Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.circuit.ai/llms.txt

Use this file to discover all available pages before exploring further.

Some work finishes in a single chat. Other work, like responding to an RFP, qualifying a pipeline of opportunities, or processing a backlog of support tickets, stretches across days and many chats. Work items are how a Circuit agent tracks that longer-lived work so nothing falls through the cracks. A work item is a durable task the agent owns. Unlike an in-chat to-do list that disappears when the chat ends, work items persist across every chat you have with the same agent. When the agent has at least one work item, a Work items panel appears in the agent’s sidebar where you can browse, search, and filter them.

What a work item looks like

Each work item has:
  • Title and description: what needs to be done.
  • Type: task, feature, bug, or review.
  • Status: pending, ready, claimed, in_progress, completed, failed, or skipped. The panel groups these as Open (pending, ready, claimed, in_progress) and Closed (completed, failed, skipped).
  • Priority: a number where 0 is the highest priority.
Related items can be grouped together under a work item group, which is useful for projects like “Acme RFP” where one parent unit of work contains many child items. Groups have their own status (pending, active, completed, or failed) and can contain sub-groups for larger programs.

Asking the agent to create work items

You create work items by telling the agent what you need, in plain language.
Create a work item to review the Acme security questionnaire by Friday. Priority 1.
The first work item you create is what causes the Work items panel to appear.

Asking the agent to update and close work items

The agent can move items through their lifecycle as you go. Typical phrases:
  • “Mark the Acme questionnaire review in progress.”
  • “Close out the Globex line items I just approved.”
  • “The supplier crossover issue is fixed, mark it completed.”
  • “Skip the items that don’t apply to this RFP.”
For states that Circuit doesn’t model directly, like “needs review” or “submitted,” ask the agent to note them in the item’s description. The agent is trained to keep sub-states there rather than inventing new statuses.

Querying work items

You can ask the agent about the state of work at any time:
  • “What’s still open?”
  • “Show me every work item with priority 0 or 1.”
  • “Which items in the Globex group are still pending?”
  • “List work items that haven’t been updated in over a week.”
The agent reads the same data the panel shows, so natural-language queries return consistent results.

Using the panel

When the Work items panel is visible:
  • Search: type into the search box to match against title or description.
  • Filter: the segmented control switches between Open, Closed, and All.
  • Sort: items are ordered by priority first, then by creation date (newest within a priority).
  • Expand: click the chevron on a card to reveal timestamps and any extra attributes the agent attached.

Working with groups and projects

For domains like RFP responses, let the agent create one group per opportunity and one child item per question or line:
I’m starting a new response for the Contoso RFP. Create a work item group for it, then create a work item for each row in the attached spreadsheet.
Ask the agent to check for an existing group before creating a duplicate:
Before creating anything, check whether we already have a group for Contoso.

Tips

Be explicit about status changes. “Mark it in progress” or “close it” is clearer than “update it.” Status is one of the few things the agent won’t infer on its own.
  • Confirm bulk actions. Creating, closing, or commenting on many items at once is a bulk action. Expect the agent to summarize and ask before it closes a large batch.
  • Use descriptions for detail. The more the agent knows about what each item requires, the better it can prioritize, route, and resolve it.
  • Review the panel. Open the Work items panel periodically to catch items the agent marked failed or flagged for human review.