Find and organise papers, pull out the ideas that matter, and plan the work around them. Your notes, tasks, files and dates live here too — so you are not stitching four apps together to answer one question.
Accounts are set up by invitation. Ask for one.
Papers, and how they cite each other. The bright ones are the papers you kept.
Gather papers, read them, and map out the ideas — and the questions nobody has answered yet.
Plan and run the work. Notes, tasks, files and dates, kept together for each project.
You pick one of the two when you make a project, so you only ever see the tools that suit the work. A client project will not show you a screening queue, and a thesis will not ask who the client is.
More than somewhere to keep PDFs. You end up with the papers you chose, the reasons you chose them, and a clear view of what your field has and has not answered.
One search goes out to OpenAlex, Crossref, arXiv and Semantic Scholar at the same time. If three of them return the same paper, you get one entry instead of three.
You can also just bring your own. Upload PDFs, paste a DOI, or drop in a list you already have — searching is a shortcut, not the only way in.
Nothing joins your list until you tick it. That matters more than it sounds: a slightly loose search can pull back hundreds of papers from a completely different field, and weeding those out later is far more work than ticking a few boxes now.
Go through papers one at a time and mark each one in or out, with a short reason. The reason is part of the same click, so you are never tempted to skip it.
When you come to write up your method, the reasons are already there. And a paper that is only loosely related gets described that way, rather than stretched to sound relevant.
Explanation strategies for a robot that must decline a request
We compare four explanation strategies used by a service robot when it cannot comply with a request, and measure the effect on perceived competence and willingness to re-engage.
Directly on explanation strategy and its effect on the interaction.
Each summary says whether it was written from the full paper or only the abstract, and how confident that makes it — so you know which ones to double-check before you quote them.
If a paper does not give a number, the field stays empty. Nothing is filled in with a sensible-looking guess, because a made-up participant count is the one mistake a literature review cannot come back from.
Explanation strategies for a robot that must decline a request
The ideas in each paper get pulled out and joined up when papers discuss them together. So a cluster means “these two come up as a pair”, not “these words look similar”. Drag anything to move it and it stays where you put it.
There are plainer charts too — papers by year, methods by count, and the citation map from the top of this page.
Sort your papers into topics, up to four levels deep. Rename a topic whenever you like and the papers stay attached to it.
Then write down the gaps. Each one records what is missing, what would settle it, which papers touch on it, and what is still open after them. These are kept as plain files — docs/gaps.json — because what you claim is missing from a field is an argument you will have to defend.
Your topics and your gaps are saved as plain files you can keep alongside your writing — they are your argument, not something the app decides for you.
The same workspace, pointed at the projects you are running. Each one is separate and can be shared. Notes, tasks and files can also sit outside every project, for the bits that belong to nothing in particular.
Notes are plain text with light formatting. Highlight any sentence and a box opens so you can ask about it. The answer is kept beside the line it came from, so picking it up next week still makes sense.
Only that note is used as background — not your other notes. Quietly reaching wider is how you end up with an answer about something you cannot see.
Highlight a sentence below — or just tap one.
The engagement brief asks for a single dashboard, but the three source systems disagree on what counts as an active account.
Finance treats an account as active until the contract end date; the CRM drops it the moment a renewal task is closed.
Reconciling the two is the actual scope of the work, and it needs a decision from the client before any of the reporting is built.
Drop a block of HTML into a note and it shows up as the real thing. Useful for a chart or table you already have, since that is usually the form it arrives in.
It runs inside a locked-down frame, kept away from the rest of the page. Projects get shared, so a note from someone else must not be able to run anything in your browser. The one on the right is live, not a picture.
One page showing every task from every project, plus the ones that are not tied to a project at all. You can add either from the same box.
A task belongs to a project or to you, never both. Yours stay private; a project's can be picked up by anyone working on it.
Send the reconciliation question to the client
Screen the remaining 42 abstracts
Renew the library proxy access
Write up the gap on trust
Invoice for March
A task is either a project's or yours alone. Yours stay private.
Upload a file to a project, or paste an image straight into a note — both end up in the same list. So a screenshot you pasted mid-sentence is still findable later, labelled with the note it came from.
Files are private. Only people on the project can open them, and that is checked every time a file is downloaded rather than once when it went up.
acme-data-dictionary.xlsx
kickoff-whiteboard.png
from the note “Scope questions”
msa-signed.pdf
migration-runbook.md
from the note “Cutover plan”
HTML and SVG files are not allowed — they can run code, so they are not really pictures.
One calendar for everything, sitting above your projects. If your meetings are spread across a few accounts, paste a screenshot of your week and the events get read off it — with the account they came from, plus meeting links and reminders.
You check the list before anything is added. Screenshots almost never say which time zone they are in, so it tells you when it is guessing instead of quietly moving three meetings by an hour.
Paste a week from any of your calendars.
Nothing is added until you approve it.
Invite a supervisor or a teammate to a single project and give them a role — owner, admin, editor or viewer. They see that project, and nothing else of yours.
Every table checks who is asking before it hands back a row. Your work cannot turn up in someone else's account, even if something in the app goes wrong.
Choose Opus 5, Sonnet 5 or Haiku 4.5 for each question, and see what one question costs on each. A quick question can use a cheap model.
The same tool, laid out for a small screen. Nothing here is desktop-only, so you can read a note or tick off a task on the way somewhere.
Dark and light are both properly designed, not one flipped into the other. Chart colours are checked for colour blindness, and nothing relies on colour alone.
Your reasons, your notes, and where each summary came from. Six months later the hard question is never what you decided — it is why, and that is the part usually lost.
I am doing a master's on how robots explain themselves, and I run projects alongside it. Both were spread over a reference manager, a few spreadsheets, a notes app and several calendars — and the thinking behind my own decisions kept falling into the gaps between them.
So this is not a product chasing a market. It is the tool I use every day, which is why the annoying parts got fixed first. The reason box exists because I could not remember why I had dropped a paper. The tick-before-you-add step exists because one careless search filled my list with papers from entirely the wrong field.
— Muhammad Abdul Aliabdulalidev.com
If you would like an account, send me a note and say what you would use it for. If you already have one, sign in below.