You can ask the same question in Orbi from three separate places. You can open the chat panel and type it, name the agent in a channel, or open a card and assign it to the agent. All three work, and they do not give the same result.
Knowing the difference saves time in the first week. Work given in the wrong place does not get lost, but it sits somewhere the team does not see.
Chat is fast and personal
You open the chat from anywhere in the app with a shortcut and the answer streams in word by word. You can attach a file too, and the document you attach enters the context in a form the agent can read.
There are two kinds of chat. A chat opened from inside a card knows that card's data, so you can ask "where did we get to on this". In the general chat, independent of the board, you pick the scope yourself. It can be visible only to you, or to everyone in the project.
Chat is the right place for trying something out, quickly reworking a text, querying a record, or thinking before you decide how to define a job.
A channel is the team's shared stream
Between chat and card there is a third place. A channel is a shared message stream that everyone in the workspace sees. Announcements, daily summaries and automated reports land there.
There are no private channels, no invites and no direct messages. If a channel exists it shows in everyone's menu and everyone can write in it. That is a deliberate simplification, because Orbi is not replacing a messaging app.
Agents do not read the channel continuously. They only answer when named, so two people talking to each other do not get an agent cutting in. The answer is prepared in the background. The agent does its research and the message lands in the channel once ready. You can expand the step block under the answer to see which steps it took.
A channel is quiet by default. A notification only arrives when you are addressed, meaning when someone names you or when an agent you called writes its answer. If you want every message as a notification, you switch that on in your profile preferences.
A channel can also have a purpose, a sentence or two answering "what is this channel for". Agents you call in the channel read the purpose too, so they know what frame to answer in. The same question deserves a different answer in a hiring channel than in an incident channel.
A card is shared and traceable
When you open a card, the work moves to where the team can see it. The card gets an owner, a state and a history.
When an agent takes the card it writes the steps into the checklist, puts the delivery in the output area, drops its reasoning into a comment and moves the card to review when done. All of that can be read later.
A card is the right place for work someone else might take over, work whose result needs keeping, work that needs several steps, and work where "who did this" will be asked.
The fourth way, calling from under a card
If you name an agent while writing a comment under an existing card, that agent does a short run limited to the card's context and writes its answer as a comment.
It is the way to ask for a quick fix without opening a full task. You do not need a new card to say "rewrite the third section shorter and more concrete".
There is a safeguard. Comments written by agents do not trigger a new run, and each comment is processed at most once. Without that, two agents could name each other into an endless conversation, with a bill growing on every message. The rule is kept simple. A trigger only comes from a human comment.
How much of what is said in chat survives
The chat itself does not carry over to the next job. What you explained in a chat last week does not appear in front of the agent on the card you open today. Chats do not enter workspace search either, because they count as personal correspondence and are not indexed.
A lasting instruction you give in chat, on the other hand, is not lost. When you say "from now on write reports as bullet points", it becomes a memory note and is recalled to the agent at the start of later runs.
The conversation itself is not permanent. The rule that comes out of it is.
There is a difference in authority too
Services you have connected personally run only in runs you start. In chat, you are the one typing, so the identity is clear and those connections load.
The same holds when you queue a card yourself. But if an agent opened the card and queued it, there is no identity, and the run continues with the workspace's tools alone.
Worth knowing if you are having work done that depends on your own account.
The rule of thumb
If only you will use the result, chat. If the team seeing it is enough, a channel. If someone will take it over or it needs finding later, a card.
Another measure helps. Will you need to find the output again in two weeks? If yes, the job belongs on a card, because both chat and channel will have flowed past by then.
Where to start
Start in chat. If you do not quite know what the job is, ask the agent, go a few rounds, sharpen the edges of the job.
The moment it is clear, open a card. Write the definition that came out of the chat into the card's description and assign it to the agent. That keeps the thinking phase fast and the delivery phase traceable.
Open a channel for the team's daily stream. Dropping a scheduled job's output into a channel means the team knows where to look in the morning.
Chat types, calling an agent from a comment and the loop safeguard are covered in Collaboration and notifications in the help center. For channel rules see Channels, and for the areas inside a card, Tasks and Kanban.
If you want to try it, you can open a free account and start your first chat.