Hi all,
I have been building a small free tool called Micronaut and would value this community’s eyes on it. It is a planner for the stage before acquisition: the point where you decide what you are comparing, what counts as one independent replicate, which controls you need, and how files will be named. Full disclosure: it is my own project, so I am posting it here mostly to find out where it is wrong compared with how other people actually work.
The reason it exists: As a core facility scientist, a lot of what sinks an imaging experiment is fixed long before the user meets the microscope. Zero single stains, no unstained controls, overlapping emission spectra, filenames invented at the scope that mean nothing months later, a missing control noticed only at image analysis. Facility folks have seen it all, and then some more. Micronaut tries to surface those decisions early, while they are still cheap to change.
A few things it does that I think are worth your look:
- Plain-language intake that does not overreach. You can write your study out in prose, and it extracts only the things it can match exactly (a marker, a replicate count, a magnification, an ISO date) and shows you the literal text it matched. Nothing is written to your plan until you accept it. Whatever it cannot match, it leaves as your narrative rather than guessing.
- A qualitative spillover check across your panel, flagging excitation/emission peaks that sit uncomfortably close, with the reasoning shown, not just a warning. Spectral values are drafted from published references and flagged as not specialist-reviewed, so treat them as a first pass.
- Suggested controls with the reason attached, specific to the readout, so you can disagree on the merits.
- A filename convention built from the finished design, plus a review page with one pass/fail readiness verdict, a study diagram, a printable bench card, and Markdown/SVG/CSV/JSON exports.
Want to consult with AI? Why not?
The review page has a “copy prompt for your own LLM” button. It copies your whole plan to the clipboard as structured JSON: every measurement, the panel with its wavelengths, each suggested control with its reason, and every planned filename. Around that data, it wraps a short instruction telling the model to reason only over what is actually there and to name what is missing instead of inventing a default. You paste it into whatever model you already use, in your own account, and read the answer yourself. Micronaut does not call any model itself, and nothing the model says comes back into your plan unless you type it in.
I had some thoughts about running the app on a local server with an open-source model, but that’s a future feature.
It runs entirely in your browser (as the app viewer only). No account, no upload, no server at any point. Your study autosaves to local storage on that device, keeps the last few versions for undo, and exports as a single file you can back up or move. Nothing leaves your machine unless you choose to send it somewhere.
It is open source, Apache-2.0.
I would genuinely like to hear:
- If you plan imaging experiments now, what do you use, and what is the first thing that breaks?
- Where is the control or spillover advice wrong or too generic for your modality?
- What would make something like this actually fit into your existing workflow rather than being one more tab?
Links:
- App: Micronaut Planner
- Manual: Micronaut Planner — User Manual
- Release notes: Micronaut Planner — Release Notes
Thanks for reading, and for any pushback.
Cheers,
Daniel