Capabilities
Agents
MAIA is a coordinated system of agents. One main agent takes your objective and directs the run. Sub-agents do the deep work on each row, in parallel, so the conversation never waits on the research.
Two kinds of agent#
| Agent | Works on | Does |
|---|---|---|
| Main agent | Works onThe whole project | DoesReads your objective, chooses the data, builds the layers, and answers you |
| Sub-agents | Works onOne row each | DoesResearch one question about one place, hundreds at a time |
You talk to the main agent. Sub-agents are its research tool, and you see their work as cells filling in.
How a run goes#
- You define an objective and dispatch it.
- The main agent queries the data and builds a layer of candidates.
- Where the data holds no answer, it defines a column that asks for one.
- It tests the column on a few rows and corrects it if the answers are wrong.
- Sub-agents research the rest, one row each. Each row spends credits, and a run above your row limit stops on an approval card first.
- The main agent reads the finished values, checks the result, and answers.
Step 4 is why a badly worded question costs a few rows, not ten thousand. A column that sub-agents fill is an enrichment.
What each one leaves behind#
| Agent | Leaves | Where |
|---|---|---|
| Main agent | LeavesLayers, columns, views, and a written answer | WhereThe map, the table, and the chat |
| Sub-agents | LeavesA value, its reasoning, and its sources, per cell | WhereThe table, under View reasoning |
While a run is in progress, a blank cell is not yet researched, and Unavailable is researched with no supported answer. Checking a result is covered on Review.
Limits#
- One run per project at a time. Several projects can run at once.
In this section
- Main agentMAIA's main agent takes a research objective and returns layers on the map and in the table, with a written answer. It answers from the data first, researches only what the data cannot answer, and checks the result before it reports.
- Sub-agentsMAIA is a coordinated system of agents. The main agent runs the conversation, and when a question needs research on every row, it hands the tasks to a sub-agent. Hundreds run at once, in the background, and none of them holds up the main agent.