The Accounting Intelligence Agents panel is the persistent surface for conversational work in DualEntry. It holds your Accounting Intelligence conversations, the connectors and skills those conversations can use, and the scheduled agents that run without you.
Open it from the Accounting Intelligence button in the top bar or with Option+Cmd+] (Mac) / Ctrl+Alt+] (Windows) by default. To rebind the shortcut, see Keyboard Shortcuts. The panel opens along the right side of the window and can be resized, so the record you are working on stays visible beside it. Close it with the X in the panel header or press the same shortcut again.
The panel is the only Accounting Intelligence surface that carries navigation. That is why you create and manage connectors, skills, and scheduled agents here rather than from the Accounting Intelligence embedded in a record. Choosing Ask Accounting Intelligence in global search opens this same view full-page, with the same navigation. For how the three surfaces relate, see How Accounting Intelligence Works.
Navigation sections
The Agents panel’s navigation holds six sections, each a separate surface: your conversations, the artifacts they produce, and the connectors, skills, and schedules they draw on. The table below lists every section and what it contains.
Scheduled and Recents lists
Below the navigation, the Agents panel lists your Scheduled agents and Recents. Each list has a See all link and a collapse control.
Each scheduled entry carries the agent’s name, a badge counting runs you have not opened yet, and a second line that reflects the agent’s latest run. The table below lists what the second line shows and where a click takes you.
A failed run never shows Ready to view. For the run statuses and where each run’s chat lives, see Scheduled Agents Reference.
Recents are conversations you started, from this and prior sessions, titled from their first exchange. Runs from scheduled agents stay out of Recents until you reply in one; from then on that run is listed like any other conversation.
Closing the panel does not end a conversation. It stays in Recents, so reopening the panel and choosing it resumes where you left off. Each browser tab holds its own conversation context, so a conversation open in one tab does not interfere with a conversation open in another, and the panel in each tab tracks only what happened in that tab.
Composer controls
The composer is where you type a request in Accounting Intelligence. Three controls sit below the input: the + menu, the accounting-basis selector, and the entity selector. The two selectors change the numbers you get back rather than just the presentation.
The + control is not only a file picker. It opens a menu that reaches most of the panel’s navigation from inside the composer, so you rarely have to leave the conversation to pick up a skill, a connector, or a schedule. The menu holds six rows:
Five of those six rows lead to a section of the panel navigation described above. The composer menu and the navigation are two routes to the same surfaces rather than separate features, so a skill or connector you reach one way is the same one you reach the other way.
The other two controls are selectors. Both apply to the whole conversation rather than to the message you are typing, and neither can be pinned by a skill or set as an organization default:
For what goes wrong when these are set incorrectly, see How Accounting Intelligence Works.
Submit with the arrow control or the keyboard. Below the composer, Try Skill press / indicates skill invocation from the same input.
Suggested prompts
New conversations in Accounting Intelligence open with suggested prompts as starting points, including a review of what needs attention today, close status, a summary of your recent activity, and a P&L refresh. See all suggestions expands the full list. The two selectors apply to a suggested prompt exactly as they do to one you type, because they scope the conversation rather than an individual question.
Suggestions are generated for you rather than drawn from a fixed list. DualEntry reads your own recent activity in the audit trail, roughly the last five days, and proposes the steps an accountant would naturally take next. Two people sitting down at the same moment see different suggestions, and someone with no recent activity sees a general starting set instead.
A generated set is held for about four hours before it is rebuilt, so suggestions do not change between one conversation and the next within a working morning. Activity from the last few minutes may therefore not be reflected yet, which is worth knowing if you expect a suggestion to appear immediately after posting a batch.
When generation is unavailable, the panel falls back to a fixed starting set rather than showing nothing. A generic-looking list is that fallback rather than a judgment that you have done nothing recently.
Permissions and logging
Every conversation in Accounting Intelligence inherits the permissions of the person running it. Role-based permissions determine which entities, modules, and records a user can open, and Accounting Intelligence operates strictly inside that grant. See user roles and permissions for the full model.
Accounting Intelligence cannot return a record you could not open yourself, and it reports that it cannot answer rather than estimating when data is out of scope or absent. Two people can therefore ask the panel the same question and get different answers, which reflects their roles rather than a fault in the assistant.
Every prompt and response in the panel is written to the audit trail, including prompts a scheduled agent runs while nobody is watching. For the fields each entry records and where to review them, see How Accounting Intelligence Works.
Connector tool calls are the exception to the permission model. A connector acts with the permissions of the credential used to authenticate it in the external system, which has no relation to the permissions granted to you in DualEntry. See how to connect an MCP connector.
Last modified on October 13, 2026
