Frequently asked questions

This page collects the questions that come up most often while using Orbi. The answers are deliberately short, and the detail is behind the link at the end of each one, in the guide section that covers the topic. If you are chasing a specific error, Troubleshooting goes from symptom to cause, while this page answers "can I do this, and why does it work this way".

#Getting started and account

#Do I have to add an API key to use Orbi?

No. If you add no key at all, agents run on Orbi's shared model pool (Gemini) and the cost comes out of your usage credit. If you add your own key, the cost goes straight to your own provider account and the same work spends far less credit. That is the way to go for regular use. More in Models and providers.

#Can I make the left menu smaller?

Yes. The collapse button in the top corner of the menu, or ⌘B (Ctrl+B), shrinks the menu to a narrow strip that shows only icons. Each icon works directly: a click goes to its page, and its name shows in the tooltip. The mouse does not widen the strip, so the icon you aim at does not move. With the keyboard, Tab into the strip opens the whole menu over the page and Esc closes it. The same button or key opens the menu for good. The choice is stored in your browser, so the menu stays the way you left it next time. You can see every shortcut in the app in the table that the ? key opens. Details: Reference.

#Can I use Orbi in English?

Yes. Pick Turkish or English under Preferences on your profile page. The choice is saved to your account, so you get the same language whichever device you sign in from. Notifications and the emails sent to you are written in that language too. If you did not choose one when you signed up, your browser's language is used. More in Roles, invites and sessions.

#Do agents count as seats?

Yes. Humans and AI are not counted separately, so creating an agent also uses one of the seats in your plan, because an agent is a team member that takes work on the board. Orbi's own Orchestrator does not count. More in Plans and security.

#What do I gain by paying yearly?

Paying yearly costs ten months' worth, so two months come free. You pick the interval from the switch above the plan cards on the Settings → Plan page, and the cards show both the yearly amount and what it works out to per month. Your monthly credit does not change and still renews monthly, because paying yearly does not hand you a year's credit at once. More in Plans and security.

#I have several workspaces. Do I pay for each of them separately?

Yes. The plan belongs to the workspace, so the price, the monthly credit and every capacity limit apply to one workspace. Moving one workspace to the Team plan does not upgrade the others, and credit spent in one is not taken from another. Opening a new workspace does not use up an allowance from your current plan either, because the new one starts on the Free plan. More in Plans and security.

#I cannot see the upgrade and credit purchase buttons on Settings → Plan. Why?

Your installation may not be accepting payments yet. When that is the case a line at the top of the page says so. Your workspace runs on the Free plan, every feature is available and only the purchase buttons are closed. You still see your credit balance, your capacity meters and your usage breakdown. Get in touch if you need more capacity, and we can set your plan and your credit up for you. More in Plans and security.

#How do I switch from monthly to yearly billing, or back?

On the Settings → Plan page, pick the tab you want and use the button on your current plan's card. Your plan does not change, only how often you are billed, and the switch happens at the end of your period. Nothing is charged at that moment and you keep the period you paid for. You can cancel the decision any time before the period ends. More in Plans and security.

#A plan card says it is not on sale in the currency I picked. What should I do?

That plan has no price in the currency you picked, which is why no purchase button is shown. You can pick another currency above the plan cards, or get in touch with us. In a workspace with a running subscription the picker is hidden, because the sale continues in the currency the subscription started in. More in Plans and security.

#I want someone on the team to only watch the board. How do I do that?

Invite them with the Viewer role, because that role reads everything and can change nothing. You can change the role later from Settings → Team. More in Roles, invites and sessions.

#Can I make an invite link expire?

Yes. When you click New link on the Settings → Team screen you choose an expiry (1, 7 or 30 days), a usage limit and the role people join with. A link that expires or reaches its limit closes on its own, and you can also delete it from the list. You can have several links at once. More in Roles, invites and sessions.

#Why am I asked to sign in again every now and then?

Your session normally stays open for weeks, and the app refreshes it quietly in the background. Devices do not sign each other out either, so a session closing on one device leaves the others alone. If all of your devices ask you to sign in at the same moment, one of the three actions that close the whole account was taken, namely changing the password, resetting the password or Close all sessions. More in Troubleshooting.

#Can members of other workspaces see my data?

No. Each workspace's data stays inside it. A member of another workspace cannot see your task, your file or your record, even if they know its identifier. More in Plans and security.

#Can we run Orbi on our own servers?

Yes. For teams whose data has to stay inside the organisation there is an on-premise installation. It uses the same container images as the cloud installation, and the only traffic that leaves is the inference request to the model provider you chose. More in Plans and security.

#Can I open more than one free workspace?

No, one person can open at most one free workspace. Workspaces you joined by invite do not count towards that number, and however many teams you join your own right to open a workspace stays. If you move your workspace to a paid plan you can open a free one again. Workspaces in the trash still count, and you can restore a deleted workspace from the trash. If that paid workspace later goes back to free, it stops receiving its monthly free credits, but the workspace keeps working and the credits you purchased are unaffected. More in Roles and invites.

#Can I download all of a workspace's data?

Yes. Admins download the whole workspace as a single zip archive from the Settings › Data tab, and it carries tasks, collections, files, agents, flows and settings. API keys, connection credentials and the search index are not part of the archive, because they are set up again when you restore it later. More in Roles and invites.

#Can I restore the archive I downloaded?

Yes. The Import from archive button in the workspace list builds a new workspace from the archive, and it never writes over an existing one. Members are matched by email, queued tasks are not restarted, and inbound webhook addresses are generated again. You have to set up API keys and connections again from Settings. The new workspace starts on the free plan, and if your archive is larger than that plan's limits its contents are not trimmed, but you cannot add a new member, space or flow until you are back under the limit. More in Roles and invites.

#If I delete a workspace, is our data gone straight away?

No. A deleted workspace waits in the trash for 30 days, and any admin can restore it from the workspace list during that time. Once the period is over, projects, tasks, collections, files and agents are deleted for good. Only admins can delete a workspace, and you are asked to type its name to confirm. More in Roles and invites.

#What happens to the comments I wrote if I delete my account?

Your comments and task history stay where they are, signed "Deleted user", so your team's history is not broken. Your name, email, avatar, password and linked accounts are deleted, and your sessions are closed. If there are workspaces where you are the only admin, you are asked to hand that role over first. More in Roles and invites.

#Why is the Trail tab on an old card empty?

The run history that shows what the agent did, step by step, is kept for 180 days and then cleaned up. The card itself, its description, its checklist, its comments and its last summary stay where they are, and the only thing removed is that old step-by-step record. Notifications, usage records and the admin activity log have similar periods. If you run your own installation, your administrator can change them. More in Plans and security.

#Can I give teammates a title, and do agents see it?

Yes. On Settings → Team every member can have a title and a short description specific to this workspace; an Admin edits everyone's, each member edits their own row. Agents see this list on every run and, when they write to someone, pick tone and level of detail from that description; when you say "assign this to the designer", they find the right person by title. The title also shows in the team list, in the @ list and on the card that opens when you click a person chip. More in Roles and invites.

#Spaces

#What is a space, and how is it different from a board?

A space splits the workspace into lines of work: sales, social media, development and so on. Every board, collection, file, flow and channel lives in a space. A board is only where cards sit, while a space keeps all the content of one line of work together. Genel is the company's shared space and everyone sees it. More in Spaces.

#Another team's work fills my Overview. How do I hide it?

Take that team's space out of your menu with the button that appears when you hover it. Overview, the Today list and the morning digest show only the spaces in your menu. The calendar and search cover every space you can access; the Space selector there narrows to one space. Your access stays; open the space from the Other spaces list whenever you need it. More in Spaces.

#How do I find a card in a space that is not in my menu?

Search and the calendar already cover every space you can access, so cards of a space outside your menu appear in the results too. Use the Space selector to narrow to one space. If you know the card's number (like #42), search finds it wherever it is. More in Spaces.

#Can someone be a manager in one space and only a viewer in another?

Yes. In an open space everyone works with their workspace role. To give someone a different role in one space only, add them under Roles in this space in the space's settings. Someone who is a Viewer in the workspace can be at most a viewer in a space. More in Spaces.

#How do I let someone see only one space?

Every other space has to be closed, because every workspace member sees an open space. You close an open space under Space settings → Visibility; nobody loses access, and you remove the people who should not see it afterwards. You set the person's place in every space from the Space access window in Settings → Team. Genel stays open to everyone. More in Spaces.

#Can I close a space I opened as open?

Yes. Turn it closed under Space settings → Visibility. Everyone who sees the space at that moment stays in; the confirmation says what changes outside the space. More in Spaces.

#I asked the agent in chat to create a file and it asked me which space. Why?

Your chat belongs to you and to no space, so new content has no default place. If you name the space in your sentence, the agent does not ask. In a chat inside a card, new content goes to the card's space. More in Spaces.

#What happens to the content if I delete a space?

You choose. You can move the content to Genel or another space and then delete the space, or delete the content too; files then wait in Genel's trash for 30 days. Genel cannot be deleted. More in Spaces.

#A card assigned to me moved with its board into a closed space. Why can't I see it?

Only the space's members see the card. When a move takes a card out of your reach you get a notification, and the board's old address says "This board moved". If you still need to work on it, ask the space manager to add you as a member. Suspended share links do not reopen on their own either. More in Spaces.

#I was removed from a closed space. What happens?

You get a notification and the space leaves your menu; a board open in a tab says "you cannot access this board". The cards you queued go back to the list, the flows you built stop and the external links you opened in the space are suspended. Cards assigned to you stay assigned to you; the space's manager reassigns them. More in Spaces.

#Who sees a closed space? Does the workspace admin see it?

In the app, only the space's members. A workspace admin who is not a member sees the space's name and members, not its content. To see the content they add themselves; the step is written to the Audit tab and the space's managers are told. The full workspace archive an admin downloads does include closed spaces; every export is written to the Audit tab for each closed space it covers. More in Spaces.

#Can an agent in a closed space write to another space or to Genel?

No. An agent working on a card or flow in a closed space writes only to that space. It knows Genel's pinned rules, but it cannot carry the space's data to Genel, to another space, to a connection that was not chosen or to a web address it has not seen. More in Spaces.

#Do a closed space's events reach a webhook or a connection?

Only when it is chosen. Events from Genel and the open spaces always reach a webhook; a closed space's events go out only when it is ticked in the webhook form's Closed spaces list. Only the space's manager turns a workspace connection on for a closed space. Both choices are written to the Audit tab. More in Outgoing connections.

#I connected Claude Code as myself but it does not see my closed space. Why?

Every token you use as yourself sees Genel and the open spaces; it sees a closed space only when you choose it. That holds for a tool connected on the consent screen and for a token you create by hand, and tokens you created earlier do not see a closed space until you choose it either. Open the token's row under Profile → Tool access → Access tokens and tick the closed spaces you are a member of with Choose closed spaces. If you leave a space, the token loses it too. More in Connecting.

#My agent can't see a space. Why?

Read the first sentence of the error: it names the workspace that was searched. There are two common causes. If several workspaces are connected to the same client, the agent may have used the wrong connection; when the space exists in another workspace, the message names that one too. Give the connections separate names that include the workspace. The second cause is an agent token with the Old token badge: with no person recorded behind it, it cannot reach closed spaces. Create a new token and revoke the old one. More in Connecting several workspaces to one client.

#Tasks and execution

#How do I stop an agent that works on its own computer?

From the menu next to the status line on the card or in the chat. "Stop after this turn" waits for the agent to finish the step it is on; "Emergency brake" stops it at once. The person who started the work, people who can edit the card and workspace admins can stop it. Details: Work on the agent's own computer.

#Can I see all the work agents do on their own computers in one place?

If you are a workspace admin, yes: the Settings › Runtime tab shows what is running and waiting now, what finished today and in the last 7 days, how it ended and the credits it used. You can also stop running work from there. Details: Runtime.

#Does work on the agent's own computer use credits?

Yes, like any other run. The price of the agent's model is charged in credits; with the workspace's own API key far fewer are charged. The detail of the status line shows the credits the work has used so far. Work that stops with "the safety limit was reached" hit the high upper limit that stops work gone out of control; normal work stays well below it. Details: Work on the agent's own computer.

#Why did a card an agent opened not start on the computer?

Work on the computer is always done on behalf of a person. A card opened by an externally connected agent runs on behalf of the person who set up that agent's connection. When no person stands behind the connection, the card says it has no one to act for; queue it yourself and it starts on your behalf. Details: Work on the agent's own computer.

#On whose behalf does an agent I mention in a channel work on its computer?

On behalf of the channel, not yours, because every reader of the channel sees it. Nobody's personal connections are used, only files every reader of the channel can see reach the computer, and credits are charged to the workspace. A squad conversation in a channel works the same way. Details: Work on the agent's own computer.

#Can I tell an agent working in a channel something more while it works?

Yes. If you called the agent, every message you write in the channel reaches it while it works; anyone else reaches it by mentioning it. Such a message does not start a second reply. Where the agent cannot decide, it can also ask the channel a question, and anyone there can answer. The question waits 10 minutes; if no answer comes, the agent goes on with its own choice and says which choice it made. Details: Calling an agent.

#Does long work on the computer stop after 2 hours?

No. One computer stays open for at most 2 hours, but if the agent is still working when that time is up, the work goes on in a new computer with the same folder and the same conversation, and the status line keeps saying "Working". It goes on only if the last stretch made progress and the card's cycle limit allows it. On a squad's shared computer each member continues its own conversation on the new computer. Details: Work on the agent's own computer.

#What happens when the agent's computer runs out of memory?

The computer closes, but the work is not lost: Orbi opens a computer with twice the memory on the same folder and continues the work once by itself. The larger memory stays for the later work of that card or chat. Work that does not fit even the highest limit stops; then make the job smaller or split it into parts. Details: Work on the agent's own computer.

#Do scheduled flows run on the agent's computer?

Yes, in workspaces where the feature is on. Each flow step decides like a card does; when the work goes to the computer, every run starts with a clean folder that its steps share, and the folder is deleted 30 days after the run ends. The run acts for the person who set up the flow; if that person has left the workspace, the run does not start and the admins are notified. More: Flow steps on the agent's computer.

#When does an agent work on its own computer?

In workspaces where the feature is on, every run starts with this decision and the agent makes it. A simple question it can answer at once is answered in Orbi; anything that needs work goes to the computer. If a file is attached to the card or the comment, or the work is code in a repository, the work starts on the computer directly. Details: Work on the agent's own computer.

#Where do the Word, Excel or PDF files the agent makes on its computer go?

The agent delivers a finished file to Orbi and the file lands where the work is: on a card in the card's attachments, in a chat as a message of the agent, in a flow on the Files screen. Only delivered files leave the computer; a file can be at most 50 MB. Details: Files the agent delivers.

#The status line says "Waiting for an answer". What should I do?

The agent asked a question and waits for your answer. On a card the question is the agent's comment that starts with "❓", in a chat it is the agent's last message. Answer with a comment on the card or a message in the chat, and the agent goes on from where it stopped. No email or notification is sent for this. If no answer comes within 20 minutes the computer closes, and an answer you write later starts the work again from the same place. If the line has no computer icon, the agent asked in work without a computer: the card is in review, and an answer comment without a mention queues the card again. Details: Talking to the agent while it works.

#Can an agent on its own computer send email for me?

Yes, if you opened that service to the agent. On its computer the agent uses connected services as it does in Orbi: an agent that can send email in Orbi sends it from its computer too, and a service you did not open to it is out of reach in both places. The account's credentials stay in Orbi and never enter the computer. Details: Work on the agent's own computer.

#How do I make agents know our brand voice in every piece of work?

Mark the collection or the file that holds your brand voice as agent context: turn on Give to agents as context in the collection's settings or in the file's details. Before the agent starts work on its own computer, that content is put there and the agent reads it first. Only admins change the switch. Details: Files and agent context on the computer.

#The agent left the checklist half done and sent the card to review. Why did it not continue?

It does now. With open checklist steps and budget left, the agent starts the next cycle in the same run and the card stays "running". Only when a cycle closes no step and adds no output does the card come to you, and the summary says so. Details: The agent works until the work is done.

#The agent ticked a few steps but the card went to error instead of review. Why?

Because it left nothing to review. If the run wrote no output, comment, record or file and the checklist still has open steps, the ticked steps are only the agent's own claim. The card then does not reach review: it goes to error without a retry, and the history shows the agent's summary next to the reason. You can queue the card again or try it with another model. Details: Retry, requeue and locking.

#The agent wrote the same report twice. Which one is final?

If the agent writes its report again in the same run, the new text becomes a new version of the first report: it stays one document in the Outputs tab and its heading shows the latest version. If you see two separate deliverables on the card, the agent added them because they were asked for separately (a post and an article, say), or the second report came from a separate run. Details: The agent works until the work is done.

#What is the "working note" on the card's Summary tab?

The note the agent leaves for its next cycle: sources read, decisions taken, the next step. Only the agent writes it and it is not a deliverable. Details: The agent works until the work is done.

#I created a card but nothing is happening. Why?

Creating a card does not start execution. Exactly four things run an agent, namely moving the card to Queued, an @agent mention in a human comment, a flow trigger, and a request from an external system to an inbound trigger. More in Tasks and Kanban. If the card sits in Queued without running, see Troubleshooting.

#I queued a lot of cards at once and some are still waiting. Will they be lost?

No. Your plan sets how many jobs run at the same time, and the rest wait their turn and start on their own as the running ones finish. While waiting, the card sits in the Queued column and there is nothing for you to do. If you want more jobs running at once, you need to move up a plan. More in Plans and security.

#Can I move a card straight into any column I want?

Yes. There is no order forced between columns, so you can mark a waiting card as done in one move, or pull a done card straight back into work. The one exception is the execution states of a card assigned to AI (the next question). More in Tasks and Kanban.

#Can I change a card's tags after creating it?

Yes. Type a new tag on the Tags row of the open card and press Enter; click the cross next to a badge to remove it. Every change is saved right away. Agents and clients connected over MCP can add and remove the same tags too. More in Tasks and Kanban.

#Can I see how much of the checklist is done without opening the card?

Yes. A segmented bar sits at the bottom of the card on the board and in the list row, one segment per step. Green is a completed step, blue the one being worked on, red a failed one, and the counter next to it says how many steps are done. A failed step does not count as completed. A card without a checklist shows no bar. More in Tasks and Kanban.

#Why is there no chat tab on a card, and how do I talk to its agent?

Conversation on a card goes one way: through comments. Your comment goes to the card's agent and reaches it live while it works, so the team reads everything about the card in one place. Chats you opened on a card earlier were not deleted and stay in your chat list. When you explicitly ask an agent in a chat to comment on a card, that comment carries a From chat mark. More in Tasks and Kanban.

#Why can't I move a card assigned to AI into the "In progress" column myself?

Because the execution states are set by the agent doing the work. If they could be set by hand, the board could say "an agent is working" while nobody was. If you are taking the work over, assign the card to yourself first, then move it to whichever state you want. More in Tasks and Kanban.

#The task went to error. Is the work that was done lost?

No. Completed checklist steps are kept, and requeue carries the task on from where it stopped. The reason for the error is in the banner at the top of the card and in the comment stream. More in Troubleshooting.

#Can I requeue a card while it is still running?

You can. A new run starts and becomes the card's only owner. The run left behind notices this at its next step and stops, writes nothing to the card's state or history, and only says so in a comment. The credit spent up to that point is not refunded, so waiting is cheaper while the step list is still moving. Orbi moves a genuinely stalled run to error by itself. More in Troubleshooting.

#Do I have to approve every step while an agent works?

No. Execution is autonomous, and the human checkpoint is the Review stage. The balance comes from narrowing an agent's tool access up front (least privilege) and from the budget ceiling per task. More in Tools and integrations.

#I don't like the output. If I write feedback, is everything done again from scratch?

No. The comment you write in Review goes to the Orchestrator first, and it decides which steps to repeat. A comment saying "the third section is weak" does not restart the whole job. More in The Orchestrator.

#The task stopped with "budget ceiling exceeded". What should I do?

It hit the cycle or agent-run limit, which is usually a sign that the task is too big. Split the task up, or skip the planning layer and assign it to an agent directly. Requeue carries on without repeating the steps that are already done. More in Troubleshooting and The Orchestrator.

#I opened the card on the wrong board. Can I move it?

Yes. Use the Board row in the card drawer to move it to another board, and the card number, its state, its comments and its history are all kept. If the target board is in another space, a move report opens first. More in Tasks and Kanban.

#Can I attach files while creating a card?

Yes. Pick files from the Attachments row of the new task form, or drop them onto the form. They are uploaded once the card is created and appear in the card's Attachments section. A card you open with the shortcut does not take you away from the screen you are on either, and you click the confirmation that appears in the corner to go to the card. More in Tasks and Kanban.

#Do I have to copy a key to connect Claude Code to Orbi?

No. You take only the address from the setup guide on the Tool access tab of your profile; when you type /mcp in Claude Code and choose Authenticate, Orbi opens in the browser, you pick the workspace and on whose behalf it should work, and allow it. For Cursor and VS Code the guide has a one-click add button. If you have more than one workspace, each is a separate connection and each goes through the same consent screen. A secret key is only needed for clients without a browser. More in Working with a coding agent.

#Can a client like Claude Code attach a file to a card?

Yes. When the connected client writes a file and names the card number too, the file goes straight into that card's Attachments section and shows up in the card's project. A document that already sits on the Files screen can be attached to an existing card as well; everyone reading the card sees it. Removing an attachment is done in the interface. More in Collaboration.

#Can a client like Claude Code post to a channel or send one person a private message?

It can post to a channel and read it. With an agent token the message appears under the agent's name, with a personal token under yours. It cannot send a private message, because Orbi has no direct messages (DMs) between people. To reach one person, ask the client to mention them with @user.<name> in a channel message or a card comment; that person is notified. More in Channels.

#Should I assign every task to the Orchestrator?

No. If you know exactly what you want, assign the task to an agent directly, because the planning layer is skipped and the work finishes faster and with fewer tokens. Keep the Orchestrator for work that needs more than one speciality. More in The Orchestrator.

#I want several agents to work on one card together. How?

Put the agents in a squad and assign the card to the squad. A coordinator splits the work into items and gives each item to a member of the squad; in a squad with two or more members, another member critiques what one member produced and the producer fixes it. Writing @squad.squad-name in a comment also gives the card to that squad. More in Squads.

#Can I talk to a squad without opening a card?

Yes. Write @squad.squad-name in a channel, or start a new conversation from the Squads section of the Chat screen. The coordinator hands the work to the members it needs, each member answers under its own name, and the discussion shows in one block. If the work grows, one button turns the conversation into a card assigned to the squad. More in Talking with a squad.

#Can squad members work at the same time?

Yes, in workspaces where agents work on their own computer. When the coordinator decides the work needs a computer, a single computer opens for the card and the members work in the same folder at the same time, each with its own model and permissions. Each member changes only the files it was given. More in The squad's shared computer.

#Who does the model I pick in a squad chat apply to?

Everyone. If you pick a model in a 1:1 squad chat, the coordinator and every member of the squad run on it for that message's turn; the members' own model settings do not apply to it. When the members work on the computer, the computer opens with that model too. Members' own models puts every member back on its own model. More in Model, files and computer in a squad chat.

#What does "The turn stopped midway and started again" mean in a squad chat?

The Orbi process running the turn stopped in the middle of the work, and the turn started again by itself. The squad answers the same message from the start, your message is not answered twice, and credits for the computer are still charged once. You do not need to do anything. If the restarted turn stops midway too, write your message again. More in Model, files and computer in a squad chat.

#Who reviews the code a squad writes?

Another squad member that runs on its own computer reviews it; that member can open the branch and look at the changes. A member that runs in Orbi only sees the summary, so it does not review code. If the squad has a single member on a computer, the card comes to review and you are asked to look at the pull request. More in Members that run on a device.

#How do I carry a squad conversation over to a task?

Press Save as task on the conversation block and choose a project. The task is assigned to the squad; its description holds the request, the squad's last answer and a link back to the conversation. The members' work and the whole conversation go into the Outputs tab. More in Talking with a squad.

#Why did the squad's text change, and which version is final?

In the Outputs tab each text is one document, and its header shows the latest version's number and status. How it got here under the document shows who wrote each version and which critique it answered; Changes shows what changed against the previous version. More in Reading a squad card.

#Why did the squad never run one of its members?

The coordinator decides who gets which work from the members' descriptions and the squad's shared instruction; it does not read the members' instructions. A member with no description shows up to it as "(no description)", so it cannot tell which work goes there. The Edit squad screen shows each member's description and warns next to empty ones. If a kind of work should always start with a member (work that needs competitor or market facts starting with research, for example), write that in the squad's shared instruction. More in Creating and editing a squad.

#Can Orbi do things without me asking, like writing a morning summary or reminding someone about a waiting card?

Yes. Turn on Proactive mode on Orbi's agent page. It writes a short briefing to the channel in the morning, leaves a reminder for the owner of a card that has been waiting a long time, and proposes a plan at the end of the week. It stays quiet when it has nothing to say, and it never changes the state of a card on its own. If you answer a suggestion you did not want with "Don't suggest again", it will not raise it again. Available on Team and higher plans. More in The Orchestrator.

#How does Orbi know what we are working on? Does it only look at card states?

No. In proactive mode Orbi reads the last seven days of cards, comments and channel talk every day and works out which subjects the team is working on. For each subject it finds similar work done before and related cards forgotten in the backlog. The briefing talks about these subjects, and reminders go to cards that belong to them. You can see what Orbi read in the Agenda panel on Orbi's agent page. More in The Orchestrator.

#Can Orbi join a conversation with its own input, or does it only summarise?

It can join. When it has evidence in hand it speaks up: it points out two statements on the board that contradict each other, quoting both; it researches a subject that needs an outside look and brings the answer with its source; and it proposes a concrete way it could do the thing being discussed. A contradiction without its quotes or research without its source never gets written. The nudge level decides which contributions arrive, and a kind of contribution that keeps getting no response is scaled back automatically. More in The Orchestrator.

#Can work Orbi started on its own send e-mails or messages to the outside world?

No. When Orbi started a piece of work itself, that work cannot take any outward-writing step on its own; research and reading stay open. When an outward step is needed, the prepared call lands as a waiting proposal in the "Recent notes" list on Orbi's page and only goes out when an admin says "Approve and send". Work you started yourself does not carry this boundary. More in The Orchestrator.

#Can Orbi open a private chat with me on its own? Can I turn that off?

It can, but only when the matter is yours to settle: a reminder kinder asked one-to-one, a decision only you can make, or a sensitive correction. You get a notification when the chat opens, and once the matter is settled Orbi carries the conclusion back to the team's shared surface. At most one chat per person per day. If you do not want this, just turn off the "Orbi may open chats with me" switch on Orbi's page. More in The Orchestrator.

#How do I start a new chat, and how do I find an old one?

On the chat screen, when no chat is selected the right side holds the list of agents: click one and the chat opens, and only you see it. Older chats are on the left, and each row shows the agent you were talking to, the first line of what was said last and when the conversation last moved; the list is split into today, this week and earlier. If you remember the sentence you are after, type it into the search box and clicking a result takes you to the place that message appears. The panel that opens from the right carries the same list, reached with the list button at the top of the panel. More in Working together.

#Can I stop a reply or have it written again?

Yes. While the agent works, the send button turns into a stop button; when you stop, the text you had seen stays in the chat. You can regenerate the last reply, and edit your last message and send it again; nothing is duplicated. If you leave the chat before the answer arrives, the agent finishes and you get a notification when the reply is ready. More in Working together.

#Can I rename a chat?

Yes. Hover a row in the chat list and the pencil button lets you edit the name in the row itself: Enter saves, Esc cancels. A chat's name is normally derived from your first message; once you set one by hand it is never overwritten automatically. The chat's owner and workspace admins can change it. Channel names are managed on the channel screen rather than in this list. More in Working together.

#How do I notice that a new notification has arrived?

While the app is open, a small card appears in the bottom right corner of the screen and disappears on its own after a few seconds, and clicking it takes you where it points. The number on the bell icon shows the unread ones. In the bell panel, notifications are grouped by the day they arrived, and every row says what it is about (Task, Channel, Flow, Plan), which workspace it came from if you work in more than one, and the time. If you have unread ones, an option appears that narrows the list down to those. More in Collaboration and notifications.

#Where do I see when we finished a piece of work?

The completed list groups tasks by the day they were completed. Today, Yesterday and older days sit under separate headings, with the number of tasks finished that day next to each heading. If you reopen a task and complete it again, it appears in the list under its latest completion date. Older releases of the product did not record this, so a few very old tasks may have no date, and those are collected under a separate heading at the end of the list. More in Tasks and Kanban.

#I can't see the search box in Backlog. Where did it go?

Search is no longer separate for each list. The row that opens when you press the Filter button at the top right holds a search box that works across Active, Backlog and Completed together, and what you typed is not cleared when you change scope. The Assignee choice on the same row also applies to all three, so you can pick a person and switch to Completed to see what they finished. When you close the row, the number of active filters stays visible on the button. More in Tasks and Kanban.

#A pull request icon appeared on my card. What does it mean, and does the card close when the PR is merged?

When GitHub is connected, the pull requests that mention a card are linked to it on their own: if the PR's title or branch name contains @task.12 or orbi/task-12, the icon appears on card 12. Grey means the PR is open, green that it was merged, red that it was closed without merging; the list in the card's Pull requests tab opens each PR on GitHub. Merging does not close the card, and moving it to done stays your decision. More in Tasks and Kanban.

#Do review comments written on the PR in GitHub show up on the card? Does the agent read them?

Yes. Every review submitted on the PR lands on the linked card as a comment, with its verdict (Approved, Changes requested, Commented) at the top; inline comments arrive one by one with their file and line number, and the author appears with their GitHub username and a "GitHub" badge. These comments do not count as human comments, so an agent mentioned in them does not start; an agent already working on the card reads the comment on its next turn. If you want an automation such as "fix the PR when changes are requested", add the "GitHub — when a pull request is reviewed" trigger to a flow. More in Tasks and Kanban.

#We use GitLab. Does the card show the pull request state?

It does, but the link is made by the coding agent or by you rather than by a connection like GitHub's. The agent connected to the board writes the merge request link to the card the moment it opens it; you can also paste the link with "Link" in the card's Pull requests tab. GitLab cloud and self-hosted GitLab, Bitbucket, Gitea and Azure DevOps links are recognised. No provider has confirmed that link's state, so the icon is drawn with a dashed border and an "Unverified" suffix. If you merged later yourself, pick "Mark as merged" from the row's menu; the icon turns green and the border stays dashed. More in Tasks and Kanban.

#Two of my workspaces use the same GitHub organization. Do pull requests land on the wrong board?

They do unless you pick repositories. A GitHub installation is one per organization; when two workspaces connect the same organization they share that installation, and @task.82 resolves to card 82 on each board. To prevent it, open the Integrations tab in Settings and use "edit" under the GitHub connection to check, in each workspace, only the repositories that belong to that board. An empty selection listens to every repository in the installation; Orbi warns you on the connection row when the installation is also connected elsewhere. More in Tools and integrations.

#Why do I have to set up a bot to connect Telegram?

Telegram has no account authorization like the other services; the identity is a bot token. In Telegram you message @BotFather, create your own bot with /newbot, and paste the token it gives you into the dialog in Orbi. The bot is yours: you pick its name, its picture and who may write to it. Then you add the bot to the group or channel whose messages you want it to read. In groups a bot only sees the commands addressed to it by default; if you want it to listen to every message, turn off privacy mode for that bot in BotFather. More in Tools and integrations.

#What is the difference between the Zapier connector and the Zapier MCP recipe?

They run in opposite directions. The connector lets an agent push an event to a Zap you built: in Zapier you make a Zap that starts with "Webhooks by Zapier → Catch Hook", paste that webhook URL into Orbi, and the agent calls zapier_send_event to hand work off — one-way, nothing comes back. The MCP recipe goes the other way: with a personal Zapier key the agent calls Zapier's own actions during a run. Both can be connected at once because they do different jobs. More in Tools and integrations.

#Can I send an email attachment or a CV with an inbound trigger?

Yes. Send the file in the attachments field of the body, with its name and base64 content. Orbi saves each attachment to Files in the flow's space; the agent gets the file's id, not its content. The agent reads the file with read_file and attaches it to the card it opens with attach_file. A request can carry up to 10 attachments, 10 MB each. Details: Collaboration and notifications.

#I removed Orbi from my Zoom account. What happens to my data in Orbi?

Zoom tells Orbi, and Orbi deletes that Zoom user's access tokens and account details from every connection they authorized. Meeting lists and recordings were never copied into Orbi. The Zoom row stays with a "reconnect" warning so that flows and agent access that use it do not break; connect again to use Zoom, or remove the row in Settings → Integrations. More in Tools and integrations.

#My Google Meet meeting ended but the flow did not run. Why?

The most common reason is that transcription was not on during the meeting: with nothing for Orbi to read, the meeting never triggers. Even with it on, the transcript is not ready the moment the meeting ends — Google produces it a few minutes later, and Orbi holds the finished meeting for up to half an hour and fires when the transcript arrives. The third possibility is permission: the Meet card added a new permission to the Google connection, so if you connected Google before this feature you need to reconnect once. More in Flows.

#How do I see what a flow or a card has produced?

With focus. Pick the flow or the card in the picker that opens with ⌘O (Ctrl+O), and the board, the collection screens and Files then show only what came from that work, while the headings in the left menu say how many items it left in each place. The same button sits on the card detail and above the flow. Focus stays until you remove it, and the bar at the top removes it with one click. More in Focus.

#How do I see notifications and due dates from my other workspaces in one place?

That is what the Inbox in the left menu is for. The Notifications tab shows what arrived from all of your workspaces in a single list, and each line's footer says which one it came from; the Awaiting action choice keeps only the ones that expect something from you. The Today box on the right shows the due cards and events of every workspace you belong to — the due date stays in sight even while you work somewhere else. The workspace menu's switch list also carries, next to each workspace, the number of notifications waiting for you there. More in Inbox.

#Will I hear about it when my card reaches its due date?

Yes. You get a notification when a task assigned to you reaches its due date and when a calendar event you created starts; every morning a short summary counts the day's due dates and events. These notifications are managed in your profile's preferences under Due dates and calendar; email arrives switched off, turn it on if you want it. More in Inbox.

#Where can I see everything the agents did, in one place?

The Activity tab of the Inbox gathers the work of your agents, your teammates and your flows into a single list. The tab opens on the Agent work view. You can narrow the list with the Who, What and When filters, and click a run to reach what the agent did step by step. For runs that ended in an error and proposals waiting for your approval there is a Needs attention view. More in Activity.

#What are the two lists I see when I open a run?

The upper list shows what the run wrote: which card it commented on, which record it updated, which field it added. For an agent's writes, the reason sits under the line. The step disclosure below it shows how it did that, which tool it ran and in what order. Day to day you read the upper one; the lower one is for working out where something went off course. A run that wrote nothing says so. More in Activity.

#The chat or a card opens on top of the page. Can I pin it to the side?

Yes. The cards and chats you open share one column on the right of the screen, as tabs. Press the pin button at the top right of the column and it sits beside the page and pushes the page aside. It stays open in the same place whichever screen you move to. Your browser keeps the choice, so you do not pick it again next time. If your screen is narrower than 1280 pixels the column still opens on top of the page. More in Tasks and Kanban.

#Can I keep several cards open at once?

Yes. Every card you click on the board opens as a new tab in the right column, and the earlier ones stay where they are. Chats sit in their own tabs too. When many tabs pile up, the ▾ menu in the strip shows them all, and Close others or Close all tidies them away. The tabs are still there after you reload the page. More in Tasks and Kanban.

#Agents and memory

#The agent said "I cannot" about something Orbi does. Why?

A feature may be missing from the agent's tool list while the product has it; the agent now checks the Help Center with read_help before saying no, and quotes the page address. If you still get a wrong answer, try the same question in the Help Center search and tell us the page. Details: The agent knows the product.

#Where do I set an agent's model and repository?

Both on the agent's Work tab. At the top you pick where it runs (inside Orbi, or in your environment), and that choice shapes the rest: with "runs inside Orbi" you get the model choice right below it, with "connects from outside" you get the repository and base branch. Connection access, setup and tokens live on the same tab. Name, title, description and avatar stay fixed at the top of the page (visible on every tab); the Definition tab only holds the squad. Whichever tab you type in, there is a single Save at the top of the page, and a tab holding unsaved changes shows an amber dot next to its name. While creating a new agent there are no tabs; the whole setup flows on one page. Detail: Agents.

#How do I make an agent think harder?

Raise the reasoning effort. Pick High with the button next to the model picker in a chat, and the choice goes with the messages of that conversation; for one card use the Reasoning effort row in the card detail, and for all of an agent's work use the setting on the agent page's Work tab. More thinking can give better answers on hard work, but it is slower and costs more. Detail: Reasoning effort.

#Why does Orbi ask me a question instead of starting right away?

Because not asking a question whose answer changes the outcome costs you far more later. Orbi looks at your workspace first, tells you what is already set up, and stops only where a decision is really needed. When the answer is one of a few clear choices it gives you clickable options and says what each one leads to. A confirmation is a different thing: before creating a new table, flow, task or document Orbi says what it is about to build and waits for your yes. It does not wait to write a record into an existing table, add a field or change a state — otherwise every small request would cost two turns. Detail: Collaboration and notifications.

#What is the confirmation card under the answer for?

Before creating a new table, flow, task or document, Orbi shows you on that card what it is about to build: its name, what goes in it, what it will do. Nothing is created until you press Build it. What you read on the card is exactly what gets built — once you approve, Orbi cannot change its mind, because the thing that runs is the very request shown on the card. If you do not like it, there is no button to press: you say what you want instead, and the old card closes as the new answer arrives. If something cannot be built for any reason, the conversation does not stop there: Orbi tells you what failed and why, and offers a corrected version on a new card. Writing a record into an existing table, adding a field or changing a state needs no confirmation; otherwise every small request would cost two turns. Detail: Collaboration and notifications.

#Can I talk about one specific record or file?

Yes. When you hover a record, a "Talk to Orbi" button appears at the right of the row; pressing it starts a conversation about that record in the panel that slides in from the right, without taking you off the screen you are on, and the top of the panel says what you are talking about. In that conversation the record is the subject, so your changes apply to it; other things that come up stay as context. Pressing it again on the same record continues the previous conversation instead of starting a new one. The same button exists on files, flows and agents. Detail: Collaboration and notifications.

#What does Orbi do itself, and what does it leave to me?

It does reversible things that only concern you itself, after telling you what it is about to do: creating a collection, adding a field, building a view, writing a draft, putting an event on the calendar. For things that cannot be undone or that others can see, you press the button: saving and activating a flow, connecting a service, running a bulk update, publishing a screen to the menu, sending anything outward. Orbi never deletes anything on its own initiative. Detail: Collaboration and notifications.

#Can I try an agent without opening a real task?

Yes. Write a sample request into the Try it card in the right column of the agent page and run it. The agent produces a single answer with the brief as it stands in the form, and you do not have to save first. No task, record or file is written to the board, because the agent can only use read tools. The run is a model call, so it is charged to your credits. More in Agents.

#Why doesn't the agent remember what we talked about in the last task?

This is a deliberate design decision. Raw history does not travel between runs, and knowledge specific to a piece of work is handed over explicitly with a mention. What does travel are the short lessons distilled from runs. They are kept in workspace memory and repeated to the agent at the start of every run, so you do not have to say "from now on, do it this way" twice. The other places lasting knowledge lives are the project description, the agent definition and collections. More in Memory and context and Workspace memory.

#In which timezone do agents read and write times?

In the workspace timezone. It is taken from your browser when the workspace is first created and can be changed under Settings → General. Agents read comment and file timestamps in this zone, talk to you in it, and evaluate phrases like "today/yesterday" against its calendar day. Times on screen always render in your own browser timezone. Flow schedules are separate: a flow runs in the timezone chosen when it was set up. More in Memory and context.

#Does the system learn it when I say "from now on, do it this way"?

Yes. Explicit instructions like that are saved to workspace memory as a pinned note (📌) and repeated to the agent at the start of every run. Lasting facts you share in ordinary conversation are also reviewed in the background and taken down as notes. You manage the notes from Settings → Memory. More in Workspace memory.

#One-off requests like "the user wants…" are piling up in memory. Why?

Work you ask for in a chat ("list the projects", "describe the image") does not go into memory, because a request says what you want done right now and nothing lasting about you. The system draws that line in the background review, and an agent cannot write notes on its own during a run, so it saves only what you explicitly told it to remember. Notes that piled up earlier can be deleted from Settings → Memory, and a deleted note goes to the archive where it can be restored. More in Workspace memory.

#Does the place, budget or health detail I give while an agent sets up a flow go into memory?

No. When an agent sets up a flow, a collection or a calendar event, the details you gave in that conversation are settings of the thing it built and stay there; they are not written to memory as notes as well. Health, belief, politics and personal money details never become notes in the background review either. If you want one remembered, tell the agent plainly "remember this". More in Workspace memory.

#Do the things I type in Claude Code go to Orbi?

Not unless you turn it on. The "Learn from my Claude Code chats" switch under Profile → Memory starts off. If you turn it on, only the messages you typed are sent to Orbi at the end of a session, and lasting preferences and rules are extracted from them. For a note, the rule has to be said with words that make it last, such as "from now on" or "always". The model's answers, the context compaction summary and files are not sent. The switch belongs to you: turning it off turns it off on all your computers, and it stays off when you renew a token. More in Workspace memory.

#What do labels like "rule", "decision" or "stale" and the topic headings on memory notes mean?

Every note is stored with a type (rule, fact, lesson, decision, constraint, reference); a decision also carries its rationale. Notes are grouped under topics, and a note that fits no topic lands in the Inbox. You see "conflicted" when an agent learned something that contradicts a pinned rule, and the decision is yours; "being checked" marks a new rule held for a few minutes because it resembles a pinned rule, and it becomes active on its own when no conflict is found; "stale" marks an agent note no run has recalled for a long time, which stays out of runs until you confirm it. Admins shape the topic headings. More in Workspace memory.

#I connected from Claude Code and Orbi suggests a "memory protocol setup". What is it, should I accept?

A client that connects from outside normally keeps its memory on its own machine; the protocol setup makes that knowledge accumulate in Orbi instead. It adds an instruction section to the repository, optionally a hook that pulls memory at the start of every session, and a gate that asks once at the end of a session whether a lesson is still unsaved. Nothing is written until you approve the files; if you do not want it, say "do not ask again" and the suggestion stops. More in Workspace memory.

#Does the agent running a card on my device see the team's memory and its own definition?

Yes. When the card goes to the device, the definition on the agent page, the pinned rules, the topic headings of memory and the notes of the card's project are sent before the card text; the agent does not have to ask for them. When the run ends it reports any lasting lesson together with its result, and the note enters memory linked to the files the run changed. More in Devices.

#I wrote a comment but the agent did not reply. Why?

Two conditions have to hold at once. The comment has to be written by a human (agent comments do not trigger another agent, which is the loop protection) and the @agent.<name> mention has to be complete. Every comment is processed at most once, and editing a comment does not run it again. More in Troubleshooting and Collaboration.

#If I change an agent's job description, who does it affect?

Everyone who uses that agent. Agents are open to the whole workspace and there are no personal agents. To try an experimental prompt, create a new agent instead of changing an existing one. More in Agents.

#Where do I find the address of the service I want to connect?

For most of them you do not have to look it up. The Add MCP button opens a ready-made list of common services (GitHub, Notion, Linear, Stripe, Supabase and others). Pick one and the address and the connection method fill themselves in. You can type in the address of a server that is not on the list. More in Tools and integrations.

#Why is a service missing from the ready-made list?

Some services accept connections only from applications they have approved themselves, and if Orbi is not on that list, the service turns you away at the authorization step. We keep such a service off the ready-made list, because it would promise a connection that cannot work. Entering its address by hand does not change the outcome. More in Tools and integrations.

#I connected my account but "Show tools" fails. What should I do?

The error text names the cause. If it says the server does not serve MCP at this address, the service is hiding a feature that is off for your account; turn the prerequisite on in the service's own settings (on GitLab, Duo and beta features). If it says your credentials were rejected, use the button next to the error to renew the authorization or replace the key. If it says the server could not be reached, the service is not answering right now; try again in a moment. More in Tools and integrations.

#I added an MCP connection but the agent does not see the tools. Why?

Adding the connection is not enough, because agents other than the Orchestrator are given access to a connection explicitly (Agents → agent detail → Connection access). If Orbi works but your other agents do not, this is almost certainly the reason. More in Tools and integrations.

#I removed a connection and added it again, and the agent no longer uses the service. Why?

The access given to agents is deleted along with the removed connection and does not carry over to the new one. Tick the agent in the "Which agents should use it?" step that appears when you add the connection, or open access on the Work tab of the agent page. Using Reconnect instead of removing keeps the access when you switch accounts. More in Tools and integrations.

#What happens when a tool of a service I connected has the same name as a built-in tool?

Both show up in the tool suggestions, and the later one is listed under the name it gets in a run: the built-in keeps create_task and your connection's tool becomes depo_mcp_create_task. The mention you write into a prompt therefore always points at the tool the agent really calls. The icon in the suggestion and in the text says which one it is too, because a built-in tool gets a wrench and a tool of a connected service gets a plug. More in Agents.

#Why does an agent "open the connection" before it uses a service?

Connections with a lot of tools are not loaded into every run up front. The agent sees the name of the connection first, opens it when the work calls for it, and keeps those tools for the rest of that run. The extra step is there so that connections nobody uses are not paid for. No tool is lost. More in Tools and integrations.

#Is my personal connection used when a teammate runs the same task?

No. A personal connection is only loaded in runs you started, and when another member queues the same task their connections apply. Scheduled work has two extra conditions. More in Tools and integrations.

#A policy event says "Warning" and the content was written anyway. Is the policy not working?

It is working. That policy is set to Watch only: content that trips the rule is written and the event lands in the list on Settings → Policies, along with which agent was writing where and the part that tripped the rule. If you want the content not to be written at all, open the policy and use Edit to switch it to Block. Details: Policies.

#How do I know a policy I wrote really works, before an agent writes something?

On the second step of the policy window, paste a sample text and press Try with the sample. The rule runs as it would in a real write and shows at once whether the text tripped it, with the reason; the trial is not written to the event list. An AI review trial spends credits, a pattern review trial is free. Details: Policies.

#Why did Orbi tell me a member has not signed in?

You are Orbi's contact (or an admin) and Orbi's housekeeping behaviors are on. Every morning Orbi checks the workspace once and writes what only admins should see: members who have not signed in for a while, unused tokens, unusual credit spending, flows whose creator left, storage that is nearly full. It changes nothing itself in any of them; the decision is yours. You close a finding you do not want with Don't suggest again, and you turn a behavior off completely with its switch on Orbi's page. More: Behaviors and housekeeping findings.

#What does Orbi do at night, and does it spend credits?

In workspaces where the agent computer is on, Orbi reads the board, the channels and the file list on its own computer at night, and prepares for the morning: promises that were not kept, work that drifts from a written decision, files in the wrong folder, unanswered questions, watched topics and your rules. It writes nothing at night; the morning rhythms deliver what it prepared. The night run spends credits and is limited by a monthly budget the admin sets (500 credits by default). More: Orbi's night notebook.

#How do I ask Orbi to keep an eye on something? How is that different from a flow?

Write one sentence in the Things to keep an eye on section of Orbi's page, such as "tell me when a card promised to a customer passes its date". Orbi checks it every night and leaves what it finds in your From Orbi tab. A rule watches for a situation; doing work at a set time (writing a report every morning) is what a flow is for. Admins write rules. More: Orbi's night notebook.

#With several admins, who does Orbi ask?

Orbi's contact, picked with Who should Orbi ask? in the proactive settings on Orbi's page; if nobody is picked, it is the admin who turned proactive mode on. The other admins see the question and the decision, but only the contact has the buttons. If the contact has not signed in for 14 days, Orbi tells the other admins. More: Behaviors and housekeeping findings.

#Coding agents

#Where do I find my tokens and connections?

On the Tool access tab of your profile (your name at the bottom of the left menu → Tool access). The Access tokens list shows the tokens of all your workspaces, grouped by workspace: the ones you created by hand, the connections you granted on the consent screen and the agent tokens you issued. In the workspaces where you are an Admin, agent tokens other people issued are in a separate Other people's view. Open a row to revoke the token or choose closed spaces. A new token is created with Create token. The agent page and Settings → Integrations have no token list; the agent page only shows the connection status. More in Collaboration.

#How do I use a coding agent on two different projects (two repositories)?

Create a separate "connects from outside" agent in Orbi for each repository and generate a token for each. One agent cannot cover two repositories, because which repository to work in is decided by the client's setup in that repository, not by the card. Cards reach the right repository through the agent you assign them to. Both can run at the same time. More in Coding agent.

#How do I give the same instruction to several agents?

Group them into a squad. The shared instruction you write on the squad is added to the start of every agent's own brief in it; the agent's own text stays below, so it can override the shared rule when it needs to. You update the instruction in one place, and every agent in the squad uses the new version right away. More in Squads.

#What happens if the agent my coding agent uses is deleted?

The connection tokens are deleted along with the agent, so on its next connection the client on your device gets "Authentication rejected" and stops. Cards assigned to that agent are not deleted but are left without an owner, and the cards wired to it on the flow canvas are removed too. To carry on, create a new "connects from outside" agent, generate a token, connect your client with the new token and reassign the cards to the new agent. More in Agents.

#Can someone who left the team keep reaching the board through an agent connection they created?

No. When a member leaves or is removed from the workspace, the agent tokens that person issued are revoked on their own. A Claude Code or script left on their device is refused on its next call. Tokens issued for the same agent by another admin keep working. As an Admin you see them in the Other people's view under Profile → Tool access; each row names its issuer, and you can revoke it there if needed. More in Agents.

#What happens to a card I queue while the coding agent's device is off?

The card waits in the queue marked "waiting for the client". The mark shows that the client is offline and when it was last seen, and when you open the card you see how long is left before it moves to error. If the device comes back within 6 hours the card stays where it is, so you do not need to queue it again. If nobody takes it within 6 hours it goes to error, the reason is shown on the card and written to the activity log, and no card is quietly forgotten. In that case you queue the card again once the device is back on. More in Agents.

#The card came back saying nothing was saved, but the agent was waiting for a helper agent. Why?

In the current version the agent reports how the run ends itself, and the card closes accordingly. If the card came back saying the agent "ended without reporting a result", it closed without reporting that the work was done, that it had a question, or that it needed you; exiting on its own while waiting for a helper agent looks the same. Its last words are in the card's comment. The device does not open a second turn in that case, so no model time is wasted: queue the card again to retry. On older device versions the card comes back saying nothing was saved, and the device continues the same session once. Details: Connecting your own device.

#I paused the device, its connection dropped, and it started taking cards again when it reconnected. Why?

In the old version the device reset the pause while reconnecting. A pause set on the Devices screen now survives a dropped connection; the device only takes cards again when you resume it. The pause on the device itself is separate: if either one is on, no card is taken. Details: Connecting your own device.

#The device shows as connected but takes no cards, and pressing Resume did not help. Why?

The device is most likely paused on its own machine (orbi-node pause). You cannot lift that pause from Orbi: the Devices screen shows the device as "paused on its machine", and the card's badge says the same. Run orbi-node resume on that machine and the device takes the waiting card right away. Details: Connecting your own device.

#Can I tell the coding agent something, or stop it, while it works?

Yes, from the code card's drawer. While a run is on, what you write in the box at the bottom with Agent selected goes to the agent; it reads it at the end of its turn and continues in the same session. This is not the same as a comment: comments are written to the team, this message goes to the agent. "Stop" on the status line halts the tool within seconds, keeps the work done so far on the device and returns the card to you. "Stop after this turn" in the ⋯ menu lets the agent finish its sentence. Details: Connecting your own device.

#The agent asked a question and the card stopped. What should I do?

On the board the card says "Waiting for your answer". The question sits in the code card's drawer, above the message box; pick one of the options or write your own answer. The agent carries on in the same session where it left off; you do not need to queue the card again. If no answer arrives within half an hour the card comes back to you with the question; the question stays on the card, and Answer and queue writes your answer on the card and queues it again. Details: Connecting your own device.

#How do I have the agent only answer, without committing anything?

In the code card's run settings, set Expected output to Answer only. The agent then commits nothing, opens no branch or pull request, and writes its answer in the card's closing comment; the card comes to review. Writing "do not commit" in the card's description is not enough, because this field decides how the card closes. The default is Code changes. Details: Connecting your own device.

#My watcher says "reached today's limit". What happened?

Every watcher has a daily event limit (500 by default) and this one reached it today. It produces no events and starts no flow for the rest of the day; it carries on by itself tomorrow, nothing to do on your side. If this happens often the watcher is reporting more than it should: in a log-file watcher narrow the pattern or lengthen the gathering time; if you really expect that many events, raise the limit in the watcher's settings. As the device's owner you also get a notification; switch off the "My connected devices" category in your notification preferences if you do not want it. Details: Connecting your own device.

#I got a "the agent's client is not connected" message while queuing a card. What should I do?

The message says the agent's client is not connected at that moment and the card will wait until the client starts, and it does not stop you from queuing the card. If you are about to turn the device on, choose "Queue it anyway" and the client picks the card up once it is running. If the message says no token has been created, first press Create token on the Tool access tab of your profile, pick this agent under Acts as and connect the client with that token, because otherwise nobody can take the card and it moves to error after 6 hours. More in Agents.

#Does Orbi see my source code?

No. The code never leaves your device, and Orbi only sees what the agent writes on the card, such as a comment, a summary or a pull request link. Your git credential and your model provider key never reach Orbi at all. More in Coding agent.

#What happens if the coding agent wants to ask something while it works?

It cannot ask the terminal, because the card is the channel. The agent writes the question as a comment on the card, assigns the card back to whoever opened it and stops. You write your answer and move the card back to Queued, and the next run reads the answer along with the rest of the card. More in Coding agent.

#Can I work with a tool other than Claude Code?

Yes. Claude Code, Gemini CLI, Qwen Code and Codex connect to Orbi on the device for every run on their own: the agent reports its result when the work is done, asks on the card when it needs to, and the card closes accordingly. Cursor, Copilot, OpenCode, Droid, Amp, Aider and Antigravity run too, but cannot report their result to Orbi yet; with these tools the card closes by the older rule, so it comes back to you when the branch has no commit. A command you wrote yourself follows the same older rule. Details: Connecting your own device.

#Can an agent that connects from outside delete a card?

Yes, a card, and also a file it wrote itself. The delete_task tool removes a duplicate or an accidentally created card from the board. An agent only does this when a person asks for it explicitly, and it does not delete a finished card but moves it to Done. The delete_file tool moves to the trash only a file the agent wrote itself, that is not an upload and is not attached to any card; the file can be restored from the interface for 30 days. Who deleted it stays in the activity log. Collection records and other people's files cannot be deleted from outside, because those need the interface. More in Collaboration.

#Can I build a flow or an agent from a client like Claude Code?

Yes. Externally connected clients can build and update flows, define agents and, with an admin token, open projects — the same rules as the interface apply. Two deliberate limits: a flow built this way is born paused and putting it live is up to you, and the access token of an externally connected agent is still issued by an admin in the interface. More in Collaboration.

#Connecting your device

#Can Orbi see every file on my device?

No. Orbi Node sees only the folders you add, and for each folder you decide whether it may only read or may also write. A request that reaches outside a folder is refused by the application itself, and shortcuts and links are resolved and checked as well. Password files, keys and repository settings are never read by default. Details: Connecting your own device.

#Can I change my device's name later?

Yes. You can change it from the Settings screen of the application on the device, or from the Devices screen in Orbi as the owner of that device. The name is also the address agents use to reach the device, so before you save you are told how many flows and cards mention the old address. That text is not changed for you: a flow left with the old address loses its permission to this device the next time it is saved. Details: Connecting your own device.

#Can an agent run any command it likes on my device?

No. An agent can only do predefined things: the capabilities that come with it (reading a file, listing a folder, looking at system status) and the ones you define. A draft prepared for a job you asked for is not written to your device until you have seen its full text and approved it. Details: Connecting your own device.

#Where do I switch ready-made packs (office files, for example) on and off?

In the Orbi Node app on the device. Every pack has a switch next to it on the Abilities screen. On a server without a screen, orbi-node verbs enable office does the same job. The Devices page in Orbi has no such switch; it only shows the capabilities of the packs that are on. When you flip the switch the list updates on its own. Details: Connecting your own device.

#A card came back saying nothing was saved. Is the work lost?

No, the work is still on your device: the copy the card ran in is left exactly as it was, and the card says where it is. The reason is almost always one of the three things that have to be set up once on the device: the coding tool being able to run commands, git knowing whose name to save under, and sending not asking for a password. If the tool was refused a command while it worked, the card says so and shows the command itself — that line is usually the answer you are looking for. Fix what was missing, queue the card on the same device again, and the work carries on from where it stopped. Detail: Connecting your own device.

#Can someone else on the team have work done on my device?

Only the people you add to the access list. While the list is empty, only you can; if someone who is not on it mentions your device on a card, the work is not done and the reason is written on the card. A flow doing something on your device is a separate permission that you approve once. Details: Connecting your own device.

#I am an admin; why can I not grant access to someone else's device?

Only the device's owner opens access. As an admin you can pause the connection, transfer it or close it, but you cannot let anyone have work done on that device. Closing is the team's safety; opening is the owner's call. Details: Connecting your own device.

#The card says there is no access to the device, but I do have access. Why?

The sentence is not about you. Which device takes the card is decided by the access of the person who queued it. If that person is not on the list, no device takes the card, even when you can reach it yourself. When you open the card you see the name of the person who is blocked, because the thing to do is add them to the access list. Details: Connecting your own device.

#Does the agent see a comment I write while the card is running on my device?

Yes. Every time the agent touches Orbi (reads the card, writes a comment, pulls a list) the new comments written up to that moment are put in front of it, and it reads them first. A comment reaches the agent on its next touch; a very short run that finishes without touching Orbi at all can miss it. It sees a screenshot you attached to the card too: when it reads the card, the image comes alongside the text. Detail: Connecting your own device.

#What happens to cards while my device is off?

They wait. A device that is off does not take cards; it picks up the next one when it comes back. If it was holding a card as it shut down, that card does not fail — it goes back to the queue and another suitable device can take it. Details: Connecting your own device.

#Can I change a capability on my device later?

Yes. On the Devices screen, open the capability and press "Edit". You change the description and inputs by hand, and for the command itself you either tell Orbi what else it should do and have it rewritten, or edit the text yourself. The change goes to the device as a new draft; the old version keeps running until you approve the new one on the device. Not available on Windows devices yet. Details: Connecting your own device.

#The card shows an "unfinished work" badge. What does it mean, and what should I do?

A run ended without pushing its work to the branch, and what was done stayed on that device. If you queue the card again, only that device takes it and continues where it stopped; nothing is lost. If the device is off, the card waits an hour and then comes back to you. If you would rather not wait, choose Release from the ⋯ menu in the drawer: when the device comes back it pushes the work to the branch, and the card can continue on another device. Details: Connecting your own device.

#Can the same agent work in two different repositories?

Yes. Add more than one repository to the agent; the first is the default. A card assigned to that agent shows a repository selector under its message box, and you pick which repository the card works in. For work that touches both repositories, open a parent card, write the shared contract in its description, open one subtask per repository and link them to the parent; when a subtask goes to a device, the parent's description goes with it. More in Coding agent.

#Can I make a card always run on the same device?

Yes. Under the code card's message box, before the run starts, pick the "preferred device"; only devices that can take the card are listed. While that device is on, only it takes the card; if it is off, the card does not wait and the next suitable device takes it. The choice stays on the card. Details: Connecting your own device.

#I installed it on my server, but the device keeps showing as offline. Why?

Connect to the server and run orbi-node status first: the output says whether the application is running and, if it is not, what it last said. There are two common reasons. You may never have started it — orbi-node start is enough, and it keeps running after your connection closes. Or you did start it and the server has since restarted; that needs a service. A service definition has to carry the application's full path, not just its name: with only the name the service cannot find it and drops into a restart loop. The step-by-step recipe is in the guide. Details: Connecting your own device.

#Can I open the interface on a server where the node runs under systemd?

Yes. The interface never runs a node of its own — it watches the one that is running. If orbi-node is already running on the server, typing just orbi-node in the terminal does not open a second connection: the screens still work, and changes you make there — adding a folder, approving a capability — reach the service too, while the live feed is read from the service's own audit log. The service's connection to Orbi is not affected. Details: Connecting your own device.

#I updated Orbi Node but the Devices screen still shows the old version. Why?

Because the version shown there is the one currently running on that device, while the install only replaces the file on disk. The running application keeps running its old version until it is restarted — so the number on the screen is right, the update simply has not taken effect yet. If you installed it as a service, sudo systemctl restart orbi-node settles it; if you started it by hand, orbi-node stop and then orbi-node start. The desktop application has no such step, it updates itself. Details: Connecting your own device.

#Models and cost

#I have an Anthropic key, but the run stopped with a Gemini error. Why was my other key not tried?

Because the key pool rotates inside a single provider. The model decides which provider is called. If the run's model is a Gemini model, only your Google keys are tried in turn, and the run stops when all of them return a quota or overload error. Orbi does not move to another provider on its own, because that would change both the model doing the work and the account that gets billed without you knowing. To run on your Anthropic key, choose a Claude model on the agent, on the flow or as the workspace default. The error message also names the model and the provider it is tied to. Details: Models and providers.

#When does my credit renew? On the first of the month?

No. Your credit renews on your subscription's billing day. If you subscribed on the 28th, your quota refreshes on the 28th of every month. On the free plan the period runs from the day you created the workspace. The date range you are looking at is written on the Plan page under Settings. Details: Plans and security.

#My credit is about to run out, can I buy more without waiting for the next period?

Yes. The Plan page under Settings has a button for buying a credit pack, and once the payment completes the pack is added to your balance right away, on top of whatever you have left. You do not need to wait until your credit is gone to buy one. Purchased credits do not expire at the end of the period, and whatever you do not use carries over. Details: Plans and security.

#Can extra credits be bought automatically before mine run out?

Yes. If you turn on automatic top-up on the Plan page under Settings, extra credits are charged to your saved payment method as soon as your remaining credits fall below the number you set, so work running overnight does not stop. You also decide how many charges can be made in a single period, and every charge is reported to you. The setting is off by default, so no charge is ever made until you turn it on. You can also set how much money may be spent in a single period. If the credit pack price goes up, automatic top-up stops and asks you to approve the new amount, and your card is never charged an amount you did not approve. Details: Plans and security.

#Are running jobs cut off when my credit runs out?

No. Work that is already running is never cut off in the middle. Only new runs stop starting, and scheduled work is skipped until the credit renews. You get an early notice at 75% and 90%, and a "ceiling reached" notification at 100%. More in Plans and security.

#I added a key but some of the provider's models are not in the list. Why?

When a key is added, Orbi really calls that provider's models one by one and only enables the ones that work with your key. Models with no quota, retired models and models with no tool-calling support are dropped, because it is more helpful for such a model to stay off the list than to appear there and stop on the first run. More in Models and providers.

#Where do I see how much a task spent?

On the card's Trail tab, which shows which agent ran with which model, how many tokens were spent, and whether the fallback stepped in. Usage and cost totals are on the Dashboard, where the breakdown by model and agent covers the last 30 days and the spend badges show this month and all time. More in Models and providers.

#The model supports images, but the agent said "this model cannot read images". Why?

The provider recently refused an image sent to that model. When that happens, Orbi sends no images to the model for 7 days, and the agent tells you it could not see the image. Once the period ends the model is retried automatically, because a temporary error never switches a model's image support off for good. If you do not want to wait, pick another image-capable model for that job. More in Models and providers.

#Collections and files

#The agent said "a similar record already exists" and did not write. Is this a bug?

No, semantic duplicate prevention is doing its job. A new record is compared with the existing ones by meaning, so the same content written in different words is caught as well. If the record really is new, the agent can deliberately push it through with force_write. More in Collections.

#Duplicate prevention is too strict and blocks records that are genuinely different. Can I adjust it?

Yes. The Duplicate protection section on the collection's Settings screen lets you turn the protection off or choose how tight it is, with the options Loose, Normal and Strict. The tighter the protection, the more records are blocked, so in collections where you keep legitimate variants of the same content, such as two platform versions of one post, choose loose. You change which field the comparison looks at from the same place. More in Collections.

#I created a field with the wrong type. Can I change it later?

Yes. On the Settings screen, use the change-type button next to the field to pick the new type. Orbi first shows how many values will survive it, how many will be emptied and a few examples of the ones that will be emptied. Nothing is converted until you confirm the loss. User, relation, file, image and agent columns cannot be converted. More in Collections.

#Can I choose how a date field is written?

Yes. The default is 30.01.2026, but the field settings let you pick 30 January 2026, 01.2026, January 2026 or just 2026, because showing a day in a founding year is pointless. The choice changes only how the value reads, and since what is stored is always a full date, sorting and filtering keep working. You do not have to match the format while writing either: 30.01.2026, 30 January 2026 and 1998 are all turned into a date, and values written by agents go through the same path. Details: Collections.

#I wrote headings and lists in a record but the table shows plain text. Why?

The table is a comparison surface, not a reading surface, and there every row has to be the same height, so cell text stays on one plain line. Open the record and the same text reads as formatted text: headings, lists, tables, quotes and code appear the way you wrote them. If the record runs long, drag the small handle in the middle of the drawer's left edge to widen it; the width you choose is remembered. Details: Collections.

#Can I link two records in the same collection to each other?

Yes. When you add a relation field, pick the collection itself as the target, because that is how you link the Turkish and the English record of one article, or a record to the one above it. A record cannot link to itself, and it does not appear in its own picker. If you want the link to work both ways, fill in the field on both records, because the reverse direction is not written for you. More in Collections.

#I changed the column order but my teammate still sees the old one. Why?

There are two separate orders. The one you set from the Columns panel in the table belongs to the view that is open, so for everyone to see it you have to save the view. To change the schema's own order, use the Fields section on the Settings screen, because the order you set there applies to everyone. More in Collections.

#I renamed a field but the records' updated time did not change. Is that normal?

Yes. The updated time moves only when the record itself is edited, whether one by one, in bulk, or when an agent column fills it in. Renaming a field, deleting one or converting its type are operations on the collection's structure, and you would not want a thousand records to all show "edited just now". When you sort the list by updated time, the record that was really edited last comes to the top. More in Collections.

#I went to a later page in a collection and refreshed, and it jumped back to the start. Can I stop that?

It no longer jumps back. The page and the Per page count you pick from the bar below the table are kept in this browser, per collection, so a refresh carries on from the same place. From the same bar you also click page numbers directly, without stepping Next/Previous one at a time. This is not a view setting and does not change your teammate's screen. More in Views, filter and table.

#How do I work out why a collection is set up the way it is?

The activity button in the header of a collection screen opens a drawer showing who did what in that collection and when. Fields being added, fields being changed and records written by agents all appear there, and agent writes come with the reason they gave. For a single record the same list sits in the record's Activity tab. No time range is applied to these lists, so the very first setup decision from months ago is still visible. More in Activity.

#A link the agent gave as a source does not open. What should I do?

An agent can now only write addresses that were really seen in that run. An address that does not appear in text you wrote, or in a search or page-read result, does not make it into a record, a comment, a deliverable, a file or a channel message. Links left in older records have to be fixed by hand. More in Collections.

#Why doesn't the agent update the file I uploaded?

This is by design, because the files you upload are read-only for an agent. The agent writes its result as a new file (an artifact), so your source always stays where it is. Renaming a file and changing its scope are yours to do as well. More in Files and artifacts.

#Can an agent produce an image for me?

A drawing, yes. An agent can write a diagram, an icon or a chart as an SVG file, and the file shows up as a picture right in the list; when you ask, the same agent can open it and change it. The other formats it can write are Markdown, plain text, JSON, CSV, HTML, Excel, PDF, XML and YAML. Do not expect photographs — what it can produce is a drawing. More in Files and artifacts.

#If I star a file, does the team see it too?

No. Favorites are yours alone: the star shows up in your list only, moves the file to the top of it, and the Favorites button brings back just the ones you starred. To organise something with the team, use a tag instead — a tag is the shared layer everybody sees. The star does not break the sort you picked either. More in Files and artifacts.

#Can I get back a file I deleted?

Yes. A deleted file waits in the trash for 30 days and can be restored during that time. If you do not want to wait, you can remove the file from the trash immediately with Delete permanently. Deleting covers the file's past states too, and so does restoring. More in Files and artifacts.

#Can I upload SQL, TypeScript or Java files?

Yes. Common code files (.sql, .ts, .js, .py, .java, .kt, .go, .cs, .php, .rb, .sh, .css) can be uploaded, agents read them like plain text, and the preview shows them in colour with line numbers. .env files cannot be uploaded, because they hold passwords and keys. More in Files and artifacts.

#I moved files to the trash but my storage did not free up. Why?

A file in the trash still has its content stored so that restoring stays possible, which means it keeps taking up your space and still counts towards your quota. To free space, delete the file permanently from the trash, and admins can empty the whole trash in one go. More in Files and artifacts.

#What happens to the files if I delete a board?

Nothing. Files, collections, folders, flows and conversations belong to the space, not to the board, and they stay where they are. Only the board's tasks are deleted permanently, and the confirmation counts them before you agree. More in Deleting things.

#How do I go back to an older state of a file?

Click the file and open the History section in its detail, then pick the state you want by its date and choose Restore this state. History starts collapsed, because day to day you only care about the current state of the file. More in Files and artifacts.

#The folder has more files but the list shows 50. Where is the rest?

The list loads 50 at a time. At the bottom it says how many of the files are loaded and Load more brings the rest; search and facets work on the whole set, not just the visible page. More: Files and artifacts.

#Can I move to the next file from the file preview?

Yes. When you open a file, the ← and → arrow keys in the detail window, or the arrow buttons in the header, take you to the previous and next file in the list, just as in collection records. A counter shows which file you are on. The arrows stay out of the way while you are typing or the editor is open. More: Files and artifacts.

#The document I just saved looks empty in the preview. Is it lost?

No. The content is on disk the moment you save; the preview shows it before text extraction has finished, and the character count arrives on its own when extraction completes. A truly empty document cannot be saved anyway; the Save button does not enable without a name and content. More: Files and artifacts.

#I uploaded the same file again and no new row appeared in the list. Is it lost?

It is not. When you upload a file whose name and extension match one in the same scope and the same folder, the upload becomes a new version of the existing file, and a message on screen says so. The previous state is not deleted: open the file, look at the History section and go back to it whenever you want. If you want a separate file, upload it into another folder or rename it. More in Files and artifacts.

#Can I put files into folders?

Yes, but it is entirely optional. A folder is only there for tidiness, so moving a file does not change its name and does not break @file links. Agents see folders too and can write a file into a folder or move it; they still reach a file by its name. More in Files and artifacts.

#Can I put collections into folders too?

Yes. Create a folder with New folder on the Collections page, then drag a collection into it or use the Move to folder button on its row. The collection's name and its @collection link do not change. Agents can also create a collection in a folder or move one. More in Collections.

#Can I write a new document directly in Orbi without preparing it on my device?

Yes. The New document button on the Files page opens a blank document, and you give it a name, write the content and save. The document is added to Files, can be mentioned with @file and agents can read it. It lands in the scope and folder you are looking at, just like an upload. Markdown is the default, and you can also pick plain text, JSON or CSV. More in Files and artifacts.

#Can a file on Google Drive be read without uploading it?

Yes. If you paste a Drive, Google Docs, Dropbox or OneDrive link into a task, the agent downloads the file, saves it in your workspace and reads its text; you do not have to think about turning a view link into a download link. The file must be shared with anyone who has the link, because for a file open to specific people only the provider sends a sign-in page instead of the file, and the agent tells you so. More in Files and artifacts.

#Can I have a scanned (photocopied) PDF read?

Yes. In PDFs with no text layer, the pages are turned into images and read with a model that supports images. This OCR step handles at most 30 pages and uses models from the providers whose keys are in your workspace. More in Files and artifacts.

#Can I turn a collection into a screen the team uses?

Yes. A collection can be added to the menu as a permanent app screen (Admin only). It works like a small CRM with a list, filters, export and bulk operations, and agents keep writing to it. More in Collections.

#Can I show a file or a table to someone who has no Orbi account?

Yes. The Share button in the title bar of the file or the collection creates a link, and anyone holding it opens the content read-only in a browser, with no sign-in. In a collection all records and all views are visible, while for an artifact it is always the latest version. You can set an expiry date and revoke it whenever you want. An Admin sees every link under Settings › Sharing and can turn external sharing off with a single switch. More in External sharing.

#How do I reactivate a suspended share link?

Press Reactivate next to the link in the share dialog; the same address works again. The space's manager, an Admin or the link's creator can do this; a link in a closed space is reactivated only by that space's manager. The same spot offers Suspend to pause a link without deleting it. More in External sharing.

#What happens if someone looking at a page I shared wants to join the team?

The Join this team button at the top of the page sends you a request, and to use it they have to sign in with an Orbi account, creating one if they do not have it yet. Admins get a notification and an email. Under Settings → Team → Join requests you pick a role and approve or decline, with Viewer preselected. If you do not want the button, you can turn it off under Settings → Sharing. More in Roles, invites and sessions.

#Can I get PDF output out of Orbi?

There are two ways. An agent can write a PDF file directly; that route produces a plain document (headings, paragraphs, lists, tables) and carries no colour, images or custom layout. If you need a presentation, a proposal or a report that follows your brand, have the agent write a page file (.html) instead: the print button in the preview opens the browser's window, where you pick Save as PDF. On that route the whole formatting is preserved. More in Files and artifacts.

#Why do external images and links in my page file not work?

On purpose. The page preview runs in a closed box: code inside it is not executed, no connection is made to the outside, and links pointing at external addresses are not clickable. This stops a page sent to you from tracking you or running something in your browser. Images inside the page come from files in Orbi; writing the image's name is enough. More in Files and artifacts.

#Can I write down what a channel is for?

Yes. The pencil button next to the channel title lets you write the channel's purpose, and it appears in the header for everyone as well as on hover over the channel name in the menu. Writing and changing it is an admin's job, because the purpose is the channel's definition. The agents you call in the channel read it too, so they know the frame they are answering in, but the purpose is context rather than an instruction given to them. Details: Channels.

#Can I send a file to a channel?

Yes. The paperclip button under the message box lets you pick a file, and the file appears by name inside the message you send. To send a file that already lives in your workspace, type @file. while writing the message and pick it from the list. The channel's Files tab collects both in the same list, because for you both are simply a file that came up in the channel. The list is split by time and every row says who brought the file in. Details: Channels.

#Can I keep my brand colours in a document?

Yes. A colour code in your text gets a small square in that colour in front of it in the preview, so you do not have to open the colours somewhere else while reading the list. Six- and eight-digit codes are recognised in plain text too; the three-digit short form is recognised only between backticks, so it does not collide with card numbers. This makes it easier to keep a design guide or a brand palette as a document. More in Files and artifacts.

#How do I make a long research output readable?

Save it as a mind model. The subject becomes a tree: headings turn into branches, findings into nodes, and what is still open shows up in colour. In a decision record the reason an option was ruled out stays written down, because a rejected branch is never deleted, only struck through. Agents read the same tree, and when they work on a branch they see the whole chain from the root down to it. More: Mind models.

#Does a decision tree I draw in a mind model run by itself?

No, and that is deliberate. A mind model is a thinking surface: it holds the decision, its reasoning and the paths you ruled out, but it executes nothing. If you want the steps of a job carried out automatically, flows are the right place. For the same reason the tree does not calculate; you can write numbers into it, but it will not add them up. More: Mind models.

#Can I see a mind model with the root in the middle instead of left to right?

Yes. Press the radial button in the controls at the top right of the canvas, next to the zoom buttons; the root moves to the centre and the branches spread around it. The button stays in place in full screen too. A map with a few main subjects of equal weight reads more easily this way. The choice is written into the document, so once you save, everyone who opens the tree sees the same shape. On the horizontal tree you can pin a card by dropping it on empty space; on the radial layout the tree always arranges itself. More: Mind models.

#How do I point at a mind model node in a chat or on a card?

Click the node, choose Copy address in the detail panel and paste it. You get an address such as @file.pilot-takip#3f9c2a71; whoever clicks it opens the mind model with that node selected and centred. Do not use the number on the card (1.2.2): it shifts when the tree changes, while the address keeps pointing at the same node. More: Mind models.

#Why does a change Claude Code made to a mind model show under my name?

You connected Claude Code with your own account, so it works on your behalf and everything it writes is saved under your name. To keep the agent's work visible, the version history shows "via Claude Code" next to your name; those versions stay on a separate line from the ones you saved by hand and land on an open canvas at once, like an agent's version. If you want a separate signature, give the tool a token in an agent's name. More: Mind models.

#Can I have an agent build a mind model from chat?

Yes. In chat, "turn this research into a mind model" or "build a decision tree for @file.report" is enough; the agent builds the tree, the file appears in Files with the .mind extension and opens on the canvas. The agent has a dedicated tool for this, so the result is a validated tree, not a plain text or JSON file. You can also ask it from chat to change a single node in an existing tree. More: Mind models.

#How do I enter the "yes/no" text on a wire and the badges under a card in a mind model?

From the node detail panel that opens on the right when you select a card. With editing on it has a box each for the note, related items, source and the label of the incoming wire; typing @ in the related box suggests people, agents, tasks and files, and you can add more than one. The wire label belongs to the wire coming from the parent to this card; under the box it says which card it comes from. What you write is applied to the tree when you leave the panel, and saving is still the Save button. The bar on the card holds only the title, type and state; content is edited in the panel. More: Mind models.

#Flows and integrations

#The agent does not use a tool of the service I connected and says it has no permission. Why?

The agent sees every tool of the connection it may use and loads the one the job needs by name. When it cannot load one it writes why: the tool was not granted to that agent, it is turned off on the connection, or the server has no such tool. You turn the first two on from the connection's tools on the agent page and from the connection's Tools to enable setting. Mention the tool in your message, for example @tool.design.post_create, and it is open from the start of the run. More in Mentioning a connection and its tools.

#My scheduled job did not run. Why?

There are three common reasons. A trigger that arrives while the previous run is still going is skipped, the flow may be paused, and cron is evaluated in the flow's own time zone. Try a new flow with Run now instead of waiting for it. More in Troubleshooting and Flows.

#Is an event lost if it arrives while a run is going?

It depends on where it came from. An event in a connected service (new mail, a Slack message, an event about to start) and a call from outside wait in line, are handled in the order they arrived as soon as the run ends, and at most 50 events can wait on one trigger. A scheduled run or an event trigger is skipped instead, because both come round again on their own. More in Flows.

#Can I build a flow from chat, without touching the canvas?

Yes, in two ways. The Flow Builder panel on the canvas asks its questions and draws the cards, the wires and the trigger in front of you, while saving is up to you. In general chat you can also describe the trigger, the steps and the order, and the agent asks about the missing pieces (how often, which agent, what condition) in a single round, then builds the flow. Either way, if you described more than one role, each role becomes its own card, and when you say "add a step to this flow" the existing cards are kept. More in Flows.

#Can I give the Builder a screenshot or a document?

Yes. Both the Agent Builder and the Flow Builder have an attach button in the message box, and you can also point at a file you uploaded earlier by typing "@" and its name. The Builder sees the image you attached and reads the text of the document, so you do not have to describe the process diagram or the brief you wrote somewhere else. Sending only the file, with no text at all, works too. When it cannot read a file, for example a scanned document with no text layer, it says so instead of guessing the content. More in Agents.

#The cards on my flow canvas are piled on top of each other. Can I fix that?

Yes. Tidy layout, in the bottom right corner of the canvas next to the zoom buttons, lays the diagram out again: every card moves under the card that feeds it, cards fed by the same card line up side by side, and none of them ends up hidden under another. The arrangement is not permanent right away, so if you don't like it, click the notice in the corner to go back. The canvas also does this on its own whenever the Builder adds a card or a wire, so you rarely have to tidy a tangled diagram by hand. More in Flows.

#How do I stop a flow step from writing to a collection?

Click that step's card on the canvas and tick "This step cannot write to the board" in the right-hand panel. The step can no longer write records, tasks, files or calendar entries; it passes what it finds to the next step through its output only. Reading and web search are not affected. Turning this on for research and classification steps gathers all writing into a single step at the end of the flow and stops the same information from being saved twice. More in Flows.

#Can I duplicate a flow?

Yes. Duplicate, in the ⋯ menu at the top bar, opens a new flow with the same cards, wires and triggers. The copy is born paused, so you review it and turn it on yourself. If the flow has an incoming webhook address, the copy gets its own new address — the original keeps working as it did, the two never mix. You will find the same duplicate button on an agent's page and in a collection's settings, too. More in Flows.

#Does a flow remember its previous run?

No, runs do not remember each other. Access to the past is set up through a collection, so a run writes its output to a collection and the next run reads that collection, which is also how semantic duplicate prevention comes into play. More in Flows.

#Can I try a flow without writing to real data?

Yes. Test really runs the flow, so the model works and the reading tools work on real data, but the tools that write outward are not executed. A "this is what I would have done" preview takes their place. More in Flows.

#Can an event in an external system (a form, a payment, a CRM record) start work in Orbi?

Yes, that is exactly what an inbound trigger is for. The external system POSTs to the generated address and the flow runs. What it sends counts as data, and what is done with it is decided only by the job description you wrote. More in Collaboration.

#Can my own system be told when a job is finished?

Yes. Outgoing connections deliver twenty-six events covering tasks, collection records, files, runs, membership and credit, signed with HMAC, and you choose which ones you want. An agent can also leave the result in a channel the team watches. More in Outgoing connections and Channels.

#Can I limit the events sent to a webhook to a single project?

Yes. While setting the connection up you can narrow it by projects, by collections and by who made the change (people only or agents only). If no box is ticked, everything is sent. The narrowing only applies to the events it belongs to, so picking a project does not silence your credit warnings. More in Outgoing connections.

#My webhook stopped working and I got a notification. What happened?

When five deliveries fail in a row the connection is paused automatically and the admins are notified. The events and narrowing you chose are not deleted. You can see the receiver's response in the delivery log, fix the address, and then Turn back on picks up where it left off. More in Outgoing connections.

#My channels all look alike in the menu. Can I put an icon in front of their names?

Yes. On the channel page, click the # to the left of the title and pick an emoji, and it takes the place of the mark. The icon shows in the menu, in the channel title and on the browser tab, and it works the same way as on collection screens. Only admins can change it. More in Channels.

#There is too much on one day in the calendar and I cannot see it all. What should I do?

Click the +3 more row at the bottom of the box and the whole day opens in a small panel. From the same panel you can also open the day in the hour grid. How many items are shown directly depends on the height of the window. More in Flows and calendar.

#How do I move an event to another day?

Drag and drop it. If you drop it on another day, its time and duration are kept, and if you drop it on a different hour of the grid in day or week view, the time changes too. Task due dates and flow runs cannot be dragged, because those change from the settings of the task and of the flow. More in Flows and calendar.

#Where do I see who added an event to the calendar?

Click the event; the card that opens says who added it, below the description. If it was a teammate you see their name, and if it was an agent you see its name and its title. Task due dates and flow runs have no such line, because they are not added to the calendar by hand. More in Flows and calendar.

#An event runs over several days. Does the calendar show it across all of them?

Yes. Next to the start of the event there is an Ends box, and you can put the end on the next day or the next week. In month view the event is drawn as a single band across all the days it covers, while in day and week view it sits in the All day strip above the hour grid. For events that stay inside a single day, the height of the box matches the duration, and you change the duration by dragging the bottom edge of the box. More in Flows and calendar.

#If I delete an agent that a flow uses, does the flow break?

No. The agent's card on that canvas and the wires attached to it are removed, and the rest of the flow keeps running. Before you delete, the confirmation names the flows the agent is used in. More in Deleting things.

#If I delete an agent, are the tasks assigned to it deleted too?

No, the tasks stay on the board and show as unassigned. You just pick a new owner. The only thing deleted is your one-to-one conversations with the agent. More in Deleting things.

#Can I see where something is used before I delete it?

Yes. When you press delete, the confirmation scans the usage and lists what will change line by line; the confirm button stays disabled until the scan finishes. If it is used nowhere, it says so. More in Deleting things.

#I deleted an agent and a flow trigger paused itself. Why?

The trigger was narrowed to run only on tasks assigned to that agent. When the agent went, the narrowing was left empty — and in Orbi an empty narrowing means "run every time". The trigger is paused so it does not quietly start reacting to every task; decide what you want and turn it back on. More in Deleting things.

#I have a question or something is broken. How do I reach the Orbi team?

Write through the contact form on the site, or email platform@orbiwork.space directly; both land in the same inbox and the reply goes to the address you write from. The form does not require you to be signed in, so it also works when you cannot get into your account. If you would rather talk than write, ask for a demo or a meeting at orbiwork.space/booking. More in Troubleshooting.