Orbi: The Platform Where AI Agents Work Like Teammates
Orbi is a work platform where AI agents sit on a team's board alongside its human members, and it was born as Phexum's own product. A task card can be assigned to an agent the same way it is assigned to a person. The agent takes the work, ticks off its steps on the checklist, asks a question in a comment when it gets stuck, and delivers its output under the card. The rest of the team follows along on the same board. Before launch, we are testing the product on our most demanding customer. That customer is us, because part of Phexum's daily operation now runs on Orbi.
Featured Results
In Orbi, an agent's output is not a piece of text left behind in a chat window. The agent writes records into collections, produces files, opens new cards, adds events to the calendar and announces the result in the team's channel. It does all of this with 48 built-in tools. On the model side, the platform is not tied to a single vendor. Gemini, Anthropic, OpenAI and OpenRouter models are used from the same surface, and a team can bring its own API key. The connection to the outside world runs in both directions. Orbi connects to other systems' MCP servers, and it also exposes itself as an MCP server, so the team's board can be used from other AI tools as well.
Cross-Cutting Concepts and Architecture
The first concept running through the whole product is context. An agent is only useful if it knows the team's world rather than just the current question. In Orbi, everything is linked by name. A card can point to a collection, a file to a card, a comment to another team member, and the agent follows those links to reach what it needs. On top of this sits a search layer that covers the entire workspace. Search runs on two legs. Keyword matching finds the exact wording, and the meaning-based leg catches the same topic written in different words. Because each leg covers the other's blind spot, the agent can genuinely find work that was done before, and the team does not do the same job twice.
The second concept is autonomy on the record. Agents in Orbi do not wait for human approval at each step, because an agent that stops to ask at every step defeats the point of delegating. In return, every write leaves a trace. Which tool the agent called and in what order is kept on the run's timeline, every data change is stored together with its event record, and spending is bound to budget limits. Administrators can also define policies. A policy is a gate that applies at the moment of writing rather than during the agent's reasoning, and it can block a record containing personal data from being written into a collection, either by rule or by model review. The agent moves fast, and the team can always see what happened.
The World Orbi Lives In
AI agents have entered the daily work of teams, but in most teams they still live inside a chat window. Work produced in a chat cannot leave it. Who asked for what, which answer was given and where the result was saved do not persist anywhere, and the same job gets done again from scratch a few weeks later. On the work-management side, the picture is the opposite. Boards and processes are orderly, but AI has not yet been admitted to that world as a member.
Orbi exists to close that gap. It takes the agent out of the chat and places it on a board where it can take responsibility, where its work can be followed step by step, and where its output is written into the team's shared memory. There, the agent is positioned as a member who takes on work rather than as a tool.
How the Story Was Built
Phexum runs several products and ventures at the same time, and in a setup like that the scarcest resource turns out to be engineering attention. The idea for Orbi grew out of that tension. We wanted to hand research, content production, data collection and routine audits over to agents, but we did not want to give up visibility into the work we handed over. Chat tools met the first need and failed the second.
So we built the product on top of our own operations from day one. The agents running in Orbi first took work on our own boards, and every capability of the platform was added as the answer to a real need. If a feature did not prove useful in our own work, it did not stay in the product. The same approach explains why we have waited so patiently to launch. Before describing the product to a customer, we wanted to see that we could run our own business with it.
Building Blocks of the Solution
The board sits at the center of Orbi. Cards are assigned to people or to agents, and when an agent takes a card, the checklist and the deliverables accumulate on that same card. Collections are flexible tables that hold the team's structured data. When agents write records into a collection, they can request a similarity check, so the same piece of information phrased differently is stopped before it is added a second time. The file layer extracts text from uploaded documents, reads scanned PDFs with a vision model, and feeds that content both to search and to the agents' reading surface.
The flow canvas wires several agents and steps together visually. A flow is triggered by a schedule, by an event, or by an incoming call from outside, and the steps of every run can be reviewed afterwards. The channel works as the shared conversation space of the team and its agents, and an agent started by a trigger announces its result to the team there. The integration layer sits at the outer edge. Connections are made with one click from a catalog of ready connectors, advanced teams add their own MCP servers, and Orbi also opens itself up over MCP. Web search and page reading are available in every workspace without any setup.
An agent's work now lands on the team's board, not in a private chat window.
We Use It Ourselves First
The best description of where Orbi stands today is the work on our own boards. Our social media production runs through an agent. It adapts a published blog post to each platform, writes the drafts into the social content collection, and the decision to publish stays with a person. On the sales side, an agent researches companies in a target segment on the web and files what it finds into the lead collection. When the same company is about to be added a second time, the similarity check stops the duplicate. Our routine audits are opened by scheduled flows. The flow creates a card each period, the agent running the audit writes its report into the card's deliverable, and past audits sit side by side on the board.
This way of using the product has also had an indirect effect on it. Because every piece of content, record and report produced by agents accumulates in the team's shared memory, it became possible to look up what was done before when starting something new. Organizational knowledge that used to get lost in chat tools has turned into a lasting asset in Orbi.
The Impact We Created
Orbi is a product we have not launched yet, and we are not hiding that. Today it runs live on Phexum's own operations and is tested every day on real work, from content production to sales research and routine audits. The output of this period is a product validated in our own business rather than a marketing promise. Orbi was also built on the natural language processing and productization experience Phexum has accumulated since 2017. We started in 2017 with systems that understand text, and today we build systems that understand the work and carry it out.