Your research record

A workspace of projects

One root folder holds one folder per project. That is the whole data model.

Point Bench at a folder — say ~/Heureka Bench — and every project inside it becomes a card on your dashboard. Create a project and you get a directory. Delete the app and you keep everything.

Projects carry their own colour and name, and can be renamed from the card menu. Inside, folders like Data, Code, Drafts, Reviews and Analysis are where your work and the agent's output land, so results never scatter.

1. On first launch, choose the folder that will hold your projects.
2. Click New project on the dashboard, or drop an existing folder into the workspace root — Bench picks it up.
3. Open a project to get the workbench view: folders in the main column, project context in the rail beside it.

Files, opened in the right viewer

Click any file and Bench renders it properly instead of offering to download it.

The file panel is a tree with tabs. Markdown opens in the editor, code in a syntax-highlighted viewer, PDFs and Word documents in a real document viewer with a thumbnail rail, images inline, and scientific formats in their own purpose-built viewers — see Data & analysis for structures and datasets.

Opening a markdown file collapses the file strip automatically so prose gets the full canvas. Everything else keeps the tree in view.

Project rationale

Why this project exists, written down once, and read by the agent every time.

Each project holds a short rationale: the goal, the background, the approach, and what success would look like. It sits in the rail next to your folders so it is in front of you rather than in a grant document you last opened in March.

ARC reads it. That is the point — you stop re-explaining the project at the top of every conversation, and analyses come back framed against what you actually set out to test.

Experiments and templates

Start an experiment from your own house format instead of a blank file.

New experiments are created from a template you control, so every entry in the notebook opens with the same headings your lab already agreed on — aim, materials, procedure, observations, whatever your discipline expects.

Edit the template once in Settings and every future experiment follows it. Existing entries are left exactly as they were written.

1. Settings → Templates to set the experiment template.
2. In a project, use New experiment to create an entry from it.

To-Dos the agent can see

A task list inside every project — and ARC starts every conversation knowing what is on it.

Quick-add items with markdown descriptions (paste a screenshot straight in), give them due dates, and drag to categorise. Each to-do is a plain markdown file in the project, so the list survives outside Bench like everything else.

ARC can add items and check them off itself. Run with ARC hands one off as a background task that updates its own status as it goes — you write "re-run the QC on the March batch", walk away, and come back to it done and ticked.

What is due, on the dashboard

Dated, unfinished work from every project gathered into one quiet digest.

The top-right of the dashboard collects dated to-dos across the whole workspace: overdue called out on its own line, today and this week in a sentence, and the rest of the month in a breakdown you can expand. Click any item to land in that project's list.

When nothing is due it stays out of the way entirely. It is a digest, not a dashboard widget demanding attention.

Notes, linked into a body of thought

A cross-project vault of markdown notes with wiki-links, backlinks and a graph.

Notes live in their own vault outside any single project, because thinking rarely respects project boundaries. Write in a Write / Read / Graph switcher: Read renders live [[wiki-links]] — hover for a preview card, click to follow, click a link to a note that does not exist yet and it is created on the spot.

Backlinks show the actual lines that reference a note, and an Unlinked Mentions section catches notes that name it without linking. Cmd+P opens a fuzzy switcher across the whole vault; Cmd+Click follows a link from inside the editor.

The graph is draggable and zooms to your cursor, highlights by tag from the legend, remembers your layout per vault, and shows dashed ghost nodes for notes you have linked to but not written yet.

Quick capture

Get a thought into the record before it evaporates.

A capture box that files a note without making you navigate anywhere first. Cmd+Shift+N opens it from anywhere in Notes (Ctrl+Shift+N on Windows). It is the smallest feature in this section and the one you will use most.

What Bench remembers about your work

Project memory and workspace memory, kept as readable files.

Bench keeps a small amount of durable context so the agent is not amnesiac between sessions: who you are and how you like to be worked with at the workspace level, and what matters in each project at the project level.

All of it is plain markdown inside the workspace, next to the projects it describes. You can read it, edit it, or delete it. Nothing about you is held somewhere you cannot open.