8/4/2026
Rory Eckel
Claude Opus 5
I built idea genesis because I kept doing the same thing by hand: read something dense, pull the terms out, write the definitions, build a deck, repeat next week with a different chapter. It is mechanical work that resists a chat window, because what you want back is not a paragraph — it is a list with the same columns every time.
So the first version did exactly one thing. You gave it a document and it gave you vocabulary with definitions. That is what went live at ideagenesis.ai on November 23, 2025, and it is a fraction of what is there now — because vocabulary turned out to be one instance of a much more general shape. Vocabulary is still one of the most used things in the product, and still the shortest path from a chapter to a Quizlet or Anki deck. It just stopped being the only thing.
idea genesis turns any document — a paper, a textbook, a lecture recording, a transcript — into structured output you can sort, export, and build on. Not a summary you read once and lose. I kept building instead of announcing; this is the post I owed it.
There are a lot of tools now that will let you ask questions about a document. They are genuinely useful, and idea genesis is not trying to be another one.
The problem with conversation as an interface is that it produces one loose answer at a time. Ask the same question of forty papers and you get forty differently-shaped replies that you then have to normalize by hand. Ask it again next week and you get a forty-first shape. Chat is great for exploring a document and terrible for processing one.
What I wanted instead was repeatable structure. Not "summarize this paper" but "give me the method, the sample size, and the stated limitation, as those three fields" — and then have that same instruction work identically on the next forty papers, so the results line up in a table instead of forty essays.
That instruction is what idea genesis calls a schema: a definition of the output you want, kept separate from any particular document you run it against. You choose one, point it at your material, and what comes back has the same shape every time. Decide what you want out of a document once, and stop re-describing it.
Everything in the product is one of three operations, and they are deliberately kept separate because they have different shapes:
The useful part is that they compose. Extract the terminology from a textbook chapter, transform every term to add the surrounding context it appeared in, then synthesize the whole set into a single study guide. Three passes, one document, and each intermediate result is saved and reusable rather than thrown away.
That progression — pull it apart, refine the pieces, put it back together — turns out to be most of what people actually do with dense material. The tool just makes each step explicit instead of asking you to hold it all in a chat window.
PDF, DOCX, PPTX, XLSX, TXT, Markdown, and HTML files. Paste a URL and the page text gets pulled in. Drop in an audio or video file and it is transcribed first, then treated like any other document.
That last one matters more than it sounds. idea genesis works on any document, not just research papers. A recorded lecture, a conference talk, a two-hour meeting — these become searchable, extractable text that you can run the same schemas against as anything else. Most tools in this space quietly assume you are holding a PDF from a journal. A lot of people are holding a recording of a class instead.
For long documents you can pick which sections to process rather than paying to run the whole thing, which is the difference between analyzing one chapter and analyzing an eight-hundred-page book.
Results are structured, so they export cleanly: Markdown, CSV, XLSX, JSON, Quizlet, Anki, calendar files for anything with a due date, and DOCX from document templates.
You can also generate a public link to a result that anyone can open without an account, and everything you run is kept in a library with its lineage intact — so you can see that this study guide came from those terms, which came from that chapter.
The principle is that output should land where you already work. Obsidian, a spreadsheet, a flashcard app, a shared doc. Structured data is portable almost by definition, and I would rather you take it somewhere else than be stuck opening my site to look at your own notes.
The built-in schemas cover the cases that come up most: vocabulary, action items, question-and-answer pairs, contacts, spreadsheet rows, study guides, summaries, exam cram sheets, deadline lists. Between them they handle a lot more than I expected them to when I started.
The flexibility is in their parameters. The vocabulary schema alone lets you set the topic to focus on, filter to nouns or verbs or phrases, and dial the complexity of the definitions up or down — the same schema producing very different output for a linguistics seminar and a certification exam. Most of the adjusting people actually need turns out to be this kind, not a new schema from scratch.
Writing your own is the next significant thing on the roadmap. For now, the built-in set and its parameters are what is available.
Everything above is something you do once. Schedules are for when you want it to keep happening.
Point a schedule at a URL, choose the schema to run against it, and idea genesis fetches that page on the cadence you set — every morning, once a week, whatever suits the material — and produces a fresh structured result each time. Those land in your library like any other run, so they accumulate: the same fields, in the same shape, dated, one after another. It behaves much more like a feed than a folder of one-off extractions.
A daily pass over a news or publications index gives you a dated series of structured rows you can scan in ten seconds, instead of a tab you keep meaning to open. A weekly run against a course or documentation page leaves you with a record of what the material said on each date you checked it. A recurring run through a schema built for dates and deadlines quietly assembles a calendar you can export. None of these are things you would remember to do by hand every Monday, which is rather the point.
Schedules are a Plus feature, capped at three running at once. Each run spends credits when it fires, and you can pause, resume, or trigger one early from the schedules page.
This one is not live yet, but it is close, and it is the feature I am most interested in.
Most in-app AI assistants are search over the documentation. You ask where a setting is and they tell you where to click. The assistant coming to idea genesis knows what is on the page you are looking at and can work the controls itself — choose the schema, fill in the parameters, select which sections of a long document to include, and start the run. "Pull the key terms out of chapter four and keep it to what I would actually see on the exam" becomes something you say once instead of six things you click.
The other half of that is trust, and it matters just as much. Anything that spends credits stops and asks first, and the confirmation tells you what it will cost before it runs. An assistant that can operate the interface is only a good idea if it cannot quietly spend your money by misreading you.
A general assistant lands alongside it — questions answered against your documents, help shaping a schema, handing work off to a run without leaving the conversation. Both are coming soon rather than available today.
Credits are priced per unit of work rather than per session: uploading a file costs nothing, extraction is charged per item produced, transcription is priced by the minute, and repeat visits to a page or recording you have already processed are free. Every action tells you what it will cost before you commit to it. Current allowances and pricing are on the Plus page.
I have tried to be straightforward about cost, because a study tool that surprises a student with a bill is a study tool they stop using.
A browser extension is in the works — capture the page you are reading straight into your library, then decide what you want made from it. It works; it is not in the extension stores yet.
Custom schemas are the bigger one — describing the output you want in plain language and getting a schema for it, instead of choosing from the built-in set. The assistants described above land in the same stretch.
There is also something larger in development that I am not ready to show yet. The short version: a pile of extracted results is not the same thing as a way to learn the material, and the gap between those two is the most interesting problem in front of me right now. More on that when it is real.
No account required to start. Upload something dense and see what falls out of it.
Try it at ideagenesis.ai. If you use it and something is missing, I would like to hear about it.