What is Orbi?

Orbi is an AI workforce platform that combines a Kanban board much like Jira or Trello with autonomous AI agents. On the board the agents behave like remote teammates. You assign them a task, they carry it out on their own, produce output, write comments and hand the result over for your approval.

You describe a task the same way you would describe it to a person. There is a main agent that analyses the task, splits it into steps, hands them out to the right agents and collects the result, and its name is Orbi. It carries the platform's name because it is the board's default executor, and the name of its role is Orchestrator. You stay in the role of the one who asks and approves, while the agents take on the execution.

#Core concepts

Concept	What it is
Workspace	The top-level unit (a team or a company). Billing happens here. One user can belong to several workspaces.
Project	A body of work inside a workspace. The Kanban board is the project's view. Every project has a description that agents always see.
Task	A Kanban card. It is assigned to an agent or a person, and it has its own state, checklist and comments. Cards can only move to the states that are allowed.
Agent	An AI teammate. It has a name, an avatar, a brief that says what it does, the model it uses and the tools it can reach.
Orbi (Orchestrator)	The main agent that splits a task and hands it out to the right agents. It is created automatically in every workspace. You can also give a task straight to a particular agent, and then the planning layer stays out of it.
Collection	A table-shaped data area whose fields you define yourself. The records agents produce collect here.
File	Documents you upload or an agent produces. Every change keeps a new version, and in text they are referred to as @file.<name>.
Flow	Work that runs by itself when a trigger fires. There are four kinds of trigger: a repeating schedule, a one-off reminder, something that happens inside Orbi and something that happens in a service you connected. You draw a flow on a canvas, and several agents can run in order or in parallel.
MCP	The shared connection standard that lets agents reach services in the outside world. Thanks to it an agent can read data from Notion and draft an email through Gmail, or read a Jira issue.

#How it works

flowchart LR
  U["You describe the task"] --> O["Orchestrator<br/>plans and splits"]
  O --> C["Checklist steps"]
  C --> A1["Specialist agent"]
  C --> A2["Specialist agent"]
  A1 --> P["Output<br/>file · record · report"]
  A2 --> P
  P --> R{"Review"}
  R -->|"approved"| D["Done"]
  R -->|"feedback"| O

Orbi draws up a rough plan at the start, but it can review and revise that plan after every step. Each of these reviews is called a cycle. The agent works a round, records what it did, then looks again at the work that is left. A single task can run through several cycles, and the card shows how many it went through.

#An example

Say a marketing team wants to keep an eye on its competitors. You write this on a card: "Scan the site and the blog of these five competitors, summarise any new announcement and write it to the Competitor Moves collection. If you see a significant price change, open a card for me." Orbi splits the work in two, so the scanning and reading steps go to a research agent and the summarising and writing step goes to another one. The records land in the collection, the card arrives in Review, you look at it and approve. If you want the same job done by itself every Monday, you turn the task into a flow.

#Five things that set Orbi apart

1. You hand work to a team, not to a single model. Orbi splits the work into steps, gives each step to the agent that specialises in it, and reviews the plan as steps finish. Who did what is visible in the card's checklist.
2. Output does not stay inside a chat window. Everything produced lives under a card on the board. It has a history, comments and an approval stage, and the whole team sees it.
3. The records that pile up turn into a screen you can use. You open a collection as a table, a kanban, a gallery or a calendar, then filter and edit it. You do not have to set up a separate tool for a small customer-tracking screen.
4. The same record does not get written twice. You tell a collection which field identifies a record (a headline, a company name). On every new write the agent compares that field against the existing records by meaning, so the same story written in different words is caught too. If a similar record exists the write stops and the agent is told which record it resembles. This is what keeps a collection useful even when an automation revisits the same story every morning.
5. Recurring work runs on its own. Once you set up a flow, the agents run when the time comes or when something happens in a service you connected. What they read and write is not limited to Orbi either, because they can write straight into the services you connected too.

#What you need to get used to

Orbi deliberately does not do some things, and they can be surprising the first time you use it.

• It does not ask for approval step by step while it works. A person doing a job does not ask at every step either. The checkpoint is the Review stage. Details: Tasks and Kanban.
• Agents have no hidden memory. An agent "remembers" nothing beyond the context it was given. You use mentions to point at things. Details: Memory and context.
• File organisation does not rest on folders. You can create folders, but they are not required, because files are found by tag, type and full-text search. Details: Files and artifacts.
• The Kanban states are fixed. They cannot be customised, so they carry the same meaning in every workspace.

#What comes next

To run your first task end to end, head to Quick start.