Active Agents

Active Agents is the live execution view — every agent currently running in your workspace, every one that's queued waiting for a resource, and a paginated list of everything that finished recently. It's the default tab on the Agents page.

📷 Screenshot: Active Agents tab with running, queued, and recent sections.

Three sections

Active Agents splits its rows into three collapsible sections so you don't have to filter mentally:

• Running — executions currently in flight. Status chip on each row reflects what the agent is doing: Running, Streaming, Generating, etc. Hidden when nothing is in flight.
• Queued & Paused — executions that have started but are blocked: waiting for a terminal-device reservation, paused on a question, or queued behind a rate limit. Hidden when empty.
• Recent — finished, failed, cancelled, or interrupted executions, paginated. Default page size is 25; you can switch to 100 or 250 from the page-size selector at the bottom of the table.

Each section header shows a count and toggles its body open / closed.

Columns

Both Running and Queued tables use the same columns:

• Status — colored chip (Running / Streaming / Queued / Paused / Cancelled / etc.).
• Agent — the agent's display name; sub-agents are prefixed with ↳ so the parent/child relationship is visible. Click the name to open the execution detail view.
• Feature — the feature this run is bound to (truncated with a tooltip when long). A dash means it's a chat-initiated or platform-hook run with no feature association.
• Current Step — the step name plus an indeterminate progress bar showing the agent is alive.
• Computer — the terminal device the run is bound to (if any), with a colored dot reflecting that device's online state. Hidden entirely when no devices are connected to the workspace.
• Elapsed — Hh Mm Ss since the execution started.
• Actions — view-detail link plus a cancel button.

The Recent table drops the progress bar (the run is over) and adds a final-status column.

Cancelling

The cancel button on a single row stops that run cleanly — agents support graceful interruption, write a checkpoint before exiting, and the row immediately moves into the Recent section with status Cancelled.

The Cancel All button at the top right of the page cancels every execution currently passing the active filters. It prompts for confirmation (Cancel all 7 running executions?) and is the right tool when an agent is misbehaving and spawning sub-agents you don't want to whack one at a time.

Filtering

A filter bar above the table narrows what's shown to:

• Status — multi-select.
• Agent path — pick one or more specific agents to focus on.
• Documents — narrow to runs that pinned specific documents as context.

Filters are URL-synced — paste a filtered link to a teammate and they see the same view. Cancel All respects the active filters, so you can use it for surgical cleanups (e.g. "cancel every running execution of analyzeRepo").

Device health

A small device status indicator in the page header summarizes the health of every terminal device the workspace knows about — green dot per online device, grey for offline, red for errored. The same colored dot appears next to each execution's Computer column, so when an execution is stuck because its target device went offline you spot it without having to dig.

If you need to free a device that a run is holding hostage, the per-device disconnect / restart actions live on Account → Settings → Terminal.

Drilling in

Clicking an agent name (or the View icon in the Actions column) opens the execution detail view at /<workspace>/agents/<agentPath>/<executionId> — the full step timeline, every action's input and output, every tool call, every model response, the cumulative cost, and any errors. That's the surface for actually debugging an agent run; this tab is the launchpad.