Introducing Heureka Bench

Count the places your current project lives. Raw data in one folder. Analysis scripts in another. Notes in a notes app that has never seen the data. Papers in a reference manager that has never seen the notes. Samples in a spreadsheet somebody emailed you. A draft in a cloud document, quietly out of date.

Nothing is missing, exactly. It's just that no two pieces know about each other, and you are the only thing holding them together. Every time you switch tools you pay a small tax in re-explaining where you were, and by month six of a project that tax is most of the day.

Heureka Bench is a desktop lab notebook that puts all of it in one place — and gives that place a research agent that can touch every part of it.

One project, one place

A project in Bench is a folder on your own disk, and everything about the work lives inside it: files, notes, linked papers, to-dos, results.

• Notes that link. Write with [[wiki-links]] and get backlinks, forward links, and a graph of how your thinking actually connects. Hover a link to preview it, click one that doesn't exist yet to create it on the spot. Notes attach to projects and papers, and to each other.
• A paper library. Your PDFs with their metadata — authors, year, tags, your own annotations — searchable, and attached to the projects they belong to.
• The lab itself. Inventory, samples, animals and their lineage, tracked next to the analysis instead of in a spreadsheet nobody has opened since the postdoc left.
• Your files, understood. A browser that knows the difference between a script, a table, a structure and an image, and renders them instead of showing you a wall of characters.
• Slides in Markdown. Build the lab-meeting deck without leaving the app and present it fullscreen.

An agent that can see all of it

This is the part that only works because of everything above.

ARC is our research agent, and in Bench it runs against your actual project — reading your files, searching your notes, checking the literature, writing and running analysis code in the context of the work in front of it. It knows what the project is for and what has already been tried, so you are not re-explaining yourself at the start of every session.

In practice that means the request changes shape. Instead of "what's known about this gene," it's look at the counts table in data/, tell me which comparison is worth following, and draft the methods paragraph for the one you pick — and the answer arrives as files in your project rather than text you have to copy somewhere.

Three things follow from that:

It keeps working when you look away. Conversations run in the background, one per project, so you can switch tasks and come back to find it still going. Recurring work runs on a schedule — a weekly data-quality check, a standing literature sweep, a monthly diff of what changed in a dataset — and the result is waiting for you rather than requested by you.

It can be taught. A protocol you explain more than twice is one ARC should already know. Write it down once as a Skill and it follows the same steps every time after, which is the part that matters when somebody asks whether two runs were done the same way.

It leaves a trail. As it works, ARC records where things came from. Make a figure from a table and the link between them is captured without anyone asking. Months later, that's the difference between writing your methods section and reconstructing it.

Because it's yours

All of this is possible because the data never had to leave in the first place.

Your projects are directories on your filesystem, in standard formats other programs can read. It works offline. There is no lock-in — you can walk away whenever you like and take everything with you, because it was always just files. And when you do want a backup, Bench encrypts the project on your machine before anything moves.

For work that shouldn't touch the network at all, a project can be locked down entirely: ARC keeps working on your local files and everything that would send data outward is switched off, with a running receipt of what was blocked and why.

And when an analysis is genuinely the wrong shape for a laptop, you can hand that one job to our cloud and have the results come back into the local project. You decide, job by job.

Get it

Heureka Bench is free to download.

Download Heureka Bench