Meeting recordings turn into readable notes without lifting a finger.
offline transcription
Recordings become documents. Automatically.
Akdira Transcribe watches your folder and turns audio and video into tidy Markdown, complete with speaker names and timestamps. Everything runs offline on your computer.
interface illustration
It finishes itself, and stays on your computer.
Who said what is recorded with name and time.
Sensitive recordings stay on your computer — no uploads, no cloud account.
The same file is designed not to be processed twice — no time or power wasted.
The output is structured Markdown you can search, read, and feed to AI agents right away.
Everything you need, without manual steps.
Watch a folder, transcribe automatically.
Choose the folder to watch, Downloads by default. As soon as a new recording arrives, Akdira Transcribe waits 5 seconds for the file to finish writing, then processes it on its own.
- Runs in the tray and starts automatically with your computer
- Notification when a document is ready
- A Teams-style interface that feels familiar
Who said what, with the events too.
Diarization separates the speakers, and voiceprints recognize the same voice in later recordings. Name them once and they are recognized automatically from then on.
- Speaker names complete with timestamps
- Audio events: music, laughter, applause, silence
- Video screen descriptions through the built-in multimodal LLM
Offline, private, and power-friendly.
All inference runs on your computer's CPU. No uploads, no cloud account, no telemetry. The built-in LLM uses only about 1 GB of RAM and runs only when needed.
- SHA-256 prevents the same file from being processed twice
- Designed to finish faster than the media's own duration
- Example: a 5-minute video finishes in about 2–2.5 minutes (Ryzen 5 PRO 2400G, 4 cores, no GPU)
OKF Markdown, ready to use.
Every recording becomes one .md document in OKF format, plus an index.md that lists them all. Easy to search, pleasant to read, and ready for AI agents.
- akt CLI with JSON output for automation
- OpenClaw skill to call akt from an agent
From recording to document in four steps.
Drop in a recording
Drop or download a recording into the watched folder (default: Downloads).
Detected automatically
The file is detected, given 5 seconds to finish writing, then queued.
Processed offline
Transcript, speaker recognition, and events (plus video description) are produced offline, within the time budget.
Saved as Markdown
The OKF Markdown document is saved along with index.md. Name a speaker once and they are recognized from then on.
System requirements.
- operating system
- Windows x64 (.exe), macOS Apple Silicon and Intel (.dmg), Linux x64 (AppImage or .deb)
- processor
- CPU only, no GPU. Test example: Ryzen 5 PRO 2400G, 4 cores
- memory
- About 1 GB of RAM for the built-in LLM, only while in use
- network
- Runs offline, no cloud account
The installer is not signed yet. Windows may show SmartScreen ("More info → Run anyway"), and macOS Gatekeeper may block apps that are not notarized.
Open transcripts in Writer.
Akdira Writer
Polish your transcripts in a visual Markdown editor in the style of Word. The OKF header stays intact when you save.
Frequently asked questions.
Does Akdira Transcribe need internet?
Not for transcription. All processing runs on your computer's CPU, with no upload and no cloud account.
Are my recordings sent to the cloud?
No. Recordings, transcripts, and the database stay on your computer. The app uses no cloud services and no telemetry.
Do I need a GPU?
No. The pipeline runs on the CPU. For example, a 5-minute video finishes in about 2–2.5 minutes on a Ryzen 5 PRO 2400G, 4 cores, no GPU.
What does the output look like?
A Markdown document with an OKF (YAML) header, a title, a description, then dialogue lines with the time, speaker name, and text. The output folder also contains an index.md that lists all transcripts.
What if the same file arrives twice?
Akdira Transcribe computes the SHA-256 of the file contents. Identical contents are not processed again unless you force it with --force, and every file location is still recorded.
How are speaker names recognized?
Diarization separates the voices and voiceprints remember them. Name a speaker once, and later recordings are recognized automatically.
Can AI agents use the output?
Yes. The akt CLI supports --json output, and an OpenClaw skill that calls it is available.
Why does Windows show SmartScreen?
The installer is not signed yet (code signing). Windows will show SmartScreen ("More info → Run anyway"), and macOS Gatekeeper blocks apps that are not notarized.
Start with a single recording.
Install Akdira Transcribe, pick a folder, and let the rest run by itself.