{"overview":"overview origami is a terminal manager that keeps each project's agents, shells, servers and pages together in one window. get started how it works a workspace is a project folder and what runs in it. its tabs sit in the sidebar, in four sections: 1. agents : claude code, codex or your own coding agent, each in a tab of its own. see agents . 2. commands : dev servers and scripts, started and stopped with one key. see commands . 3. terminals : shells that open in the folder. see terminals . 4. browsers : websites beside your code, with shared or separate logins. see browsers . over the tabs sit its loops : conversations in which origami drives your own agent through one piece of work. each workspace also has its git changes and history to review and commit, a kanban board, and the artifacts your agents hand back: plans, reports, diagrams and screenshots. sidetrack keeps a second tab in view beside the one you work in. agents can work origami too: open tabs, run commands, write artifacts, move cards and reply to review comments, all where you can see it. a loop has these tools built in. an agent in a terminal tab or another app reaches them through the mcp server . good to know - origami is free for up to 5 workspaces with 10 tabs each. origami+ removes both limits. - when origami opens again, every tab comes back stopped, except those set to start on launch.","install":"install download origami, open it, and point it at your first project folder. get started how it works 1. go to tryorigami.app/download and press download free . origami needs macos 14 or later. 2. on linux, press choose your build and pick an appimage or a .deb, for x86 64 or arm64. linux builds are experimental. 3. open origami. it starts on its home page, with new workspace in the sidebar. 4. press choose a folder and pick a project. that folder becomes your first workspace . good to know - origami looks for a new version when it opens and every minute after. when one is out, a cloud button shows in the titlebar: update available! . press it and choose update , and origami installs it and restarts. see updates . - browser tabs are not available on linux. - the free version has every feature, for up to 5 workspaces with 10 tabs each. origami+ lifts the limits.","the-window":"the window the rail, the sidebar, the stage and the titlebar, and what each one is for. get started how it works 1. the rail is the narrow column on the left: one tile for each workspace, then a + to add one. press a tile to go to that workspace, on the tab you left it on. at its foot are the origami mark, which opens the home page, then trail , usage and settings . 2. the sidebar holds the open workspace's loops , then its tabs in the agents, commands, terminals and browsers sections. press the workspace's name at its head for its folder and editor actions. ⌘b collapses it. 3. the stage is the main area, where the tab you picked runs from edge to edge. 4. the titlebar says where you are as a path: the workspace, the view, then what is open in it. on its right are the views ( stage , explorer , git , kanban and artifacts ), the bell for notifications , and the tiles that show or hide the tray and sidetrack . 5. the home page is your first screen: origami's edition, its version, and a row for each thing it can do, each opening a page about it. good to know - views such as git or settings turn the sidebar into their own list. its back button returns to your tabs. - a long path gives way to the buttons, which never move or fold into a menu.","workspaces":"workspaces a workspace is a project folder and everything that runs in it. workspaces how it works 1. press + under the workspace tiles on the rail and pick a folder. the workspace takes the folder's name and opens with its sections empty. 2. right-click its tile and choose edit to change its name, its appearance (initials, an emoji, an icon or an image), its colour, its working directory and the environment variables every tab in it starts with. 3. press a tile to go to that workspace. ⌥⇧↓ and ⌥⇧↑ step to the next and previous one, and tiles are dragged up and down the rail to reorder them. 4. in the commands section's menu, start all auto-startable starts every command set to start on launch, and stop all auto-startable stops them again. good to know - origami free holds 5 workspaces with 10 tabs each, not counting loops. origami+ has no limit on either. - settings › workspaces lists every workspace, with what each holds and the same edit and remove actions. - settings › general › sidebar can hide the agents, commands, terminals or browsers section. a hidden section's tabs are still reached from the command palette and sidetrack. - removing a workspace asks first. it removes the workspace from origami, never the folder on disk.","tray":"tray throwaway terminals along the foot of the window, for a quick command without adding a tab. workspaces how it works 1. press ⌘j, or show tray at the titlebar's right end, beside show sidetrack . the tray opens under the stage and sidetrack with a terminal already running in the workspace's folder, with its environment variables. 2. press + after the chips for another. each terminal is a chip, terminal 1 , terminal 2 and so on; press a chip to show it. 3. press ⌘⌥↓ to move the keyboard into the tray, opening it if it is hidden, and ⌘⌥↑ to give it back to the stage. clicking either side works too. 4. point at a chip's icon and press its ×, or press ⌘w while the keyboard is in the tray, to close a terminal. if something is still running in it, origami asks first. closing the last one hides the tray. good to know - its terminals are not tabs: they have no sidebar row, do not count toward the tab limit, and are gone when you quit. - each workspace has its own. hiding the tray or switching workspace leaves them running. - drag its top edge to make it taller or shorter, and drag chips to reorder them. - copy reference on a chip's menu gives an agent the terminal to read and type into. only you can close it.","sidetrack":"sidetrack a panel on the right that keeps a tab in view beside the one you work in. workspaces how it works 1. press ⌘d, or show sidetrack at the titlebar's right end. it pushes the stage aside rather than covering it. 2. right-click a tab and choose open in sidetrack , or press + in sidetrack's head and pick one. each tab you add is a chip in the head; press a chip to show it. 3. the same list holds its widgets : the scratchpad ; tasks , the open loop 's plan, subagents and background commands; and pull requests , the repository's pull requests with their checks and reviews. 4. press ⌘⌥← or ⌘⌥→ to move the keyboard between the stage and sidetrack, or click the side you want to type into. good to know - drag chips to reorder them; esc while dragging puts them back. drag sidetrack's edge to make it wider or narrower. - a tab is drawn in one place at a time. while the tab sidetrack shows is also the one on the stage, the stage has it, and sidetrack gets it back when you switch tabs. - what it holds is saved with each workspace; whether it is open follows you from one workspace to the next. - the plan's tab on a loop's message box opens tasks . what it lists is kept with each loop and comes back after a restart.","files":"explorer the workspace's files in a view of their own, to find, read and change them. workspaces how it works 1. press ⌘⇧e, or explorer in the titlebar. the sidebar shows the folder tree under filter files , and search beside it. with no file open, the view lists your recent files. 2. files also open here from a loop's file badges, open in files in git or on an artifact, a path ⌘-clicked in a terminal, the command palette 's files , or an agent. 3. each open file is a tab at the head. ⌘w closes the one showing, and ⌘⇧t reopens the last one closed. the count at the row's end lists them, with open with and save all . 4. press ⌘s to save; nothing is saved otherwise. right-click selected lines for add to loop , which hands them to a loop with their line range. 5. right-click the tree for new file , new folder , rename , duplicate and remove , which moves it to the trash. good to know - if a file you are editing changes on disk, it says changed on disk : reload takes the new one, and keep mine keeps your edits. - a file outside the workspace's folders is read-only until you press allow editing . pictures and pdfs are drawn, and films and sounds play. other binary or oversized files offer show in folder instead. - quitting, or removing the workspace, asks before unsaved edits are lost.","editor":"editor where a file opened in the explorer is read and changed, with vs code's keys. workspaces how it works 1. ⌘d adds the next occurrence of the selection, or of the word at the cursor, as another cursor. ⇧⌘l takes every occurrence, ⌥⌘↑ and ⌥⌘↓ add a cursor above or below, and ⌥-click adds one where you click. esc goes back to one cursor. 2. ⌥↑ and ⌥↓ move a line, ⇧⌥↑ and ⇧⌥↓ copy it, ⇧⌘k deletes it and ⌘/ comments it out. 3. ⌘f finds in the file and ⌥⌘f finds and replaces, with match case , whole word and regex . ⌃g goes to a line, or to line:column. 4. ⌥⌘ and ⌥⌘ fold and unfold, ⌥z turns word wrap on or off, and ⌃space completes a word. 5. the status line at the bottom right shows the cursor's line and column, the language, the indentation and the line endings. press the position to go to a line, or one of the others to change it for this file while it is open. good to know - settings › editor sets what saving does, indentation, word wrap, whitespace, font size, closing brackets, word completion, and the files hidden from the explorer and search. - a repository's .editorconfig goes over those settings for indentation, line endings, the final newline and trailing whitespace. - while a file has the keyboard, ⌘d and ⌘⏎ are the editor's, not sidetrack's or the tab's. the shortcuts dialog lists every key under editor .","search":"search find text across the workspace's files, and replace it in all of them at once. workspaces how it works 1. press ⌘⇧f, or search at the head of the explorer's sidebar. a word or line you have selected, or the word at the cursor in a file, fills the field. 2. the tiles in the field match case ( aa ), whole words ( ab ) and a pattern ( .\\ ). with aa off, a query with a capital in it still matches case. 3. include and exclude narrows it with globs separated by commas. files your ignore files leave out are skipped unless you turn on include ignored files . 4. press a file to unfold its matches, and a match to open the file at it. 5. ⌘⇧h opens the same page with replace unfolded. each match shows what it becomes, and leave out skips a file or a match. replace all changes the rest. good to know - a file open with unsaved edits is searched and changed as the editor holds it, and stays unsaved. - a file changed on disk since the search is skipped, and show skipped on the toast says why. undo on the toast puts each file back, unless it has changed since. - files over 1 mb, binary files and the files hidden in settings › editor are not searched. results stop at 20,000.","scratchpad":"scratchpad a page of notes for each workspace, kept in sidetrack beside your work. workspaces how it works 1. open sidetrack , press + and pick scratchpad under widgets . 2. type. markdown is drawn as you write it: a heading, a checkbox to tick or a block of code. 3. what you type is saved with the workspace as you go, so each workspace has its own pad. 4. ask an agent to note something down or read your notes, and it uses the pad through the mcp server. good to know - an agent can add to the pad or change part of it, but never replace or empty all of it. - anything you typed is saved before an agent's edit is applied, and the pad on screen shows the change at once. - every agent edit is a scratchpad edited row in the agent trail , with a button that opens the pad. - the scratchpad is your own notes, so it is left out when workspaces are exported.","selection":"selection pick several rows of the sidebar at once and start, stop or remove them together. workspaces how it works 1. ⌘-click a tab's row to pick it, and ⌘-click it again to let it go. 2. ⇧-click a row to pick every row from the last one you picked, or from the tab that is showing, down to it. 3. while rows are picked, the foot of the sidebar says how many are selected, with buttons to start those not running, stop those that are, remove them all, or clear the selection. only what applies is there. 4. right-click one of the picked rows for a menu that acts on all of them. good to know - esc, a plain click on a row, another workspace or another view lets the selection go. - removing several rows asks once for all of them. - loops are picked the same way, among themselves: the foot copies their references, closes them or removes them. - the same works in the artifacts list, where ⌘a picks every row and the foot can also group them. see artifacts . - a right-click on a row that is not picked is that row's own menu, and lets the selection go.","agents":"agents a coding agent running in a tab of its own, in the workspace's folder. tabs how it works 1. press the agents section's name and choose add existing agent , then pick one. its tab is added stopped. 2. press ⌘⏎, or start on its page, and the agent starts in the workspace's folder. ⌘⇧⏎ stops it. 3. origami comes with claude code, codex, copilot, opencode, antigravity, grok and pi built in. each starts the agent's own command line tool, so it needs to be installed. 4. for an agent of your own, choose create custom agent in the same menu and give it a name and the command that starts it. good to know - each row shows its agent's mark. a custom agent that starts claude code or another known agent gets that agent's mark too. - the mark also says what the agent is doing: see agent status . - settings › agents lists every agent. built-in ones can be turned off but not removed; built-ins also have options for what they are told as they start. see agent options .","terminals":"terminals a shell in a tab, opened in the workspace's folder. tabs how it works 1. press the terminals section's name and choose add terminal . give it a name and, if you like, a launch command to run first and a working directory other than the workspace's folder. 2. the terminal starts at once, in your system shell with the workspace's environment variables. a launch command runs, then leaves you in the shell. 3. rest the pointer on a running tab and a dock slides out from its right edge: its cpu and ram , unfold, stop , and in a git folder, the worktree and branch pickers. 4. press ⌘k to clear the terminal. good to know - a stopped terminal shows its own page: what it would run, where, and start and edit . - a shell that exits keeps what it printed last on screen, with start again and hide output on its dock. - turn on auto-start on launch in its edit dialog and it starts whenever origami opens. - agents can open terminals, type into them and read what they print with origami's tools, built into loops and offered to other agents by the mcp server .","commands":"commands a dev server or script saved as a tab, started and stopped with one key. tabs how it works 1. press the commands section's name and choose add command . give it a name , the command to run, and a working directory if it is not the workspace's folder. 2. the command is added stopped. press ⌘⏎ to start the tab on the stage, ⌘⇧⏎ to stop it, or use start and stop at the end of its row. 3. turn on auto-start on launch and it starts whenever origami opens. 4. start all auto-startable in the section's menu starts every command set to auto-start, and stop all auto-startable stops them. good to know - when a command ends, what it printed stays on screen, so an error is still there to read. its dock has start again and hide output . - commands run with the workspace's environment variables. - put a server's command in sidetrack to keep its log in view while you work in another tab. - agents can create and run commands and read their output with origami's tools, built into loops and offered to other agents by the mcp server .","browsers":"browsers a real browser in a tab, beside your code, for you and your agents. tabs how it works 1. press the browsers section's name and choose add browser , then give it a name and a url . it opens at once. 2. under sharing , choose whom it shares cookies and sign-ins with: origami , every browser; workspace , this workspace's browsers; browser , none; or incognito , which keeps nothing once it stops. 3. its toolbar has back, forward, reload, home and zoom, an address bar, and developer tools. set as home makes the page you are on its start page. 4. while a browser has the keyboard, ⌘r reloads and ⌘⇧r, pressed twice, clears the site's data, cache and cookies and reloads. ⌘+ and ⌘− zoom, ⌘0 resets it, and ⌥⌘i opens developer tools. 5. agents use the same tab with origami's tools, built into loops and offered to others by the mcp server . they read the page, click, type, take screenshots, and read the console and network log. good to know - except in incognito , a login survives a restart. - removing a browser that keeps its own data forgets it, and removing a workspace forgets its browsers' shared data. - when a page needs you, for a login or a captcha, an agent hands it over: the tab shows waiting for you with the reason, and continue hands it back. - browsers are not available on linux.","command-palette":"command palette a search box for everything origami can do, with a second one for jumping between tabs. keyboard how it works 1. press ⌘⇧p to open the command palette. type to search the actions for the workspace you are in: add a tab, start or stop one, open git, the board or the artifacts, switch workspace, and more. 2. press ⌘p instead for the tab list: every tab and open loop of every workspace, under its workspace's name. pick one to go straight to it. 3. move with the arrow keys and press enter to run a row. a row whose name ends in \"...\" opens a page of its own, such as settings... or switch to workspace... , and the page you are on shows as a path before the search. 4. press backspace in an empty search to go back a page, or esc to close the palette. good to know - other parts of the app open the palette on a list of their own when it is too long for a menu: filing an artifact in a group, choosing a card's parent or an epic, finding a loop, or picking the branch a loop's worktree starts from. search it and press enter. - some can also make what you type when nothing matches, such as a new group. - files opens a file in files ; loops lists loops, closed ones too. - keyboard shortcuts opens the full list of keys.","shortcuts":"shortcuts the keys that move you round origami without the mouse. keyboard how it works 1. open the full list with keyboard shortcuts in the command palette , or keyboard shortcuts in the app's menu. 2. the keys you will use most: - ⌘p: the tab list. ⌘⇧p: the command palette. - ⌘1 to ⌘9, and ⌘0: go to tab 1 to 10 of the workspace. - ⌘⇧↓ and ⌘⇧↑: the next or previous tab. - ⌥⇧↓ and ⌥⇧↑: the next or previous workspace. - ⌘⏎ and ⌘⇧⏎: start or stop the tab you are on. - ⌘d: open or close the sidetrack . ⌘⌥← and ⌘⌥→: move the keyboard between it and the main stage. - ⌘j: show or hide the tray , the throwaway shells under the stage. - ⌘b: collapse or expand the sidebar. ⌘,: settings. - ⌘⇧j, ⌘⇧e, ⌘⇧g, ⌘⇧o and ⌘⇧a: the stage, explorer, git, the board and the artifacts. - ⌘f: find in a terminal. ⌘k: clear it. ⌘r: unfold . 3. ⌘esc takes the keyboard away from the terminal or page you are typing in. good to know - the app's shortcuts work while a terminal has the keyboard, so you never need to click out of one first. - in a browser tab, ⌘r reloads the page instead of opening unfold. - the full list also has the keys for loops, browser tabs and moving along a terminal's line.","find":"find a search bar over a terminal that finds text anywhere in its output. keyboard how it works 1. press ⌘f in a terminal, or right-click it and choose find . a bar opens over the terminal. 2. type what you are looking for. the bar counts the matches as it reads the output and shows which one you are on, such as \"3 of 12\". 3. press enter to go up to the match before, since a terminal is read from the bottom. press shift+enter to come back down. the arrows beside the field do the same, and both wrap round at the ends. 4. press esc, or the close button, to put the bar away. the keyboard goes back to the terminal. good to know - the three tiles before the field turn on regex , match case and whole word . - find searches the whole scrollback, not only what is on screen, and never holds up the terminal while it reads. - a \"+\" after the count means it is still reading, or that there are more matches than it keeps count of. - find and unfold share one place on the terminal, so opening one closes the other.","unfold":"unfold describe a command in plain words and get the shell command back on your prompt. keyboard how it works 1. choose the agent that answers in settings › agents , under unfold . only built-in agents are offered. 2. in a running terminal, press ⌘r. you can also right-click the terminal and choose unfold , or press the unfold tile on the terminal's dock. 3. type what you want in plain words, such as \"re-encode this video at half the size\", and press enter. 4. the agent answers with one shell command, which is pasted onto your prompt. read it, change it if you like, and press enter yourself to run it. good to know - the agent runs in your login shell, in the terminal's working directory and with the workspace's environment, so your paths, aliases and variables are all in play. - nothing runs until you press enter: the command only ever arrives as pasted text. - with no agent chosen yet, opening unfold takes you to settings › agents instead. - of the agent's options , unfold uses only its model and how hard it thinks. - it is for terminal tabs only, not agents or commands. esc closes the bar.","agent-status":"agent status the mark on an agent's tile that says whether it is working, waiting on you or done. agents how it works 1. each tab's tile carries a small mark on its corner. violet means something for you, green means nothing to do, and a turning ring means the agent is active: - violet ring : the agent is waiting on you, for an answer or a permission. - violet dot : it has finished and you have not seen the result yet. - green ring : it is working. - green dot : it is running with nothing for you. - red dot : the session ended with an error. start it again to clear it. 2. point at a mark to read its state in a word. 3. each other workspace's tile on the rail shows the mark of its most pressing tab, so you can see where you are needed. 4. when the sidebar shows a view's own list, such as git's, the tabs waiting on you sit as tiles at its foot. press one to go to it. good to know - built-in agents report what they are doing themselves. to turn that off for one, open its options in settings › agents and turn off report status on the origami page. its state is then guessed from how quiet its terminal is. - antigravity has to be set up once, with set up on that same page. grok reports through its terminal's title alone, with no switch.","agent-options":"agent options the flags and variables a built-in agent starts with, chosen from a list instead of typed. agents how it works 1. open settings › agents and press the options tile on a built-in agent's row. 2. the dialog has three pages. options holds the agent's flags, such as its model or how it asks for permission. environment holds the variables it reads. origami holds what origami adds to its start, such as report status . 3. each setting is a choice: a list, a switch or a short field. default leaves it out, so the agent decides. 4. press the preview command tile beside the pages to see the whole line the agent will run, and copy command to copy it. 5. press save . every tab of that agent uses the options from its next start. reset clears them all, ready to save. good to know - options are for built-in agents only. an agent you added yourself runs exactly the line you wrote. - each option is checked against the release you have installed. one it does not take yet says \"needs … or later\" and cannot be set, and one that is going away says so. - a page with nothing to set for that agent cannot be opened. - unfold uses only the model and how hard the agent thinks.","agent-updates":"agent updates a toast that tells you when a newer release of one of your built-in agents is out, and updates it for you. agents how it works 1. origami checks the built-in agents you have enabled when it opens, and again every hour. 2. when a newer release of one is out, a toast names the agent and its new release, with the release you have installed under it. 3. press update on the toast. it shows updating while the update runs, and cannot be closed until it is done. 4. when it finishes, the toast says which release you now have and goes away. if the update fails, it says so, and details shows what went wrong. good to know - close the toast and you are not told about that release again. the next release brings a new toast. - settings › agents shows each built-in agent's installed release, and \"available\" beside it when there is a newer one, with its own update button. updating there takes the toast down. - updates run one at a time. - agents you added yourself, and agents you have turned off, are never checked.","notifications":"notifications a word from origami when an agent needs you or has made something for you. agents how it works 1. when an agent is waiting on you, or has finished, while you are elsewhere, origami sends a system notification. click it to go to the tab. 2. when an agent makes something for you in the workspace you are in, a toast says so, with open : a new or updated artifact, a new card, a review comment, or a reply to your comment or annotation. 3. when an agent hands you a browser page, such as a login, a toast says \"a page needs you\" with go to page . 4. one toast shows at a time and goes away by itself. what still needs you, or went wrong, moves to the bell in the titlebar, marked with a teal dot. 5. press the bell to see every workspace's notifications, sorted into updates , pull requests , errors , from agents and other . clear closes all but the updates. good to know - turn either kind off in settings › agents , under notifications : notify when waiting for the system notifications, and show what agents make for the toasts. - the app's icon counts the tabs waiting for you, and bounces once when an agent needs an answer while you are elsewhere. - each thing gets its own toast. news about the same thing updates it. - closing an update in the bell skips that release.","usage":"usage what is left of your agents' plans, and what your agents have used. agents how it works 1. press usage at the foot of the workspace rail, or choose usage in the command palette . 2. limits opens first. it shows a card for each account your enabled agents are signed in to: how much of each plan limit is used, and when it resets. 3. switch to activity to see what was used. pick a range from the menu beside the switch: today , this week , this month , 24 hours , 7 days or 30 days . 4. activity shows the cost and the tokens side by side, with a calendar of tiles, a day or an hour each, shaded by what it cost. the totals are split by provider, and under the calendar is every model's share of both. 5. click a tile to narrow the page to it. click it again to let it go. good to know - the cost is an api estimate : what those tokens would cost at api prices. it is not what a plan charged you. - activity covers claude code, codex and opencode, and goes back a month. - an account signed in with an api key has no plan limits to show. - grok's card shows how much of its plan's credits is used this period, and when the period ends. - press refresh in the page's head to read everything again.","mcp-server":"mcp server the way an agent in a terminal tab or another app works origami the way you do, from opening a terminal to answering a review comment. agents how it works 1. origami runs an mcp server on your machine while it is open. through it, an agent can list your workspaces and tabs, open and stop tabs, run a command and read its output, read the git status and diff, write artifacts, work the board, answer review threads, drive a browser tab, show you a file in files , and read and add to the scratchpad. 2. open settings › mcp . under connect there is a row for each agent you have enabled, with the line that adds origami to it. 3. press install on a row to add origami to that agent, or copy its line and run it yourself. a row says installed once it is there, and remove takes it out again. 4. if you change the server's port , the agents already connected point at the old one. their rows say so and offer fix . 5. once an agent has called origami, its row says when it was last used. good to know - a loop needs none of this: origami's tools are built in. - agents stay connected after a restart, but can only reach origami while it is open. - enable mcp server turns the whole server off. tools lists every tool it offers. - every call an agent makes is recorded in the agent trail .","agent-trail":"agent trail the record of every call an agent made to origami, and what each one changed. agents how it works 1. press trail at the foot of the workspace rail, or choose trail in the command palette . 2. the trail opens on changes : what agents created or changed, such as a tab opened, an artifact saved, a card moved or a review comment posted, newest first and grouped by workspace. 3. switch to all calls to see everything agents asked for, including what they only read and the calls that failed . one call made again and again on the same thing is one row with a count. press the count to unfold it. 4. on a row, details shows the call's input and output, and go to opens what it is about, while it still exists. good to know - the workspace menu beside the switch narrows the trail to one workspace. calls that named no workspace are grouped under no workspace . - each workspace keeps its newest 250 rows, so a busy one never pushes another's out. - clear all empties the trail. - a toast is only for what an agent made for you. everything else it does is here and nowhere else.","handoff":"handoff hand a card, your review comments or your annotations to a loop or an agent as a ready-made prompt. agents how it works 1. open the handoff menu where the work is: - a card: handoff at the top of its page, or right-click the card on the board; - review comments: review at the head of the git view, which hands over all the open threads; - annotations: the arrow beside annotate at the head of an artifact. 2. choose handoff to loop , then new loop or one of the workspace's open loops, or handoff to agent , then one of the agents running in the workspace. 3. origami puts the prompt in that loop's message box, after anything already there, or pastes it into that agent, and takes you to its tab. nothing is sent: read it, add to it if you like, and send it. 4. to use an agent outside origami, choose copy prompt for agents instead and paste it wherever you like. good to know - the prompt tells the agent which of origami's tools to use to read the work, do it and report back. a loop has them built in. any other agent needs the mcp server connected. - handoff to agent is held while no agent is running. start one first, or hand off to a loop. - the review menu is off while there are no open threads, and the annotation handoff while the artifact has no annotations.","loops":"loops a conversation in which origami drives your own installed agent through one piece of work. loops how it works 1. press loop at the top of the sidebar, or ⌘n from anywhere, to open the workspace's new loop. once you send, it is listed under loops , over the tabs. 2. origami drives claude code, codex or github copilot as you signed it in, so no key is needed. 3. a loop's line shows where it works: its worktree, the checkout's branch, or its folder. a new loop in a repository can have a worktree of its own. 4. when the work is finished, press close on its line. reopen , typing in it, or a question reaching it brings it back, with the agent's memory of the work. 5. the arrow beside loop has find loop , which searches names and messages, and closed loops . good to know - a loop is named from your first message, then by its agent where it can. rename in its menu changes it. - a workspace has one draft. what you type or attach in it is kept, and the button reads draft meanwhile. a loop left with a message unsent shows a yellow dot. - claude code's terminal commands, such as /clear , are not offered in a loop, and one typed is not sent. - the new page offers starters, such as a failed tab. a press fills the box and sends nothing.","loop-worktrees":"loop worktrees a new loop works in the workspace's checkout, or in a worktree and on a branch of its own. loops how it works 1. in a repository, under a new loop's message box, choose where it works: current checkout , shown as its folder and branch, or new worktree . 2. with new worktree , the button beside it, from and a branch, says what the worktree starts from: the checkout's branch, until you pick another. picking one checks nothing out. 3. the worktree is made when you send the first message, on a new branch named after that message under loop/ . the loop works there from then on. 4. the loop's line in the sidebar shows the worktree's name, and in the workspace's own checkout its branch, following it when the agent switches it. good to know - only a new loop chooses: once its first message is sent, a loop stays where it began. - if the worktree cannot be made, nothing is sent, and your next message tries again. - removing the loop asks whether to remove its worktree too, unless a workspace or another tab still works in it. a worktree with uncommitted changes is kept. - loop worktrees are kept outside the project's folder and are listed with the others on the worktrees page.","attachments":"attachments what you hand a loop beside your words, from artifacts and cards to files, tabs and pictures. loops how it works 1. press attach , the paper clip in the message box, or type @ where a word would start. a dialog opens with the kinds on its left, everything , artifacts , cards , files , loops and tabs , and the things on its right, the most recently touched first. 2. click rows to pick several, or press enter to add the one you are on with any you picked. shift+enter picks and stays, and tab steps through the kinds. add to loop adds what you picked. 3. files and closed loops are listed only once you type. 4. with text before the caret, a thing goes into your sentence as a token; otherwise it sits over the text. drag a token to move it into or out of the text. 5. paste a picture, or drop files on the loop's pane: each goes in where the caret is, a picture as a token such as image 1 and any other file by its path. a pasted reference becomes the thing it names. good to know - the agent is given a reference to each thing, and a file's path from the loop's folder. - pictures are png, jpeg, gif or webp, up to 5 mb. - press a token in a sent message to go to its thing, and a picture to see it whole.","shell-commands":"shell commands run a command in the loop's folder from its message box, and hand the agent what it printed. loops how it works 1. type ! first in an empty message box. it turns to shell , and enter runs what you type with your login shell, in the loop's folder or its worktree. 2. it runs at once, even while the agent works. a sheet on the page shows the command, the last lines it printed and how it ended, with the rest a press away. 3. once it ends, the agent is told what ran and what it printed, and answers. while the agent works, it is told when the turn ends. 4. the stop tile on the sheet stops the command. one you stop goes to the agent with your next message. good to know - backspace on an empty line, or esc, goes back to a message. ↑ and ↓ bring back the loop's earlier commands, and shift+enter starts a new line. - a command has no terminal, so it cannot ask you anything. use a terminal for one that does. - the agent is given at most the last 12,000 characters of what was printed.","models":"models the model a loop runs, how hard it thinks, how fast it runs, and which agent runs it. loops how it works 1. press the model at the foot of the message box, or type /model . its sheet searches every installed agent's models, and its left edge has all models , pinned and a mark for each agent. 2. press the star on a row to pin it. pinned lists every agent's pinned models. 3. beside the model, effort and speed ( normal or fast ) are set where the model has them. effort carries over to a new model that has it. 4. each row names its agent. picking another agent's model hands the loop to it, marked handed to on the page; your next message tells it the conversation so far. good to know - a workspace's next new loop starts with the agent, model, effort, speed and mode last set in its loops. - a new workspace starts from the defaults in settings › loops . until you pick a model there, claude code starts on sonnet 5.5 at medium effort where your account has it. - a loop can only be handed over while no turn is running. - where your account cannot run fast, speed stays normal and the menu says why. github copilot has no speed to set. - only models the agent offers your account are listed; an agent still reading them shows as loading.","modes":"modes how far a loop's agent goes before it asks you. loops how it works 1. the mode is at the foot of the message box, after the model. there are four, the same for every agent: ask first asks you for leave, edit freely changes files without asking, plan only plans and changes nothing, and full access never asks. /plan switches to plan only. 2. when the agent needs you, its question sits at the top of the message box. press a choice to answer. several questions are answered a step at a time and sent together. 3. a permission shows what it is for, with deny , allow for this session and allow . enter allows and esc denies. 4. a plan the agent wants to go ahead with offers approve and keep planning . 5. anything you type in the box and send is your answer, in your own words. good to know - what you answered stays in the conversation. - allow for this session is offered only where the agent can remember it. - in plan only, codex answers with its plan and does not ask to go ahead. - the mode you choose is the one the next new loop in that workspace starts in.","queue":"queue messages you send while the agent works, held until its turn ends. loops how it works 1. send a message while the agent is working and it is queued, listed at the top of the message box. 2. when the turn ends, the next queued message goes by itself. nothing goes after a turn that was stopped or that failed. 3. press a queued message to take it back into the box, or press ↑ in an empty box. send now at its end, or ⌘⏎ as you send, cuts into the turn that is running. 4. press esc, or the stop button, to stop what the agent is doing. good to know - a turn that fails at a usage limit of your plan carries on by itself shortly after the limit resets, where the reset time is known. the top of the box says usage limit , with resume now and don't resume . - sending a message yourself takes the place of the resume. - a large conversation left idle until its agent's cache has run out offers compact and continue or keep full history before your next message. - right-click a queued message to remove it.","blocks":"blocks figures, charts, tables and controls an agent draws in its answer, which you press to reply. loops how it works 1. where a picture says more than a sentence, the agent answers with a block: figures, meters, bar and line charts, a whole split into parts, heatmaps, tables, steps or tabs. a chart's corner tile shows it as a table. 2. a button or a choice answers for you: one press sends it as your next message. a choice that takes several picks, and a form, send with their own button. 3. next steps , at the end of an answer, offers up to four things to ask for next. 4. a checklist sends nothing as you tick. what you ticked goes to the agent with your next message. 5. other buttons open a thing, copy text, or open a terminal with a command typed for you to run. good to know - once you send anything after it, a block's choices, forms and answering buttons are done. buttons that open or copy keep working. - what you picked, filled in and ticked is kept with the loop. - a block that cannot be drawn says couldn't draw this block , over what the agent wrote.","goals":"goals a goal that keeps a loop going, turn after turn, until the agent says it is met. loops how it works 1. type /goal followed by what done looks like, and send it. the agent is told to work toward it. 2. each time a turn ends, origami sends the agent on toward the goal, until it says the goal is met, or that it is blocked and needs you. 3. goal sits at the top right of the message box with where it stands: working , met , blocked or paused . press it to read the goal whole and what the agent last said of it. 4. send /goal on its own, or press clear goal , to let it go. good to know - after 30 turns the goal is paused. keep going takes it up again, as it does a blocked one. - only a turn that finished keeps it going: after one you stopped, or one that failed, the next step is yours. - the messages origami sends for you show as a quiet line, kept going toward the goal . - goals work the same with claude code, codex and github copilot.","rewind":"rewind say one of your messages again from where it was, with the files put back if you want. loops how it works 1. at the end of every turn, what it changed is shown as a tree of folders and files with their lines, and a file opens to its diff. review opens the git view . 2. point at one of your messages and press the pencil, edit from here , while nothing is running. its words and attachments go back into the box, and everything after it goes quiet. 3. turn on restore files at the top of the box to also put the files back to how they were before that message. 4. send. the conversation, the agent's memory and, if you chose it, the files go back to that point, and your new words are sent. good to know - in a git repository, a turn's changes are read off the folder, so what a command or a script changed is included, and restore files is offered. origami keeps its checkpoints apart from your branches and staging. - outside a repository, a turn's changes are only the edits the agent says it made. - the agent forgets in the same session, except codex and github copilot for a message from before origami last started: they start a new session, told what was said before it. - press esc, or the cross at the top of the box, to cancel.","fork":"fork a new loop that starts from everything said in another, then goes its own way. loops how it works 1. right-click a loop under loops and choose fork , or type /fork in its message box. words after /fork are the new loop's first message. 2. the fork is named after the loop, with \"(fork)\" after it, and its page marks where it parted with forked from and the loop's name. 3. it runs on the same agent and settings, with the agent's memory of the work, and its files can still be put back with rewind . 4. from there, each loop goes on alone: what you say in one never reaches the other. good to know - a loop is forked once its agent is at rest. /fork typed while a turn runs waits for it to end. - a loop with nothing said in it cannot be forked, and says so. what you typed stays in the box.","instructions":"instructions your own words for every loop's agent, and a look at everything a loop loads. loop settings how it works 1. open settings , then loops , and write in custom instructions . they are appended to the system prompt of every loop, whichever agent runs it. 2. they always come after the agent's own prompt and never replace it, so the agent keeps what it runs on. 3. in a loop, loop context , at the foot of the message box beside attach , opens a sheet of what the loop loads as it starts. 4. its rows include origami's prompt, your custom instructions, the agent's own prompt, instruction files, memory, skills, subagents, mcp servers and hooks. each says whose it is, such as built-in , user or project , and how many tokens it takes. 5. press a row that has a file to open it in files . good to know - a loop loads what its agent would load in a terminal, its own settings, skills and servers, but not the connectors that come with your claude or chatgpt account. - tasks a loop hands to another agent are given your instructions too. - github copilot does not say what else it loads, and the sheet says so.","subagents":"subagents your own helpers for a loop's agent, each running on one agent, model and effort. loop settings how it works 1. open settings , then loops , and choose new subagent from the subagents heading's menu. 2. give it a name , when to use it and its instructions , then pick what it runs on : one agent, one of its models and an effort , or default for either. 3. a loop of that agent runs it as its own; a loop of another agent hands the work to it. either way it runs in the loop's folder and mode. 4. turning off allow use of subagents stops every loop's agent starting them. only origami subagents leaves out the ones claude code and github copilot bring with them. 5. a loop's subagents are listed, with their state, model, effort and tokens, in the sidetrack 's tasks widget. good to know - a loop's agent can also hand a task to any installed agent and model by itself. the task runs in the background, at most four at once, and its result comes back once the loop is at rest. - a task's permissions are asked at the loop's message box. - each row shows the agent, model and effort it runs on. an info mark after a name lists what else its file sets. - agents can add them with origami's tools, built into a loop and in a terminal through the mcp server .","servers":"servers your own mcp servers for claude code, codex and github copilot, and how each is doing in a loop. loop settings how it works 1. open settings , then loops , and choose add server from the servers heading's menu. 2. give it a name, then pick what it runs as : a command , with its environment , or a url , with its headers . 3. pick the agents that get it. every loop of those agents has it straight away, open ones included. 4. in a loop, loop context at the foot of the message box lists its servers with their state, such as connected , needs sign-in or failed , and their tools. 5. a server's row offers what it needs, such as sign in or reconnect . right-click it to turn off , sign out or edit in settings . good to know - a tile beside loop context counts the servers that are down. press it to list them over the box. one you hide stays hidden in every loop until it is up again. - a server that signs in asks the first time a loop starts it, in your browser. - servers are kept in each agent's own files in your home folder, and origami keeps no copy. the list shows your own, not a project's. - codex only takes a server over streamable http. - a loop has origami's own tools built in, with no server to add. agents in terminals get them from the mcp server .","skills":"skills your own skills for claude code, codex and github copilot, managed in one place. loop settings how it works 1. open settings , then loops , and choose new skill from the skills heading's menu. 2. give it a name , when to use it and its instructions . 3. under available to , pick the agents that get it. a skill for several agents is one skill they share, not a copy each. 4. press a skill's row to change it. taking an agent off removes the skill from that agent, and with none left the button becomes remove . good to know - skills are kept in each agent's own files in your home folder, so the agent has them outside origami too. origami keeps no copy. - the list shows your own skills, not a project's or a plugin's, and each row shows which agents have it. - right-click a skill for open in files , show in folder and remove . - agents can add skills with origami's tools, built into a loop and in a terminal through the mcp server .","hooks":"hooks a command your agents run when something happens in their sessions. loop settings how it works 1. open settings , then loops , and choose new hook from the hooks heading's menu. 2. pick when it runs: before a tool runs , after a tool runs , when you send a message , when a session starts , when a session ends or when a turn ends . 3. for a tool's event, for tools matching narrows it to some tools. left empty, it runs for every tool . 4. write the command under run , pick the agents it is for under available to , and press create . good to know - a hook is stopped after 30 seconds. before a tool runs, exit code 2 stops the tool. - a hook an agent adds is written nowhere until you allow it. a toast says a hook is waiting for you , and its row says added by an agent , with review and allow . - hooks are kept in each agent's own files in your home folder, and origami keeps no copy.","changes":"changes the git view's first page, where you read, stage and commit what has changed in a workspace's folder. git how it works 1. press ⌘⇧g, or the git button in the titlebar, which shows how many files changed; its tip adds the lines added and removed. the view opens on uncommitted changes . 2. each changed file is a sheet under unstaged , untracked or staged . press a file to open its diff; a changed picture shows as a picture. 3. stage or unstage a file, or a whole section with stage all and unstage all . discard throws a change away only after a second press, on confirm . 4. write a message in the commit bar at the top and press commit , or enter. only staged files are committed. good to know - shift+enter starts a new line in the message, and a message you have not sent is kept when you leave the view. - a file's … menu has open in files , copy file path and show in folder . - a folder that is not a repository offers enable git instead.","commits":"commits the branch's history in the git view, with each commit's details and files, and the button that pushes it. git how it works 1. open the git view. the sidebar turns to its history: uncommitted changes first, then the branch's commits under a heading for each day, newest first, each led by its author's initials. 2. press a commit to see its hash, tree, author, date and parent, and under them every file it changed, each opening to its diff. 3. load more , at the foot of the list, reads further back. 4. push , at the head of the view, sends your commits to the remote and shows how many are waiting. a branch with no upstream yet offers publish branch instead. good to know - the arrow beside push holds force push with lease and force push , which asks first, since it overwrites the remote's history. - a warning on push says when the branch and its remote have both moved on. - right-click a commit for copy commit message , copy commit hash and soft reset head to this commit , which keeps your files and uncommits everything after it. - origami does not pull: run git pull in a terminal .","branches":"branches switch, create, rename and delete a workspace's branches from one page of the command palette. git how it works 1. open the branches page: the branch tile on a running terminal's dock, the branch button at the head of the git view, or git: branches in the command palette . 2. the page lists the branches with the current one marked. type to search them. 3. pick create new branch... , type its name and press enter. origami creates the branch and switches to it. 4. pick any branch for switch to branch , rename branch or delete branch . good to know - delete branch asks first. a branch with commits that were never merged asks again before it is force deleted. - the current branch cannot be deleted. - a branch checked out in another worktree is greyed out. switch that worktree to another branch, or remove it, to free this one. - the dock's branch tile is there on the main stage only, and only while the workspace's folder is a repository.","worktrees":"worktrees choose which checkout of a repository the git view works on, and add or remove worktrees. git how it works 1. open the worktrees page: the folder tile on a running terminal's dock, whose tip names the worktree, the worktree button at the head of the git view, or git: worktrees in the command palette. 2. the workspace's own folder is listed first, as main , then every other worktree with its branch, including those loops made. 3. pick one for switch to worktree , open terminal here , copy path or remove worktree . switching points the git view and the branches page at that checkout, and the choice is kept for the workspace. 4. add worktree... asks for a folder and a branch, existing or new, and switches to the worktree once it is made. good to know - switching moves no tab: a terminal keeps running where it started. open terminal here starts a new one in the worktree. - removing a worktree with uncommitted changes asks again, with force . - main cannot be removed. a worktree whose folder has gone is tagged missing and offers prune . - a worktree inside the workspace's folder may show up as untracked files in the main checkout, and the dialog warns you of it.","stashes":"stashes put uncommitted changes aside and bring them back later, from the git view. git how it works 1. in the git view's commit bar, switch commit to stash . 2. choose what goes: all , tracked (everything but untracked files) or staged . a message is optional and names the stash. 3. press stash , or enter. 4. switch the sidebar from commits to stashes to list them. press one to see the files it holds and their diffs. 5. right-click a stash for pop stash , which applies it and takes it off the list, or apply stash , which applies it and keeps it. good to know - drop stash , in the same menu, asks first and cannot be undone. - copy stash message is there too. - the stash button stays off while nothing falls in the scope you chose.","review-threads":"review threads comments pinned to lines of a diff, which you and your agents read, answer and resolve. git how it works 1. open a file's diff in the git view and press the + beside a line. drag from it to take in several lines. 2. write the comment and press ⌘⏎, or comment . escape cancels. 3. hand the open threads to a loop or an agent with review , at the head of the view, which counts them. see handoff . 4. the agent reads the threads with origami's tools, built into loops and offered to other agents by the mcp server . it replies on them, resolves them, and can start threads of its own. 5. open a thread to reply, resolve it, or reopen one that was resolved. good to know - a thread follows its lines as the file is edited. when they are gone it says anchor lost , and can be removed. - while threads are open, the titlebar's git button is marked and its tip says how many. each file's sheet shows its own count. - threads belong to files under review: uncommitted, or changed by the branch. once a file's changes are discarded or its branch is merged, its threads stop showing. - a thread cannot be started on the staged section.","branch-review":"branch review everything the current branch changes since it left its base, committed or not, as one diff. git how it works 1. in the git view's sidebar, press branch changes , under uncommitted changes , or run git: branch review from the command palette. 2. each file the branch changed is a sheet with its diff, headed by how many commits the branch is ahead of its base and behind it, and the lines changed. 3. the base is the repository's default branch. to compare against another, press vs and the base's name at the head of the view, and pick one. the choice is kept for the repository. 4. comment on any line as you would on the uncommitted changes: the review threads are the same ones. good to know - the page only reads. stage, commit and discard from uncommitted changes . - where the repository names no default branch, the page asks once: what does this workspace merge into? - in the commit list, branched from marks where the branch begins. - agents read the same diff with origami's tools, built into loops and offered to other agents by the mcp server , so a thread on a committed line means the same to them as to you.","pull-requests":"pull requests a workspace's github pull requests and gitlab merge requests, in a sidetrack widget. pull requests how it works 1. press + in sidetrack 's head and choose pull requests under widgets . if no account is connected for the repository's host, it offers to connect one . 2. it lists requests under this branch , yours , review requested , other open and recently merged . each row shows its checks, its reviews and how many comments it has. 3. press a row to open it on overview , checks , conversation and files . overview lists what still blocks the merge, the reviews, and the description, labels and assignees. 4. files opens the changes in branch review when the checkout is on the request's branch. otherwise it offers check out in a new worktree , or start a loop on its branch , which makes the worktree and puts the request in a new loop's message box. good to know - gitlab's merge requests are called pull requests here too. their checks read pipeline and their reviews approvals . - agents can list a workspace's requests and read one, with what blocks it, the logs of its failed jobs and its open threads. - the bell says when checks on your requests finish, one is approved, has changes requested or is merged, or your review is asked for.","accounts":"accounts the github and gitlab sign-ins origami reads and changes pull requests with. pull requests how it works 1. in settings › pull requests , open the menu on accounts and choose github , github enterprise , gitlab or gitlab self-managed . 2. for github or gitlab, sign in with github or sign in with gitlab opens the site in your browser with a code already copied, for you to approve origami. 3. other ways takes a personal access token instead, or uses the sign-in of gh , glab or git where one is signed in to that host. 4. github enterprise and self-managed gitlab connect with a token or a tool's sign-in. type the host , and origami checks what answers there. good to know - a github token needs the repo and read:org scopes, and a gitlab one api . an action a token cannot take says what it lacks, with reconnect . - tokens are kept in the system's keychain. a tool's sign-in is asked of the tool each time, never copied. accounts are never exported. - the same page sets whether new requests start as drafts, pushing before creating, what happens after a merge, notifications, and which browser opens a request's page.","checks":"checks a pull request's ci jobs, their logs, and re-running the ones that failed. pull requests how it works 1. open a request's checks . jobs are grouped by workflow, or by stage on gitlab, each with its status, how long it took and its attempt. required marks one the merge waits on. 2. press a job for its steps. the step that failed is open, and the ones that passed are folded. 3. press a step for the end of its log. load full log reads the rest, and copy log copies it. 4. when a check failed, re-run failed jobs is in the band under the request's head. a run's menu also has re-run all jobs and cancel run , and a job's log has re-run job . good to know - handoff on a failed job's log gives it to an agent, with the end of the log. see handoff . - only the host's own ci can be re-run from origami. other apps' checks are listed under other checks . - on gitlab a re-run is retry , a manual job has run , and there is no re-running all jobs.","reviews":"reviews comment on a pull request, answer its threads, and approve it or ask for changes. pull requests how it works 1. on overview , reviews lists who is asked, what each said, which are code owners, and how many approvals are still required. request reviewers asks more people, teams or bots. 2. under them, approve , request changes and comment send your review. approving needs no text; the others do. 3. conversation holds the code threads by file, then the comments in order. write at its foot to comment, and reply to or resolve a thread where it is. 4. press add reaction under a comment or the description to react. press your own reaction to take it back. good to know - origami answers the code threads a request already has, but does not start a new one on a line. - a comment's menu has quote reply , edit and remove for your own, and hide for github moderators. - handoff on an open thread gives it to an agent. - on gitlab, reviews are approvals , and teams cannot be asked.","merging":"merging merge a pull request from origami, by the repository's own rules. pull requests how it works 1. the band under a request's head says where it stands, such as ready to merge or merging is blocked , and what is left in the host's terms: approvals, required checks, open threads or conflicts. 2. its button is the next step: ready for review for a draft, update branch when it is behind, merge when ready where the repository has a merge queue, or merge . 3. the arrow beside merge picks a method the repository allows, whether the branch is deleted, and the squash commit's message. where the merge must wait, enable auto-merge merges it once it can. 4. after a merge, origami can delete your local branch and remove the worktree it made for the request, as settings › pull requests says. good to know - bypass rules and merge is offered only where the host lets you merge as an admin, and asks first. - gitlab's queue is the merge train, joined with add to merge train . a request can join a train but not skip it. - the head's menu has convert to draft and close pull request , and delete branch once it is merged or closed.","new-requests":"new requests open a pull request for the branch you are on, from the git view. pull requests how it works 1. in the git view , press create pull request beside push, or new in the pull requests widget. it is there for a repository on github or gitlab with a connected account, on a branch other than the default. 2. the sheet holds the base branch, a title from the commit or the branch's name, and the description from the repository's template, with template to pick another where there are several. 3. add reviewers , and assignees , labels and milestone where you may. the code owners it will ask are named. create as a draft opens it as a draft. 4. create pull request pushes the branch first if the remote lacks its commits, asking unless settings › pull requests says always. the new request then opens in the widget. good to know - fill with agent has an agent write the title and description. an agent can also fill the sheet on its own; only you create the request. - if the branch already has an open request into that base, the sheet shows it instead. - templates and code owners are read from the base branch, not your working copy.","board":"board a kanban board for each workspace, which you and your agents keep together. board how it works 1. press ⌘⇧o, or kanban in the titlebar. a new board has three statuses: backlog , in progress and complete . 2. drag a card from one status to another. complete shows what was finished this week; show more lists the rest. 3. the sidebar holds the board's controls: the layout ( board , epic or list ), the search, which looks through keys, names, descriptions and tags, and beside it the squares for statuses and new card . 4. under the search, narrow the board to one epic , or open filter for reporter , type and priority . each filter that is on shows as a token; press it to take it off. 5. recent lists the latest changes to cards, each marked by whether you or an agent made it, and opens the card when pressed. good to know - statuses adds, renames, reorders and removes statuses. backlog and complete stay first and last, and a status still holding cards cannot be removed. - list shows the cards as an outline of what sits under what. - the kanban button is marked when the board has changed since you last looked at it. - agents read and change the board with origami's tools, built into loops and offered to other agents by the mcp server .","cards":"cards a piece of work on the board, with a key, a page of its own and a conversation under it. board how it works 1. press new card beside the board's search, or n on the board. name it and press create card . it goes in backlog unless you set another status, with a key made from the workspace's name, such as ori-42. 2. press a card to open its page. move to , at its top, takes it to the next status. 3. under the name, pills set its status, priority and parent. press the kind's tile to make it an epic , story , task or bug . tags go on the line below. 4. the page has a description , subtasks (cards whose parent it is, ticked off in place) and attachments : a file, an artifact, a link or a browser, or files dropped on the page. 5. comments go under the page. good to know - a card says who reported it, you or an agent. for five minutes after an agent reads or changes it, it says an agent touched this card . - handoff gives the card to a loop or a running agent of the workspace, or copies the prompt for one. see handoff . - press the key on a card's page to copy it. on the board, a card's … menu also has copy reference . - remove asks first, and its toast has undo .","epics":"epics a card that gathers the stories, tasks and bugs of one larger piece of work. board how it works 1. make a card and set its kind to epic , or change an existing card's kind from the tile on its page. 2. make the epic another card's parent, with that card's parent pill. a card belongs to the nearest epic above it, so a task under a story under an epic is the epic's too. 3. narrow the board to one epic from the epic field under the search, which opens the epics on the command palette, newest first. its tile clears it again. 4. switch the layout to epic to see each epic as a lane across the statuses, with how much of it is complete. cards with no epic share a no epic lane. good to know - epics are not on the board layout, since they organise work rather than move through it. filter by type › epic to see them there. - in the epic layout, drag a card along its lane to change its status. - list shows each epic with everything that sits under it.","artifacts":"artifacts documents your agents publish to origami, drawn as they are meant to be read and kept with every version. artifacts how it works 1. an agent publishes markdown, html, json or a mermaid diagram as an artifact, with origami's tools: built into loops , and offered to other agents by the mcp server . a browser screenshot you ask an agent to keep becomes one too. each artifact belongs to the workspace it was made in. 2. press ⌘⇧a, or artifacts in the titlebar. the sidebar turns to the list, newest first, under today , last 7 days and older , which is folded under its heading. 3. press a row to open it. each change an agent makes is a new version: the version button at the right of the head lists them with their dates, so you can go back to any. 4. right-click a row, or press the ellipsis at the head, for copy reference , open in files , open in browser , group , rename , export and remove . good to know - copy reference copies a link an agent can turn back into the artifact, so you can point one at it in a prompt. - an artifact that is new or changed since you last opened it is marked on its row. - search artifacts looks through all of them at once, with no sections. - remove asks first, then deletes the artifact and all its versions for good.","live-artifacts":"live artifacts an html artifact that keeps what you choose on it and tells the agent that wrote it. artifacts how it works 1. an html artifact is drawn in origami's own design, the origami kit, which follows your theme and the workspace's colour, unless you asked the agent for a look of its own. 2. its page can hold choices, forms and buttons. what you pick or type is kept with the artifact and is there when you come back. 3. what you do on it reaches the loop that wrote or changed it by itself, as do your annotations and replies. a loop at rest hears a moment later, and one at work once its turn ends. the loop then shows told about your feedback on and the artifact's name. 4. open in browser opens it in your own browser, where it stays live: what you do there reaches the agent too, and a new version reloads the page where you were. good to know - an agent working through the mcp server is not told. it reads what you did on the page, and can wait for it. - a loop is told by itself at most 12 times in a row. after that, what you do goes with your next message. - an exported artifact keeps its look but sends nothing.","groups":"groups a name artifacts share, so the work that belongs together is found together. artifacts how it works 1. right-click an artifact and choose group . the palette opens on the workspace's groups: pick one, type a new name, or choose add to new group . for an artifact already in a group, add to new group becomes move to new group , and remove from group is added. 2. to file several, pick their rows with ⌘-click or ⇧-click and press group on the bar at the sidebar's foot. 3. switch the list from recent to groups to see each group under a heading with its count, the one written to last first and ungrouped last. press a heading to open or close it. 4. right-click a group's heading for rename group and ungroup artifacts . renaming a group to another group's name merges the two. good to know - an artifact is in one group or none. a group exists only while an artifact is in it, so there is nothing to create or delete. - a name that differs from an existing group only by case, spacing or a plural's last \"s\" files into that group. - agents are told to reuse the groups a workspace already has. a browser screenshot an agent keeps for you goes in screenshots . - filing an artifact does not change when it was last updated, and its reference stays the same.","annotations":"annotations feedback you pin on a version of an artifact, for an agent to address in the next one. artifacts how it works 1. open an artifact and press annotate at the left of the head. it reads stop annotating while it is on. 2. click anywhere on the artifact to drop a numbered pin, type your comment and press enter. shift+enter starts a new line, and esc cancels. 3. press a pin to read its thread, add a reply , or take it away with remove annotation . an agent replies there too, which is how it asks about a pin it finds unclear. 4. press the arrow beside annotate to hand the review to a loop or an agent: see handoff . the prompt asks the agent to read your annotations, ask about any that are unclear, and publish a new version. good to know - pins belong to the version they were made on, so a new version starts with none. - a row in the list shows how many annotations its latest version has. - no handoff is needed for the loop that wrote or edited the artifact: it hears of new pins by itself once it is free. agents on the mcp server are told to check your annotations before writing a new version. - while annotating, a click places a pin and follows no link.","appearance":"appearance how origami looks, from its theme and its terminals' colours to the size of everything in it. setup how it works 1. press ⌘, to open settings at general . under appearance , theme is light , dark or system , which follows the system's own setting. 2. terminal colors picks the palette terminals draw in. it lists the schemes made for the theme in use: o-dark, dracula and mocha in dark, and o-light, solarized and latte in light. 3. ui scale sizes the whole interface at 90%, 100%, 110% or 120%. ⌘⇧+ and ⌘⇧− step through them from anywhere, and ⌘⇧0 goes back to 100%. 4. under terminal , font size sets terminal text from 10 to 24 pixels. good to know - the titlebar, the rail and the main area take the terminal scheme's background colour, so a terminal runs edge to edge with nothing framing it. - theme... in the actions palette (⌘⇧p) switches the theme without opening settings. - every change applies at once, to every open terminal.","storage":"storage what origami keeps on this computer, with ways to give the room back, move it elsewhere or start over. setup how it works 1. space used is one bar for the total, split into loop worktrees, browser data, loops, tab output, artifacts, boards, screenshots and everything else. 2. under clean up , left behind removes the data that workspaces and tabs you have removed left on disk. a loop worktree with changes not committed is kept, and origami says so. 3. closed loops and browser data open a list first: tick what to remove or clear, and press the button at the foot. removing closed loops can take their worktrees too. browser data is listed by what shares it: all of origami, a workspace, or one browser. 4. older artifact versions , tab output , screenshots and agent trail each ask once. every artifact keeps its latest version, and running tabs keep their output. screenshots are the pages agents looked at, kept for a week. 5. transfer has export and import : see export and import . good to know - a row's button is off while it has nothing to act on. - reset settings puts every setting back as it first was, and keeps your workspaces, custom agents and the mcp server. - remove everything removes every workspace with its tabs, board, artifacts and loops, resets the settings, and cannot be undone. it asks first.","export-and-import":"export and import your workspaces, boards, artifacts and settings saved to one file, to bring back later or onto another computer. setup how it works 1. in settings › storage , under transfer , press export . the kinds are on the left, workspaces , boards , artifacts , custom agents and settings , and their rows on the right. untick what to leave out, press export , and choose where the file goes. 2. to bring it back, press import and choose the file. origami checks it before anything is added, and refuses one that is not an origami export. 3. pick what to bring in, kind by kind. a workspace whose folder you already have in origami is added as a copy, replaces yours or is merged into it, as you choose, with merge as the default. 4. press import . the foot says what will come in. good to know - a workspace goes with its tabs and the custom agents they run. its board and its artifacts, at their latest version, go with it when ticked. settings are left unticked, and importing them replaces yours. - a board or an artifact already in your workspace is left as it is, so importing the same file again adds only what was missed. - a workspace whose folder is not on this computer can be pointed at another with choose folder . every tab arrives stopped. - the scratchpad is not exported: it stays on this computer. - an import stops at your workspace limit.","origami-plus":"origami+ the subscription that lifts the free limits on workspaces and the items in each. setup how it works 1. free, origami has every feature, for up to 5 workspaces and 10 items in each. origami+ makes both unlimited, and adds early access to new features and priority support. 2. in settings › general , the card at the top says get origami+ . learn more opens the plans, monthly or yearly. 3. to use a licence you have, press activate , paste the key and press enter. 4. once it is active, the card shows your email, and manage opens https://tryorigami.app/plus, where you look after your subscription and its devices. good to know - a licence runs on up to 5 devices. activating on one more says the device limit is reached: remove a device on the manage page, then try again. - origami checks the licence while it is open, and keeps working offline for 7 days after the last check. - origami+ enterprise gives a team origami+ from 5 seats up, with 2 devices a seat and an admin roster. the organisation's name shows beside origami+. - origami from setapp is already origami+, and shows no licence card.","updates":"updates origami keeps itself up to date, and shows what changed once it has. setup how it works 1. origami looks for a new release when it opens and every minute after. when there is one, a cloud button appears in the titlebar: update available! 2. press it, and origami asks whether to update to the new version. update downloads and installs it, then restarts origami; later leaves it for now. 3. to look yourself, choose check for updates from the app menu, or check for updates in the actions palette (⌘⇧p). with nothing new, it says you are on the latest version. 4. the first time origami opens on a new version, it shows that version's changelog. open it again from changelog in the app menu, where every version is listed beside it. good to know - after the restart your workspaces are as you left them, with their tabs stopped as after any quit; commands set to start on launch start again. - your agents are updated apart from origami: see agent updates . - origami from setapp is updated by setapp, and has no check for updates .","troubleshooting":"troubleshooting fixes for the problems that come up most, and where to ask when they do not help. setup how it works 1. an agent cannot use origami's tools. in settings › mcp , find the agent under connect . not installed offers install . installed for port … means its entry points at an old port, and fix points it at this one. if the server says it failed to start, try another port , then fix each agent. 2. an agent's row never shows it working or waiting. in settings › agents , open the agent's options , then its origami page, and check that report status is on. if it says it is off until origami is reopened, quit and open origami again. antigravity needs set up there, once. 3. a tab exited with an error. its output stays on screen, and the last lines usually say why. fix the cause, then press ⌘⏎ or start on its row. 4. activating a licence says the device limit is reached. remove a device at https://tryorigami.app/plus, then activate again. good to know - changing the mcp port means fixing every agent that was connected to the old one. - only built-in agents report what they are doing. a custom agent's state is read from its terminal instead. - for anything else, write to us at https://tryorigami.app/support."}