AI Agents
How agents work
Behind every question, report, or automation in Statement, an agent (or a small team of agents) is doing the actual work - retrieving data, interpreting it, and producing the output you asked for. Rather than a single monolithic system trying to do everything, Statement breaks work down by kind of task, and routes each part to the type of agent suited to it. Statement uses specialized agents to handle different parts of a request, such as retrieving data, performing calculations, and preparing a report or application. You describe what you need, and Statement coordinates the work automatically. You can see progress updates in the chat while this happens. An expandable view shows steps and messages such as “Retrieving data,” “Applying calculations,” and “Preparing the summary,” helping you follow what’s happening. For example, a revenue comparison may involve retrieving figures for two periods, calculating the change, and presenting the results. These appear as steps within one conversation. You don’t need to choose or manage individual agents. For everyday use, the useful distinction is the work each one handles and the progress you can see as Statement prepares your answer.
Research agents
Research agentsResearch agents find the information needed to answer your question using your connected services and selected uploaded files. They identify the relevant source and retrieve records for the accounts and time period you’re asking about. For example, ask “Which invoices are still unpaid?” Statement retrieves the relevant invoice records from your connected accounting service, including details such as amounts, customers, and due dates, where available. When a question involves several sources, Statement gathers the relevant information from each. That evidence can then support calculations, comparisons, and a clear explanation, helping you explore your business data without manually searching through individual records.
Data agents
Data agentsData agents work with the numbers in your connected financial services and selected uploaded files. They retrieve the relevant records, apply the requested dates and filters, and prepare the data needed to answer your question. For example, ask “How much did we collect from customers last month?” Statement retrieves the relevant payment data from your chosen connected service and calculates the total. If you ask for a comparison, it can also retrieve the previous month’s figures and calculate the change. Retrieval and calculation are handled as distinct steps. Statement uses the retrieved values to produce totals, breakdowns, and comparisons before turning those results into a written answer, table, or chart. The results depend on the data available to Statement, including how recently the connection was synced. When required information is missing or the source is unclear, Statement can flag the gap or ask you to clarify.
Workflow agents
Workflow agents bring multiple steps together into a repeatable process. They can retrieve data, apply calculations, prepare a summary, and carry out a supported action through a connected app. For example, you could ask: “Every Monday morning, pull last week’s revenue and email a summary to our finance team.” With the required connections and recipients specified, Statement can prepare a workflow that retrieves updated figures and creates the message each time it runs. Workflows can run immediately, once at a future time, or on a recurring schedule. They can also include a condition-for example, checking overdue invoice totals each morning and sending a notification only when the total exceeds an amount you specify. You can ask Statement to list, change, pause, resume, or delete schedules. This lets you manage recurring work through conversation as your needs change.
Reasoning engine
The reasoning engine coordinates Statement’s work. It interprets your request, considers the conversation so far, and identifies the information, calculations, and actions needed to deliver the result. It brings together the capabilities described as research, data, and workflow agents. It decides what can happen next, passes results between steps, and reviews what each step returns before choosing how to continue. Independent tasks can run together; tasks that need earlier results wait for those results. For example, ask “Every Monday, email me a comparison of last week’s revenue with the week before.” The engine coordinates retrieving both periods, calculating the change, preparing the summary, and setting up the recurring delivery. If a required detail is missing, such as the destination or timezone, it asks you to clarify. The reasoning engine handles coordination, while specialized capabilities retrieve data, perform calculations, and create the final output. This connects your original question to a complete answer, application, or workflow.