Drop the recording. Get a full transcript, identify who said what, and a ready-to-send report, in the format the audience needs. Minutes instead of hours.
curl -fsSL https://raw.githubusercontent.com/Lesur-ai/transkryptor/v6.1.1/scripts/install.sh | bash
macOS / Linux · self-hosted on your machine · works offline once installed
Transkryptor is built around the moment after the recording — when you still have to listen back, decide what mattered, and write it up. Pick who you record for and the report comes out in the right shape.
Standups, steering committees, client calls. Get a structured recap with decisions, owners, and next steps, in the same format your team already reads.
Lectures, workshops, onboarding sessions. Get a thematic synthesis attendees can actually read, plus the full searchable transcript for later.
User interviews, podcasts, field recordings. Cleaned-up verbatim, speaker turns identified, recurring themes surfaced, with quotes you can lift straight into a deck.
Same recording, multiple audiences. Pick a preset to ship a report in a click; switch to custom when you've got a house template you want to keep using.
A short, decision-grade brief for leadership.
Sectioned by topic, attendees, decisions, open questions.
A flat list of commitments, owners, and dependencies.
Filler words removed; meaning and structure preserved.
Recurring themes across the recording, with grounding quotes.
Bring a house template. The prompt editor lives in the sidebar.
An hour of audio shouldn't mean an hour of waiting. Three things make Transkryptor work for real, long-form recordings:
Recordings get processed in parallel instead of one long queue. You watch progress live, then download the report.
Optional speaker detection labels turns of speech, so a three-person interview reads like a conversation, not a wall of text.
Record in French, send the report in English. Or the other way around. The two languages are picked independently.
Four stages, run in one pass. You see everything as it happens: per-chunk progress, throughput, and live logs, so it never feels like a black box.
The file is split into overlap-safe windows so long recordings can be processed in parallel.
Each chunk is transcribed concurrently, then stitched back in order, with optional speaker detection on top.
The transcript is split into semantic batches, each analysed in parallel against the model you chose.
Batched analyses fold into one report, in the format you picked. Download as Markdown, share anywhere.
Transkryptor is optimised for the Cloud Temple LLMaaS (SecNumCloud, EU-bound) — and works out of the box with any OpenAI-compatible endpoint. Bring whichever inference provider already fits your security and budget constraints.
A self-hosted vLLM, an EU-sovereign API, a US frontier-model API, an on-prem appliance — if it speaks the OpenAI chat / transcription protocol, Transkryptor connects to it. One URL, one key, in your .env.
Default settings, prompts, and retry policy are tuned against Cloud Temple's SecNumCloud-qualified LLMaaS. Plug in the key and it just works — audio, transcripts and prompts stay inside the EU perimeter.
Prerequisites: Node.js 18+ and an API key from your provider of choice. The installer drops the app in ~/Applications/Transkryptor, sets up the config, starts the service on localhost:3000, opens your browser.
$ curl -fsSL https://raw.githubusercontent.com/Lesur-ai/transkryptor/v6.1.1/scripts/install.sh | bash
# installs to ~/Applications/Transkryptor (macOS)
# or ${XDG_DATA_HOME:-~/.local/share}/transkryptor (Linux)
# starts the service, opens http://localhost:3000$ git clone https://github.com/Lesur-ai/transkryptor.git
$ cd transkryptor
$ npm install
$ cp .env.example .env
# edit .env — point it at any OpenAI-compatible endpoint
$ npm start
# → open http://localhost:3000$ git clone https://github.com/Lesur-ai/transkryptor.git
$ cd transkryptor
$ cp .env.example .env
# edit .env — point it at any OpenAI-compatible endpoint
$ docker compose up --buildPS> wsl bash -lc "curl -fsSL https://raw.githubusercontent.com/Lesur-ai/transkryptor/v6.1.1/scripts/install.sh | bash"
# a native Windows installer is on the roadmap
# for now, WSL is the supported path
.env before you startSet the API endpoint and key of the OpenAI-compatible provider you want to use, plus the comma-separated list of models the UI should expose. The first model becomes the default.
Self-hostable, vanilla stack, contributions on GitHub.
Any OpenAI-compatible endpoint — sovereign, self-hosted, or commercial.
The API key lives on your host. Audio never leaves your provider.
Per-chunk progress, throughput, server logs streamed live to the UI.
Transkryptor is live and open-source. The fastest way to support the project is a star and an issue when you find a rough edge.