Revisenda — the guide

Revisenda (formerly Research Colab; pronounced re·vi·sen·da, Latin — the things to be revisited) is a live writing workspace for multi-author papers. Collaborate live. Catch up anytime. You write LaTeX together in real time, and the project remembers what changed, who changed it, and why.

Home page

After signing in you land on Home. It has two tabs. "My Research" lists your projects, most recently active first — typing, a discussion, a decision, a version, by you or by any co-author, moves a project to the top, like Overleaf's list; each card shows a chip counting updates since your last visit, the project's latest ⚑ milestone (its current stage — "Submitted to JASA" — however old), a few recent happenings from the last week, and one main button ("Catch me up" when there's news, "Open manuscript" when there isn't). "Projects" is a status board: one line per project — review progress, open discussions, who work is waiting on, and how recently anything happened. "Needs your word" at the top collects everything across projects that waits on you personally: proposals to review, suggested decisions to confirm, tasks, and reviewer comments assigned to you. New accounts see a "New here? Three things to try" card until they dismiss it; the sample project also contains a "👋 Start here" discussion that walks through the five core gestures. The "New project" button creates a project; "Import project" brings one in from a zip file or from Overleaf (via its Git address) — or simply drag a project zip anywhere onto the home page and drop it.

Finding a project: the search box under the header (press "/" from anywhere on Home) filters titles as you type and, after a moment, also searches the text of every manuscript, discussion, message and decision you can open — so a phrase you remember writing, a co-author's name in a thread, or a figure name all work. Each result links straight to the place: a manuscript hit opens the file at that line, a discussion hit opens the thread. If the memory is vaguer than a word ("the paper with Dong about variance… last spring?"), press "✦ Let the AI look": the workspace's AI reads a short card of each of your projects and names up to three that fit, with the reason. It only runs when you press the button, never as you type; Esc clears the search.

The workspace (Manuscript page)

The Manuscript page is where you write. Left to right: the FILES list (with FIGURES below it), the LaTeX editor, and the compiled PDF. The left menu navigates between project pages: Catch Up, Overview, Manuscript, Discussions, Review, Decisions, History, and More (Proposals, Tasks, Activity, Members). On a laptop narrower than about 1366 px the menu opens as an icon rail and the FILES panel starts folded, so the manuscript gets the room; the research-discussion panel starts closed on screens under 1600 px and opens itself the moment you press Discuss or open a thread. Expand or fold anything and it stays that way on this browser. Panels are resizable — drag the dividers — and collapsible: click the chevrons, or drag a divider all the way. Double-click a divider to reset it. On a phone the menu is a tab bar along the bottom (More holds the rest), the Manuscript page shows the PDF or the editor one at a time, and the discussion opens as a full-screen sheet from the speech-bubble button. The "Aa" menu above the editor gathers the editor preferences: text size (A− / A+), spell-check, formula preview, the dark editor theme, and Vi keybindings — each remembered on this computer, affecting only you. The PDF's page count, zoom (− / +), and open-in-new-tab (↗) live in the same toolbar whenever the PDF is showing.

The editor helps as you type. Brackets close themselves: typing { [ ( or $ inserts the matching closer with the cursor inside, [ and ( pair with ] and ), and left( brings its ight). Commands that take several arguments lay them all out — extcolor{ becomes extcolor{}{}, rac{ becomes rac{}{} — and typing } steps over an existing closer instead of doubling it, while { just before a group hops into it. Press Enter at the end of a egin{...} line and the matching end{...} is added underneath (only when one is actually missing). Backspace inside an empty pair removes both halves; typing an opening bracket with text selected wraps the selection. The bracket matching your cursor's is highlighted so nesting stays legible. Autocomplete is also built in and works like Overleaf's: suggestions appear as you type, and Enter (or a click) accepts one complete with its argument slots — egin gives egin{} with the cursor inside and the environment list already open, rac gives rac{}{} with Tab hopping to the next slot, and the egin menu also offers whole blocks like "egin{align} …" that drop the matching end in one press. Inside cite{, ef{ or includegraphics{ the project's own bibliography keys, labels, and figure files are suggested automatically.

The editor shows the paper's structure. Fold arrows in the gutter collapse any section, subsection, or environment (theorem, proof, figure…) — click the arrow, or the "…" placeholder to expand again; Ctrl/Cmd-Alt-[ folds everything, Ctrl/Cmd-Alt-] unfolds. Heading lines carry a soft background tint by level (section, subsection, subsubsection each their own), and theorem-like blocks — theorem, lemma, corollary, proof, plus anything the document defines with \newtheorem — get a faint wash with a thin left stripe, so you always know what kind of text you're in.

Sharing outside the team: the project owner can create a read-only link (Invite → Read-only link) that anyone can open without an account — they see the compiled PDF and the source, nothing else: no editing, no discussions, no history. Replace the link to kill the old one instantly, or turn it off entirely. Good for showing a draft to an advisor or committee member who isn't on the project.

Editor comfort: open the "Aa" menu above the editor to switch the editor pane to a dark theme or turn on Vi keybindings — both remembered on this computer.

The word count lives in the editor toolbar ("123 w") — live as you type, and LaTeX-aware: commands, math, and comments aren't counted, while headings, captions, and footnotes are. Click it for this file and the whole project. The OUTLINE panel in the left sidebar (under FIGURES) lists the open file's sections and subsections as a tree — click any heading to jump there, and the heading your cursor is under stays highlighted.

Formulas preview live: put the cursor inside any math — $\dots \dots$, […], or an equation/align environment — and a floating card above it shows the formula rendered, updating as you type, before any compile. It honors the paper's own \newcommand and \DeclareMathOperator definitions. While an edit is half-finished the card keeps the last good rendering, slightly dimmed. The "Formula preview" switch in the Aa menu turns it off and on; the compiled PDF remains the authoritative rendering.

Spell-check is on by default (the Aa menu above the editor turns it off and on). It understands LaTeX: commands, math, citation keys, labels, package names and file names are never marked — only actual prose gets the red wavy underline. Right-click an underlined word for suggestions, or choose "Add to my dictionary" for field terms; added words follow your account across all projects. Six dictionaries are available — English (US), English (UK), Deutsch, Français, Español and Português — chosen with the "Dictionary" selector that appears under Spell-check in the Aa menu (remembered on this computer; a dictionary downloads the first time you pick it). German compound words are accepted when they split into two known parts; a rare long compound that still gets marked can be added to your dictionary with one click. Chinese has no spell-checker of this kind — Chinese "typos" are wrong characters, not misspelled letters, so the checker simply stays quiet on Chinese text. It works in .tex, .txt and .md files, runs entirely in your browser, and uses no AI allowance.

Live editing

Everyone on a project edits the same manuscript at the same time, like Overleaf: you see collaborators' cursors with their names, and their typing appears as they type. The toolbar chip says "● Live · N here" — N is how many people are in the project right now; rest your pointer on it to see who, with the file each person has open. Next to the file name the status reads "Live · saving…" for the second or two while your latest keystrokes are being written to the project, then "Live · saved" once the server has them — it says "saved" only when the writing is actually done, not merely when you are connected. There is no save button anywhere.

Losing the connection is safe. Keep writing: your edits are kept on this device and sent the moment the server is reachable again — the chip says "● Offline" (no network) or "● Reconnecting…" and the status next to the file name says "Saved on this device". Closing the lid, changing Wi-Fi at the office, even closing the tab before the connection returns: the next time you open the project on the same browser, what you typed is merged with whatever co-authors did meanwhile — nothing is dropped and nothing is duplicated. Only a hard refresh while still offline shows nothing (the page itself can't load); the text is still there and appears once you're online. If the network works but the live connection is blocked (some campus or hospital networks block WebSockets), the chip reads "● Live · slow link": edits travel every few seconds instead of instantly, and everything else works. Compile always sends your latest text to the server first, so the PDF matches what you see; when you're fully offline it says so instead of building an old version. Note for shared computers: the copy lives in that browser — use a private window on a computer that isn't yours.

Versions and checkpoints

You never save versions by hand. The manuscript checkpoints itself: after about 10 quiet minutes, when the last person closes the project, and at least every 30 minutes during a long session. Each checkpoint appears in History with an automatic description of what changed and who edited. Checkpoints are quiet bookkeeping — they don't clutter Catch Up or the home page, which speak in human events instead: who edited what, comments, decisions, and milestones. To name a moment yourself ("Submitted to JASA", "Draft sent to coauthors"), open History and press "⚑ Create milestone" — the name appears as a violet flag on that version. Any version can be restored from History; restoring never deletes anything, it creates a new version with the old content, and everyone's screen updates immediately.

History and comparing versions

The History page lists every version: who, when, what changed, and why — with links to the discussion or proposal a change came from. Compare any two versions with the Compare panel; a header above the diff lists who changed what across that span, and "just this change" narrows to a single step. Naming a version: press the ⚑ on any version row (or "⚑ Create milestone" for right now). Downloads from History: "⬇ Full history (.zip)" contains every version plus a changelog; each version row also has its own "zip".

Downloads and exports

To download the compiled paper as PDF: open the Manuscript page, and in the PDF pane press the ↗ button (top right of the PDF) to open it in a browser tab, where you can save or print it. If there is no PDF yet, press "▶ Compile" first. To download the whole project as a zip: press the ⬇ at the top of the FILES panel, or ask the assistant to fetch it. To download an older version, use its "zip" link on the History page. The full history zip (every version + changelog) is the "⬇ Full history (.zip)" button on History. You can also ask the assistant for any of these and it will start the download for you.

Compiling and the PDF

Press "Compile" at the right end of the toolbar (or Ctrl/Cmd-S) to build the PDF — that end of the toolbar is pinned, so Compile and the view switch never scroll out of sight however narrow the window. Editor and PDF sit side by side in Split view; the Editor / Split / PDF switch changes layouts. The bar above the PDF shows how long the last compile took (e.g. "compiled in 3.2 s"); the first compile of a session is the slowest and later ones are much faster. Recompiling keeps your place — the PDF stays at the page you were reading instead of jumping back to page 1.

Which file gets compiled: the project's main file is detected from the LaTeX itself — the .tex that declares documentclass and egin{document} and pulls the others in with input or include; an old draft in an old/ folder or a standalone figure file is never mistaken for it, and a "% !TeX root = …" comment settles it outright. Once detected it is remembered for everyone on the project; the Compile button's tooltip names it, the FILES list marks it with a "main" chip (press ★ on another .tex to change it), and opening the project lands on it — never on a README or checklist just because it sorts first. After that, the project opens on whichever file you had open last in that browser. Opening a project that has no PDF yet compiles it for you, so the PDF is there when you arrive. Under the Aa preferences there is an "Auto-compile" switch: on, the PDF rebuilds itself a few seconds after you (or a co-author) stop typing — off by default.

The standard LaTeX package collections are installed — including journal and publisher classes, TikZ, biblatex with biber, fonts, Chinese/Japanese/Korean, and the major European languages — and university templates work by bringing their own class files with the project. pdfLaTeX, XeLaTeX, and LuaLaTeX are all supported: the right engine is picked automatically, and a "% !TeX program = xelatex" (or lualatex/pdflatex) line at the top of the main file forces one. Bibliographies, indexes, and cross-references resolve automatically across passes. Not supported: the minted package (use listings instead). If a compile reports a missing package, report it via Help — missing packages can usually be added quickly. The ⇄ button on the bar between the editor and the PDF jumps from the line you're editing to the same spot in the PDF (double-clicking in the editor does the same); double-clicking anywhere in the PDF jumps back to the matching source line. Text in the PDF is selectable — drag to select and copy it like any document (double-click still jumps to the source). The bar itself still drags to resize and double-clicks to a 50/50 split. The page count, compile time, − / + zoom, and ↗ open-in-new-tab for the PDF sit in the main toolbar whenever the PDF is visible (zoom is remembered on this computer). If the compile fails, a red "Compile failed" panel appears under the PDF: the first error in plain words, a link that jumps to its line, the list of all errors (click one to jump there), and "raw log" for the full TeX output. "✦ Fix with AI" on that panel asks the workspace's AI to diagnose the failure and propose the smallest edits that make it compile; each proposed edit is shown as before/after text with a checkbox, and nothing changes until you press Apply. Applied fixes go into your own working copy only (never into the shared Current version) and recompile automatically; if a new error appears, press "✦ Fix with AI" again. The AI is told never to rewrite prose or change mathematical meaning — read the before/after anyway, as you would a co-author's edit. Missing packages on the server and missing figure files cannot be fixed by an edit; the diagnosis says so and points you to Help or to the Figures upload.

Files and figures

The FILES panel lists the project's .tex and .bib files. + creates a file (also .sty, .cls, .txt, .md); the ⬇ downloads the project as a zip. Deleting a file removes it for everyone (live projects) and needs a confirmation. File and folder names can't start with a hyphen ("-") or contain ".." — those read as commands to the LaTeX toolchain, so the workspace refuses them; rename with a letter first (e.g. "s-intro.tex"). Figures (images, PDFs) live in the FIGURES panel below — drop files anywhere on the workspace to upload, then reference them with \includegraphics{name}. Import references: the 📚 button in the FILES header brings a Zotero collection (or the whole library), or pasted BibTeX from any reference manager, into a .bib file in the project. Zotero needs a one-time connection under Account → Reference manager (Zotero user ID + a read-only API key from zotero.org/settings/keys). Importing into an existing .bib adds only entries whose citation keys aren't there yet, so existing \cite{} keys keep working; "Replace the whole file" is the alternative. New keys appear in \cite{ autocomplete at once.

Catch Up

"Catch me up" (top right on every project page, or Catch Up in the left menu) answers "what did I miss?". If nothing needs you, it says so and steps aside. Otherwise it shows, in order: Needs your input (proposals, decisions, tasks, reviewer items), Open questions (unanswered discussions, however old), and What changed — a short AI briefing where every claim links to the event it came from. Catch Up never counts something as read until it has actually been shown to you. A row of window buttons — since my last visit, 1 day, 7 days, 1 month, 3 months, 6 months, 12 months — re-runs the same briefing over any period; the 12-month view reads like the skeleton of an annual or grant progress report, built from the project's actual records. Very busy windows show the 120 most recent events and say so. If the workspace's AI does not answer just now (a hiccup on its side), the What-changed part is written by the built-in summarizer instead and the page says so; the next visit asks the AI again — nothing is cached on that path.

AI allowance

The ✦ Ask panels show a small line — "412 of 500 AI actions left this month" — whenever the workspace sets a monthly allowance; it turns amber under 20 % left, and the same meter is on the Account page. Actions reset on the 1st. When the allowance is used up, the free built-in helpers take over until then; press Help and ask if you need more.

Discussions (threads)

Discussions attach scientific conversation to the manuscript. Select any text in the editor and press Discuss to open a thread anchored to that passage (the anchor stays attached as the text moves); or start a general thread from the DISCUSSION panel. You don't have to name a thread: write your message and the first sentence becomes the title (when the workspace has AI, a sharper short title replaces it a few seconds later); a thread started from a highlighted passage with no message is titled with the passage itself. To rename any thread, click its title in the thread header, edit, press Enter. Threads have types — question, method, result, writing — and live until someone resolves them. Inside a thread, hover any message and press "↩ Reply" to quote it (like WeChat or Slack): your reply carries a small quote of the original, and clicking that quote jumps back to the original message. The × on the composer's quote chip cancels the quote without losing your text. Drop a file anywhere on a thread — or simply paste a screenshot from your clipboard (Ctrl/Cmd-V) — to attach it: images appear right in the conversation, other files as a 📎 link, and everything is also kept in the thread's Evidence & artifacts list (25 MB per file). Pasted screenshots are auto-named with a timestamp. The 🙂 button opens an emoji palette; a message that is only an emoji or two shows big, WeChat-style. Co-authors writing in another language? Hover a message and press "Translate" to see it in yours, right under the original (the original is never replaced), or press "🌐 Translate all" in the thread toolbar to do the whole thread in one go — messages already in your language are left alone. Set your language under Account → Translation (default English). Each translation is done once and then shown free to everyone reading in that language; "Translate all" costs one AI action for the whole thread. Resolving asks "What did you decide?": the answer is saved as a Decision, so the reasoning is findable months later. Reopening a resolved thread keeps the old resolution visible.

Decisions

The Decisions page (left menu) is the project's memory of settled questions: what was decided, why, who confirmed it, and the discussion it came from. The workspace's AI can suggest a decision from a converged discussion, but nothing becomes a Decision until a human confirms it. Decisions are usually born from discussions — resolving a thread asks "What did you decide?" and saves the answer — but for a decision made outside the app (a meeting, a call, an email), press "+ Record a decision" at the top of the Decisions page: it is saved as confirmed and attributed to you. You can also just tell the assistant (Help) — "record a decision: …" — and it drafts the card for you to confirm.

Editing a confirmed decision: if you confirmed it, fixing your own wording leaves it confirmed. If someone else changes what a confirmed decision says, it moves to "Amended — needs reconfirmation": the previous wording and who confirmed it are kept, the change shows up in Activity and Catch Up, and the decision waits for a human to reconfirm the new wording — so a settled conclusion is never quietly attributed to someone who didn't agree to it.

Proposals

A proposal is a reviewable bundle of changes: a diff plus an explanation. The workspace's AI always goes through proposals — nothing it writes touches the manuscript without a human pressing Accept. Accepting applies the changes to the live manuscript for everyone; rejecting records why.

Review rounds (reviewer feedback)

On the Review page, paste a decision letter and the workspace's AI splits it into individual reviewer comments; each becomes a tracked item you can assign, discuss, and resolve. If the split got something wrong, press "Correct wording" on a comment to fix or delete it. Write your reply in "Your response"; "✦ Draft response" offers an AI suggestion in a separate box — "Use this" adds it to your draft, it never overwrites what you have written, and if a co-author saved a different response while you were typing, their version appears beside yours and you choose — "Use theirs" or "Save mine over theirs" (the other text is kept in the history either way). The page shows progress ("6/9 resolved"), and "Draft response letter" assembles a reply from what actually changed in the manuscript — you edit and approve every word before it leaves the building.

Tasks

Tasks (left menu → More) are lightweight to-dos with an assignee, often created from a discussion or a reviewer comment. Your open tasks appear in "Needs your word" on Home and in Catch Up.

Inviting people

Press "Invite" (top right). Start typing a name or an address and people you already share a project with are suggested — pick one with ↑/↓ and Enter; nobody outside your projects is ever listed. Someone who already has an account joins the moment you add their email; otherwise they get an invitation link. Everyone on a project can read and edit everything in it; only the project owner can rename or delete the project. Renaming: double-click the title in the header. Deleting a project — three places, same owner-only rule: (1) the ⋯ button on the project's row in the "Projects" tab on Home → "Delete project…", (2) inside the project, Members (left menu → More) → the "Danger zone" card at the bottom, or (3) tell the assistant (Help) "delete this project" — it opens the same confirmation card. Either way you type the project's title to confirm, because deletion removes the manuscript, versions, threads, and decisions for every member and cannot be undone. If you don't see a Delete option, you are not this project's owner. (Note: the Danger zone on your Account page is different — that one deletes your account, not a project.)

Account, AI allowance, and passwords

Creating an account takes a name, an email address, and a password typed twice (the form tells you at once if the two don't match). You are signed in the moment you press the button — no waiting for an email. We do still ask you to confirm the address: a strip at the top of the page says so until you click the link we sent. Writing, compiling, importing, threads, and decisions all work before the click. Three things wait for it, because they reach other people or spend the workspace's money: sending invitations, creating a read-only share link, and the workspace's AI (the built-in helpers keep working). If you try one of them early, the dialog says so and offers to resend the email right there. Sign-in itself needs the confirmation within seven days; after that, the sign-in page offers a fresh link. Mistyped the address? Change it under Account and press Resend. Accounts created with Google (or a personal Microsoft account) arrive confirmed and skip all of this; ORCID and university Microsoft sign-ups confirm their address by link in the same way. An invitation link opens only for the address it was sent to: register or sign in with that address to accept it, or ask the owner to invite the address you use. Clicking the link also confirms that address.

If the sign-in page shows "Continue with Google", "Continue with Microsoft", or "Continue with ORCID", you can sign in with those instead of a password — same account either way. Google matches by verified email automatically. A university/work Microsoft account that already has an account here connects under Account → Linked sign-ins after a password sign-in (a safety rule for organizational email). ORCID doesn't share your email, so its first sign-in asks for one and confirms it by link; from then on it's one press. The Account page's "Linked sign-ins" section shows what's connected and lets you connect more. Click your initials in the top-right corner of the home page to open Account. It shows your details, the Reference manager section (connect Zotero with your user ID and a read-only API key; Disconnect forgets it), and this month's AI allowance: how many AI actions you've used (each Catch Up, suggestion, fix, or assistant question is one action) and the safety-net dollar meter. Change your password there too. When you've used about 80% of the month's allowance, the meter turns amber and Catch Up shows a heads-up with your exact count, so the limit never lands as a surprise. If the allowance does run out, AI features fall back to simpler built-in summaries and say so — nothing breaks, and everything resets on the 1st. Need more before then? Press Help and ask — during the free preview, limits are raised freely. A project can also use its own API key (project Settings → AI), which never counts against your allowance.

Turning the AI off for a paper, or bringing your own key

Each paper has an AI mode in Settings → AI (owner's choice): the workspace default, Off, the workspace's AI, or your own key. With AI off, nothing from that paper is ever sent to a model: the paper still remembers everything and still briefs you; it just won't write anything for you. Catch Up keeps its "needs your input", open questions and evidence sections unchanged and shows "what changed" as a factual summary instead of prose; reviewer-letter import splits by numbering; compile help uses the built-in fixes; Ask the paper, AI edit suggestions, the polished response letter, suggested decisions, image-to-LaTeX and translation are unavailable. With your own key, every AI feature in that paper — Catch Up included — runs on the provider and model you chose (Anthropic, OpenAI, Moonshot, DeepSeek, or any OpenAI-compatible endpoint), billed by that provider to you, never counted against the workspace allowance, with an optional monthly budget cap you set; the key is stored encrypted and is never shown again. The plan price does not change with the AI mode.

If your journal asks authors to declare AI assistance: response letters and suggestions here are drafted from your project's record and you edit every word, but they are still machine-drafted text — declare it if the journal's policy asks you to, or switch the AI off for that paper.

Forgot your password: on the sign-in page press "Forgot your password?", enter the address your account uses, and a reset link arrives from no-reply@revisenda.com; it works once and expires in an hour. The page says the same thing whether or not the address has an account, on purpose. If nothing arrives within a couple of minutes: check spam (a new sender often lands there the first time), make sure you typed the address the account was created with (an invitation may have gone to a different one), and if you signed up with Google, Microsoft or ORCID you can simply use that button — no password needed. Still stuck? Press Help and write to the admin, who can re-send the link or hand you one directly.

Email notifications: being assigned a task or reviewer comment emails you right away. New discussions and replies are bundled into one email per project, about twenty minutes after the first message, and only if you haven't been in the app since — if you already saw it, no email comes. Ordinary edits never trigger email. The Account page has one switch that turns all notification email off or on. Separately, the workspace admin can occasionally send an announcement to all members (what changed, planned downtime); routine announcements respect that same switch, and only essential notices — an outage, a policy change — go to everyone.

Ask the paper (manuscript AI)

"✦ Ask the paper" is about the manuscript and the project's record — the same feature on Overview ("Ask this project anything"), on Catch Up, and in the manuscript's discussion panel ("✦ Ask the paper"); it is different from Help (top right), which answers questions about the product. Open it from the discussion panel's header, or select text and press "✦ Ask" in the small tooltip. Ask a question about the file you're looking at, or ask for a change ("tighten this paragraph", "turn this list into a table"); the selection, if any, is the focus. Answers are prose. Changes come back as suggestions IN the editor: the old text struck through in red, the replacement in green beneath it, with Accept and Reject on the card. Nothing changes until you press Accept; accepting is an ordinary edit (collaborators see it, Ctrl-Z undoes it). Ctrl/Cmd-Enter accepts the suggestion nearest the cursor, Ctrl/Cmd-Backspace rejects it; the panel offers Accept all / Reject all for the file. A suggestion that changes the words a reader would read (not just LaTeX or math) is flagged "changes wording — read carefully". The AI is told never to invent citations or results; when it can't know something it says so and leaves a TODO placeholder. Equation from image: the 📷 button in the same panel (or pasting an image into its box) turns a photo, screenshot or whiteboard snapshot of an equation, table or diagram into LaTeX, shown with "Insert at cursor" and Copy — check it against the picture. The compile error panel's "✦ Fix with AI" also offers "Review in editor", which shows each fix the same inline way. Every question, image or fix is one AI action against the monthly allowance.

Help (the assistant)

The Help button (top right, on project pages) opens the assistant. It answers questions about the product and this project's state, and can fetch downloads for you — "download the PDF", "get me the v12 zip", "take me to History". It answers only from this guide and your project's real state; when it doesn't know, it says so and points you to the guide. The line under its composer — "Write to the maker" — opens the same message box as before: a bug, a question, something missing, sent straight to the person who runs the workspace, with the page you're on attached automatically. If you tell it a decision — "record a decision: we use the 3:1 allocation" — it drafts a decision card from your words; you review it and press "Record decision", and nothing is saved unless you do. If you're the project's owner and tell it "delete this project", it opens the same permanent-deletion confirmation used everywhere else: you must type the project's exact title, and nothing is deleted unless you do. It never edits the manuscript and never changes anything on its own — downloads, navigation, and confirmation cards that only you can complete. Questions you ask may be reviewed by the workspace admin to improve the product; don't paste secrets into it.

The name

Revisenda was called Research Colab until September 2026. Nothing else changed: same accounts, same projects, same co-authors. Links to the old address (researchcolab.app) redirect to revisenda.com; you sign in once more on the new address because sign-ins are tied to the domain. If you had installed it to your dock, remove it and add it again to get the new icon. Emails now come from no-reply@revisenda.com. The word is Latin (re·vi·sen·da) and means "the things to be revisited" — the same family as agenda and addenda.

Install it like an app

Revisenda can live in your dock without any download: Safari on macOS → File → Add to Dock; Chrome or Edge → the Install icon in the address bar. You get an app icon and its own window. It keeps itself current: when a new version is deployed, the app notices the next time you return to it and reloads (in the manuscript workspace it shows a small "Reload now" bar instead, so you choose the moment — your text is already saved either way). No need to quit and reopen it. There is no separate desktop download — the web app is the app.

When something is wrong

Press "Help" in the header to report a problem or contact the team — a human reads every report. If the page itself breaks, reload once; your writing is safe on the server. For anything the assistant can't answer, Help is always the escalation.