2.5 The agent system: ask or automate

Above the data layers sits the part of Orbit that does the work. The Knowledge Base holds the connected data; the agent system is what turns it into answers and outputs. There are two ways to put it to work — ask it a question, or set it to run a process — and both draw on the same foundation underneath.

When a request comes in, an orchestrating agent interprets it, works out which data and which specialist agents it needs, draws on the structured and document layers as appropriate, and assembles the result. The user does not have to direct any of this — which agent runs, which layer to read, how to combine the pieces. They state what they want; the orchestration handles how it is produced. This is the layer your team experiences as simply "asking Orbit," and it is the same machinery whether the request is a one-off question or a process that runs every night.

Two modes

Ask is the interactive mode. A user puts a question in natural language and gets an answer drawn from across all of the data at once, grounded in primary sources and traceable back to them. This is the mode for the question that arises in the moment — during earnings season, ahead of a meeting, in the middle of building a view. The work happens now, and the answer comes back now.

Automate is the standing mode. The same agents can be assembled into repeatable workflows that run on their own — on a schedule, or in response to a trigger such as a new filing or a piece of news — and produce structured data and finished reports without anyone restarting the process each time. This is the mode that turns a piece of research a team does by hand, again and again, into something the platform does for them in the background. A morning briefing on a coverage list, a standing watch on a portfolio, a report refreshed every quarter end: each is a workflow defined once and run indefinitely.

The two modes are not separate products; they are the same agent system, used two ways. A workflow is, in effect, a question worth asking on a schedule. Teams typically begin in the ask mode — exploring, getting comfortable, seeing what the data can answer — and move work into the automate mode as patterns emerge and certain questions prove worth running on their own. That progression, from asking to automating, is how a firm shifts from using Orbit occasionally to running part of its research process on it.

Off-the-shelf, or built by you

Orbit provides a set of off-the-shelf agents for common research work, so a team can get value immediately without building anything. These are covered in 2.6. But the more important point is that agents are not a fixed catalogue handed down by Orbit. Every user can build their own — encoding their own logic, their own view of what matters, and their own way of working — and can link agents together so that the output of one becomes the input to the next, automating a chain of research from end to end. This is covered in 2.7.

This is what makes the platform personal as well as shared. The data underneath is common to the whole firm; the agents on top are each user's own. Two analysts working from the identical Knowledge Base can run entirely different processes over it, each reflecting how they research, what they cover, and what they care about. The platform does not impose a single way of working — it gives each person the means to build theirs and run it at a scale no individual could manage by hand.

Throughout, the agents work the way the rest of the platform does. They draw on the structured layer where it answers, and on the document layer where a question needs the words themselves, exactly as described in 2.4 — and every result they produce carries its sources, so it can be checked. The agent system adds reach and repeatability on top of the connected data; it does not change what the data is or how it is grounded.