Press record: your microphone on one channel, everything your Mac plays on the other. Thunderegg transcribes it and writes the note you would otherwise have taken — summary, decisions, and action items with an owner and a due date. Recording and transcription never leave your Mac; the write-up runs on a model you choose, on your Mac or your own Claude, OpenAI or OpenRouter account. No Thunderegg account. Recording needs macOS 15 or later.
Most “convert to Markdown” tools upload your documents to a server. For contracts, medical records, financials, or anything under NDA, that’s a non-starter.
Thunderegg does all conversion, OCR and transcription on-device — documents through a bundled converter, images through Apple’s built-in Vision engine, speech through an engine that holds no network code at all. No setting sends any of them anywhere.
The AI layer is where you get a choice, and it is a real one. Until you connect an account or paste a key, it runs on-device and none of your notes go anywhere. Point it at a model you run on your Mac, or connect your own Anthropic, OpenAI or OpenRouter API key and the same features run on the better model — billed by that provider, never by Thunderegg. Change your mind and switch back in one setting.
AI models are trained on Markdown. Feed them clean .md instead of PDFs and Word files, and you get sharper answers for fewer tokens — while your knowledge stays portable forever.
Markdown strips the formatting bloat of PDF and DOCX. Same meaning, ~30% fewer tokens — lower API costs and more room in the context window.
Headings, lists and tables survive the conversion. That preserved hierarchy is exactly what models use to reason — clearer answers, fewer hallucinations.
Locked in a PDF, a document is a dead end. As .md it becomes a node — wiki-linkable in Obsidian, Notion or any vault, so every file connects to the rest of your thinking instead of sitting in a silo.
Documents, spreadsheets, email, images and recordings — all converted to clean, LLM-ready Markdown.
Email becomes a tidy header + body (attachments listed, never dumped as raw code). Screenshots & scans are read with Apple’s on-device Vision OCR. Audio and video are transcribed on your Mac — Zoom, OBS, Discord and browser recordings included.
Whatever fits your flow — including a file you don’t have yet.
Drop one or many files onto the Thunderegg icon in your Dock. Thunderegg converts them and shows you what it made.
Select any files → Quick Actions → Convert to Markdown. System-wide, no app to open.
Drop files into your MarkItDown Watch folder and they convert automatically — originals tidied away for you.
Hit record and Thunderegg captures the call itself — then turns it into notes. No file to hunt down, no second app in the loop. Recording needs macOS 15 or later.
Press record. When you stop, Thunderegg transcribes the whole thing on your Mac, then writes the note you would have had to take yourself — using a local model, or your own cloud account if you want the sharper summary.
Thunderegg records quietly: it never joins the call and never announces itself, so the other people are not told. Recording laws differ by place, and many — California and Florida among them — require everyone’s consent. Tell the people you are recording, and get their agreement where the law asks for it. The app reminds you every time.
Your microphone is recorded on one channel and everything your Mac plays on the other — so the transcript can tell your words from theirs without guessing, and without a second AI model.
Executive summary, action items with owner and due date, decisions, open questions, risks and the quotes behind them. Action items join your to-do list across the whole vault.
You cannot switch this one off. Decisions, open questions, risks and notable quotes are each optional; admitting what it could not make out is not. Nothing unclear that time? The section is removed rather than padded. And when an on-device model reports a size below the point where summaries turn confidently wrong, the transcript is kept underneath, so every line can be checked against what was actually said — a local server that reports no size at all is assumed to be big enough. Drop it when you’re happy.
Transcription runs entirely on your Mac — the audio is never uploaded. The write-up needs a model you supply: one running locally, or your own Claude, OpenAI or OpenRouter account. Choose cloud and the whole transcript goes to it — not a summary of it — along with the recording’s filename, its length, and, if you have set them, your vault’s correction glossary and your name. Connect neither and you keep the transcript, unsummarised. Pick Local and the whole meeting stays on the machine.
A summary you read once is a summary you forget. Every action item Thunderegg pulls out of a meeting lands in one queue across the whole vault, and on a timeline built only from dates your notes already state.
Overdue, due this week, later, yours with no date, and waiting on someone else — every open item in the vault in one list, worst first. It reads the checkbox in the file rather than asking a model what is still open, so it is instant and it cannot be wrong about what you already ticked.
A bar runs from the meeting where the commitment was made to the date it is due — two real dates, never an estimate. No due date means no bar: the item is listed as unscheduled instead of being drawn at an invented width. Overdue work grows a second segment to today, so slippage shows up as length.
A deadline written in prose is invisible to the date buckets, and an item with two owners is in nobody’s queue. Thunderegg reports both rather than guessing, and offers you the date the note already states — one click to accept. Inventing a date from prose is the one thing worse than having none.
This runs over the notes already in your vault, so anything you record or convert feeds it. No model is involved at any point and nothing leaves your Mac — it is reading your own files.
Converting is step one. Thunderegg then reads what it converted — a one-line summary, tags, a topic folder, a map of what connects to what. You choose the engine: a model running on your Mac, or your own Claude or OpenAI account. The features are the same either way; a cloud model is faster and handles longer notes and transcripts. You pick per vault, or for the whole install.
Each note gets one line saying what it is, plus tags — and links to the related notes already in your vault, so a PDF you dropped in March is findable in November. With nothing connected it still runs, on a plain on-device pass: a first-sentence summary and a tag for the file type, marked rough so you know which it was. Very large batches — 200 files or more at once — convert fast and skip the AI pass by default.
Point it at a pile of converted files and it proposes folders by subject — reversible, so a sort you dislike costs nothing. This one needs a real model behind it: a local one, or your own key. The maps that link topics together are built on your Mac from the summaries already written, without a second model call.
Local points the AI at Ollama on your Mac — you install Ollama and pull a model once, then it is private, free to run, and slower. Cloud runs the same features on your own Anthropic, OpenAI or OpenRouter API key — faster and more accurate. One setting, switchable whenever you like.
Everything the upload-a-file converters do, without handing them the file — and without the sign-up or the monthly bill for the app.
| Feature | Thunderegg | Upload-to-a-server tools |
|---|---|---|
| The file itself | Never uploaded to convert — conversion, OCR & transcription run on your Mac | Uploaded to convert |
| Where the AI runs | Your choice — on your Mac, or on your own cloud account | Their server, no choice |
| Converts offline | Yes | No |
| Account required | None for Thunderegg — cloud AI runs on your own provider account | Sign-up |
| Pricing | Converting is free · Pro $19.95 once | Per-page / monthly |
| Who pays for the AI | A local model is free to run · cloud is your own account, billed by them | Their account, their price |
| Image OCR | Apple Vision | Varies |
| Record & summarise a meeting | Built in · recording & transcription on-device, summary on the model you connect | Separate subscription |
| Email (.eml / .msg) | Clean parse | Rare |
| Where the Markdown lands | Your vault — or beside the file | Download |
Convert as many documents as you like, for ever, without paying anything. Pro adds the two things that go beyond a converter: recording your meetings and the AI write-up. One payment, no subscription, ever, and three recordings free to try.
report.pdf becomes report.md. By default it is filed into your vault; set Where converted notes go in Settings to Beside the original file and it is written next to the source instead. Two files that would claim the same name never overwrite each other.Stop pasting documents into web tools. Convert on your Mac, and decide for yourself where the AI runs. Converting is free, for ever; recording and the AI write-up are $19.95 once, with three recordings free.
Download — converting is free