This page is the operational guide to running, watching, and stopping Deep Research jobs.
Starting a run
The first time you open a property, the workspace shows a setup panel at the top:
- (Optional) upload documents — rent roll, OM, P&L.
- (Optional) toggle the Units pipeline version (v1 Stable vs v2 Beta).
- Click Start Deep Research.
The setup panel disappears and section cards begin populating in real time.
For a property that's already had a run, you can re-run from the property header (full re-run) or from any individual section card (per-section re-run — see Per-section reruns).
Watching progress
Each section card has a status pill:
- Idle — never run.
- Queued — waiting on capacity.
- Running — agent is actively working.
- Done — finished successfully.
- Failed — finished with an error. Click for details.
The chat panel is disabled while sections are queued or running, and re-enables as each finishes.
Queue position
Behind the scenes Deep Research is queued — there's a finite pool of agent capacity per organization. If your queue position is >0, the section card shows "Queued (position 3)". Self-Serve customers occasionally see queue positions during peak times; Enterprise customers with higher concurrency rarely do.
If you have a Private Server add-on, you have your own dedicated agent pool with no shared queue. See Private / Dedicated server add-on.
Canceling
There are three ways to cancel:
- Cancel one section. Click cancel on the section card while it's running. Its status flips to Idle and the agent's resources are released.
- Cancel everything. From the property header, Cancel all. Drops everything currently queued or running for this property.
- Org-wide cancel. If you've started runs across many properties and want to stop everything, contact support — there's no UI button for this.
Canceled runs do not consume a pro-forma credit. (Pro-forma is a separate explicit action anyway.)
What happens to partial data
If a run is canceled or fails mid-way, sections that already completed are kept. Sections in progress are discarded. You can re-run only the failed sections later — see Per-section reruns.
Common reasons a run is slow
- A rent roll with 1000+ units takes longer to extract.
- A property in a tertiary market with thin data coverage triggers more retries on the v2 units pipeline.
- A document upload is still indexing — the run will wait for indexing to complete before starting any document-dependent agents.
- The org's queue is busy.
See Why is Deep Research slow? for the full diagnostic.