Skip to main content
Open Evaluations in the Graph Studio sidebar (?tab=eval). Panel title: Agent Evaluations. Header link History jumps to workspace History / Sessions.
Connect to Build, Run on draft, Experiment, Signal

Agent Evaluations panel

Entry points

Copy under the title: Choose how you want to evaluate this agent. Run history and analytics live on the workspace Evaluations page.

Signal panel

After Signal:
  • Copy explains enabling/disabling signal ingest per build.
  • Columns: Build · Version · Since · Signal (toggle).

Experiment

Selecting Experiment shows placeholder copy (Compare different builds, or build vs draft.) without a full wizard yet. Prefer Connect to Build or Run on draft for production scoring.

Deep dives

  1. Connect to Build — Mode → Build → Metrics → Review.
  2. Run on draft — Dataset → cases/sessions → Metrics → Launch.
  3. Workspace Analytics / History — results after runs complete.