Research Agent
Custom Agent Runtime
Product
See the work before you ship it.
Follow activity, inspect outputs and find the decisions that need you. Your runtime keeps execution.
Actual product shellIllustrative sample
Open full-size workspace image ↗ (new tab)A view for each operating question.
Activity & Runs
Know what started. See what changed.
Follow named work, reported progress and terminal outcomes without opening every runtime log.
Attention & Approvals
Give human decisions a clear place.
Find failures and reported review requests beside the work they affect.
Outputs
Put the output beside the work.
Inspect the result and understand the run that produced it.
Health
Know the signal. Preserve the unknown.
Keep runtime health, event freshness and activity separate.
Client View
Show selected work. Keep the operational context.
Give an authenticated client a focused read-only view of reviewed work.
Explore five questions with a sample.
Illustrative sample · These scenes explain the interface without executing a runtime.
Northstar AIDemonstration dataRun
Competitive Landscape Research
Running
Running
Activity history / UTC
- Research started
run.startedGathering source material
Artifact
No artifact reported. A started run is not an output.
Attention
No human decision reported in this sample state.
Explore the sample
Inspect the run, artifact metadata or review request. These controls reveal demo context and send no runtime action.
An active agent. A named piece of work.
run.started reports work underway. No percentage or estimated finish is invented.
Each question selects a deterministic sample. Completed and failed outcomes use separate run identities; they are not successive states of one run. Manual selection stays paused until you choose Play.