turn on the Conversation trigger, and pepito records when people start talking, first words included. You stay in the discussion, and you leave with the same minutes as after a call.
Off by default. pepito listens on your Mac: it sends no audio and writes nothing to disk before a recording starts, and macOS's orange dot stays on while it listens. Recording the people around you may require their consent: check that you use it in line with local regulations.
Your meetings, remembered.
pepito records your calls and meetings on your Mac, from the first word, and turns each one into a transcript and minutes written by the AI you choose.
Meeting after meeting, it keeps their memory on your Mac, in plain files. Turn on the MCP server, and your AI tool uses it to prepare your next meeting, recap a week or a project, tell you where things stand, and list decisions and actions.
Free. macOS 26 or later. In English and French.
Records calls from
- Teams
- Zoom
- Google Meet
- FaceTime
- iPhone calls
- and more
Works with
- ChatGPT
- Claude
- Gemini
- Ollama
- Your company's LLM
Before the next meeting, ask the last ones.
With the MCP server turned on, your AI tool reads your meetings and answers from them. A few things to ask it:
Prepare a meeting
“Prepare my call with Northwind: what we agreed last time, what is still open.”
Recap
“Recap this week's meetings in ten lines.”
Where things stand
“Where does the Avenir project stand since September?”
Actions
“What did I promise in my meetings this week?”
Decisions
“What did we decide on the budget, and when?”
Every meeting, kept and in order.
Every meeting lands in the Recordings window: minutes, transcript and notes together, by day, or week by week in the calendar.
Clients, projects, types
clientacme projectavenir decision
File each meeting under its client, its project and its type, with a color and a description for every tag.
Tagged for you
Turn it on, and the AI that writes your minutes picks among your tags, guided by their descriptions.
Find
Any meeting by title, tag or participant, or week by week in the calendar.
Write the minutes again. The old ones stay.
Change your instructions or the AI and write them again. Earlier versions stay listed with their date, ready to read or restore.
- Transcribe again in another language, without recording again.
- Attach files, images and links to a meeting.
- Bring in audio files, or a transcript exported from Teams, Zoom or Webex.
Recorded too much? Cut it.
Trim the wait before the meeting or the minutes after it. Two meetings back to back? Split them: each gets its own transcript and minutes.
Ask from the meeting you're reading.
Type an instruction under the minutes, and your AI tool answers without leaving pepito, from this meeting or any other.
- Ask what was decided, here or in earlier meetings.
- Add what the transcript missed, and write the minutes again with it.
- Have the minutes rewritten for another audience. The old version stays.
With Claude Code, Codex or opencode and the MCP server on. How the Prompt Bar works
Transcription runs on your Mac.
pepito does not send the audio. Only the transcription models are downloaded, by macOS.
- Recording
- Written to disk on this Mac
- Audio
- Deleted automatically 24 hours after the meeting by default, imported audio included; choose 48 hours, 7 days or 30 days, or turn it off in Settings, Privacy. Deleted audio stays recoverable for 7 days in pepito's trash. The transcript, the minutes and the notes stay.
- Transcription
- macOS speech recognition, on this Mac
- Transcription models
- Downloaded by macOS from Apple's servers
- Minutes
- Sent to the AI you choose. With a local model such as Ollama or LM Studio, they stay on the Mac.
Your meetings, in your AI tools.
Connect Claude Desktop, Claude Code, Codex, opencode or Witsy: your AI tool finds your meetings, answers from them and keeps them organized.
It stays off until you turn it on, and it is reachable only on this Mac.
With Claude Code, Codex or opencode, you can also ask from inside pepito, on the meeting you're reading.
- Codex
- Claude
- Gemini
- Ollama
- Your company's LLM
What did we decide on Avenir this week?
list_recordings tag project::avenir, October 19 to 23: 3 recordings
get_summary SteerCo Avenir, Thursday 14:30
Three meetings touched Avenir. The decisions came from Thursday's SteerCo:
- Go-live moves to 14 November, after a last rehearsal on staging.
- The internal LLM server replaces the pilot API from phase 2.
- The phase 2 budget is approved as proposed.
Wednesday's load test retrospective explains the date: the first run failed on the connection pool.
Friday's sprint planning added the rehearsal. File it under Avenir too.
update_tags Weekly sprint planning, Friday: added project::avenir
Free to use. Pay once to lift the time limit.
Free
0 EUR
Forever.
- All features
- Unlimited recordings
- Up to 50 minutes per recording
Pro
35 EUR
35 EUR
Once, with every future update.
Free with the code during the launch. Ends 31 December.
- Everything in Free
- No time limit on recordings
No subscription, no account. Payment by Paddle, our reseller. VAT included where applicable.
What goes over the network.
Choose your settings and see where each piece of data goes. Check it yourself with Little Snitch or LuLu.
- Stays on this Mac
- Leaves this Mac
- Downloaded by macOS
Every route, in a table
- RecordingNowheremacOS captures the system audio and the microphone, pepito writes them to disk. No bot, no browser extension, no screen recording.
- TranscriptionThis MacmacOS speech recognition runs on your Mac. pepito does not send the audio.
- Transcription command (optional)Wherever the command sends itpepito hands the audio and the locations of the recording's files to your command, on this Mac. What the command does with them is up to the command. How it works.
- Post-processing command (optional)Wherever the command sends itAfter each processing, pepito hands the locations of the recording's files to your command, on this Mac. What the command does with them is up to the command. How it works.
- External sources (optional)Read on this Macpepito reads the audio files of the folders you choose, on this Mac, and offers their import; nothing is imported without your confirmation. A Bluetooth recorder (FOCASE REC) is transmitted by the device without encryption. How it works.
- Minutes, with Ollama or LM StudioThis MacThe transcript goes to a model running on localhost.
- Minutes, with your serverThe URL you enterAny OpenAI-compatible endpoint, such as your company's LLM server. Nothing is sent until you configure it.
- Minutes, with Claude, Gemini, Codex or opencodeThat providerClaude, Gemini and Codex send the transcript to their vendor, under the account you signed in with; opencode sends it to the provider of the model you set in it.
- Your AI tool, through MCPThe model it usesWhat your AI tool reads from pepito goes to the model it runs on, possibly online; pepito does not know which. It also gets the paths of the folders where pepito keeps your meetings, and can read their files with its own tools. Nothing is sent until you turn the MCP server on.
- Prompt Bar (optional)The model your AI command-line tool usesWhen you send an instruction, pepito starts your AI command-line tool (Claude Code, Codex or opencode) on your Mac, with your configuration, and lets it use pepito's tools without asking you. It can read the whole recordings folder, not just the meeting on screen, audio and trash included, and what it reads goes to the model it uses, possibly online; pepito does not know which. pepito does not restrict its other tools, and what was said in a meeting can influence it. Configuration files at the root of the recordings folder apply too, and can make it run commands. Nothing is sent until you send an instruction. How it works.
- Software updatesGitHubIf you allow automatic update checks (pepito asks on second launch; Settings, General), pepito downloads the update feed from GitHub (raw.githubusercontent.com) about once a day, at launch when the last check is more than a day old. Check for Updates does the same on demand. Installing an update downloads it from GitHub (github.com). GitHub receives your IP address and pepito's name and version; no system profile is sent.
- Transcription model, per languageApple's servers, through macOSmacOS downloads and updates the model, not pepito: the first time you use a language, and whenever macOS reinstalls it. The network must let your Mac reach Apple's servers, without TLS inspection or proxy authentication.
Companies in the EU
With a local model, no audio, transcript or minutes leave the Mac. There is no processor to declare and no transfer outside the EU.
Legal, medical, finance
When recordings cannot leave the firm, pepito keeps them in a folder on the machine that recorded them.
Restricted networks
Allow your Macs to reach Apple's servers so macOS can install the transcription model, and connect your internal LLM server for the minutes. Once the model is installed, transcription runs on the Mac.
From download to first minutes in ten minutes.
- Download the DMG, or run
brew install vjeantet/tap/pepito. - The setup assistant asks for microphone and audio access, then asks for the language spoken in your meetings. macOS installs the matching transcription model.
- Start a meeting from the menu bar, or turn on auto-start and let pepito do it.
- Version
- pepito 26.67
- System
- macOS 26 Tahoe or later
- Mac
- Apple Silicon recommended for transcription speed
- App
- Native Swift, signed and notarized by Apple
- Files
- WAV, M4A, SRT, Markdown, JSON
- Languages
- English and French interface, transcription in the languages macOS supports
What comes next.
A meeting matters for what you do with it afterwards. So no frills: a feature gets in only if it helps that work. Our priorities, in this order:
The recording. It starts when it should, as it should, and no recording is ever lost. Everything else comes after.
Your data. Kept safe, with a plain account of what goes where.
Openness. Files you can work with anywhere: minutes in Markdown, transcripts in SRT, details in JSON, and an MCP server for your AI tools.
- Next
Full-text search
Search every transcript and every set of minutes at once, to find a decision, a name or an action.
- Shipped in 26.67
Ask from the meeting you're reading
Ask a question, add what the transcript missed or have the minutes rewritten, without leaving pepito.
- Shipped in 26.53
Meetings in the room start on their own
Turn on the Conversation trigger: pepito starts recording when people start talking around the table, first words included.
- Shipped in 26.36
Ask Codex or Claude about your meetings
With the MCP server, your AI tools find your meetings, answer from the minutes and keep your recordings organized: titles, tags, notes.
- Dropped
Live transcript
Months of use and hundreds of comments from people who use pepito taught us the same thing: a transcript scrolling during a meeting draws your eyes away from the people in it, and gives the illusion of following the discussion. It looks appealing on paper and gets in the way in the room. pepito stays in the background so that you can stay in the conversation.
Questions.
Does my data really stay on my Mac?
Recording and transcription run on your Mac. The minutes do too with a local model such as Ollama or LM Studio. If you choose a remote server or a cloud tool like Claude, Gemini or Codex, the transcript goes to that service, and only because you configured it. The same goes for the MCP server: once you turn it on, what your AI tool reads goes to the model it uses. The same goes for the Prompt Bar: each instruction you send starts your AI command-line tool, and what it reads, anywhere in your recordings folder, goes to the model it uses.
How is pepito different from cloud meeting tools?
Most of them upload your audio for transcription, or join the call as a bot that everyone sees. pepito transcribes on your Mac with macOS speech recognition by default and sends no audio, and captures the Mac's audio directly: no bot, no browser extension, no screen recording.
Does it work without an Internet connection?
Yes, depending on what you choose. For transcription, you pick the engine. macOS speech recognition runs on the Mac, but needs Apple's servers to install the model for your language, the first time and whenever macOS reinstalls it. A transcription command never reaches them: it can run a local model such as whisper.cpp, or send the audio to your company's server. For the minutes, a local model such as Ollama or LM Studio, or your company's LLM server on your internal network, needs no Internet access either.
Which call apps are supported?
All of them, because pepito records the Mac's audio output: Teams, Zoom, Google Meet, Webex, Slack, Discord, FaceTime, phone calls from your iPhone that you take on your Mac, and others. For meetings in the room, it records the microphone.
Which AI can write the minutes?
pepito detects the Claude, Gemini, Codex, Ollama and opencode command-line tools installed on your Mac. It also connects to any OpenAI-compatible server, such as LM Studio, vLLM or your company's API, with the URL, key, model and temperature you set.
Which AI tools can connect to pepito?
Claude Desktop, Claude Code, Codex, opencode and Witsy, through pepito's MCP server. Turn it on in Settings, MCP, then use Connect an AI Tool… to get the steps for yours.
Is pepito GDPR-compliant?
The app sends no audio, transcript or minutes to any server of ours, and we have no access to them. If you configure a remote model, or connect an AI tool that uses one, your obligations toward that provider apply.
Do I have to tell participants that I record?
Consent rules depend on the country and on your organization, and pepito does not handle consent for you. Since no bot joins the call, participants see no recording indicator: tell them yourself.
How does auto-start work?
Turn it on from the menu or in Settings, Auto-Start, and choose the apps to watch. Recording starts after 10 seconds of call audio. The Conversation trigger works on a microphone you choose: pepito listens to it continuously, on your Mac, and starts recording when a voice keeps going. It sends no audio, keeps less than a minute in memory only, while it listens, and writes nothing to disk before a recording starts. macOS's orange dot stays on while pepito listens; turn off auto-start to go off the record. With auto-stop, pepito offers to stop after a silence you set, 30 seconds by default. In Mini mode it stops without asking. If it starts too early or runs too long, trim the recording afterwards.
What if a recording starts too early, or I forget to stop it?
In the recording's … menu, choose Edit Audio… and drag the yellow handles around the meeting. If two meetings ended up in one recording, add a split: the part after it becomes a new recording, at its real start time. pepito transcribes again and writes new minutes; the earlier minutes stay in the version history, and the original audio stays 7 days in the folder's Trash.
Can I import existing recordings?
Yes, M4A and MP3 files. They go through the same transcription and minutes, and you can add context when importing. You can also import a transcript, such as a Teams, Zoom or Webex export, in SRT or VTT, or as plain text or Markdown: pepito creates the recording and writes its minutes.
Your next meeting can start from the last one.
macOS 26 or later, Apple Silicon recommended. Native Swift app, signed and notarized. Files in WAV, M4A, SRT, Markdown and JSON.