← All changes
getting startedorganizationevaluationsagentsskills and tools

One interface, workspaces, named fixtures, and run review with the assistant

The dashboard drops its view modes for one sidebar with a Developer menu, "workspace" replaces "organization" throughout, fixtures take a label, and the assistant can review a run.

New

  • Fixture names. A fixture now takes an optional label — "Order not found", "Payment declined" — shown in place of the tool id in the suite's fixture list and in each case's picker. Two fixtures on the same tool no longer read as duplicate rows. Existing fixtures are unaffected and keep showing their tool id until you name them. See fixtures.
  • Run review in the assistant. Ask the in-dashboard assistant how a run went and it will find the run, report the overall result and each case's scores, and open a failing case to show the AI judge's reasoning and any human review recorded against it. Reading results is free — no interactions are charged.
  • Where a case failed because the test case was wrong rather than the agent — a stale reference answer, an unrealistic expectation, a fixture it never ticked — the assistant can revise it. The correction is saved as a new version, so earlier runs keep the definition they were scored against. Starting a run is still done from the Evaluations tab, so you always see the cost first. See reviewing a run with the assistant.
  • Reasoning effort on an agent's model settings — on models that support it, choose how hard the model thinks before it answers. Low is faster and cheaper, High helps on the questions your agent gets wrong today. See models.
  • Resolve Relative Date, a new tool on an agent's Tools tab, works out dates a calendar phrase only implies — "the second Tuesday of next month", "the last Friday of this month", "in 10 business days", with weekends skipped. See built-in tools.

Improved

  • One interface. The Simple / Developer / Assistant view toggle is gone. Every feature is available to everyone, and everything technical now lives under a single Developer section in the sidebar. Nothing was removed — only regrouped. See finding your way around.
  • Developer holds three groups: Operate (Live Sessions, Contacts, Conversations), Build (Skills, Tools, Data Tables, Assets, Sandbox) and Analyze (Analytics, Usage). Sidebar sections remember whether you left them open.
  • Workspace replaces "Organization" and "Tenant" everywhere in the dashboard — one word for the thing your agents, members, knowledge and billing belong to, distinct from the personal account you sign in with. Nothing to change on your side. See workspace.
  • The workspace danger zone now reads Close workspace rather than "Close account". Closing a workspace never closed your login; the old label said otherwise.
  • Usage gains a Top agents chart, and token usage and rate limits moved behind a disclosure so the quota bars and activity charts lead. Find it under Developer → Analyze → Usage. See interaction counting.
  • The dashboard Home page is retired. Its charts moved to Usage, and a workspace with no agents now lands on the welcome view under My Agents. Existing links to it open chat.

Fixed

  • Agents could pick the wrong day when you said "next Tuesday" or "this Friday". Every agent now knows the current weekday and the week ahead, on every channel. Nothing to change on your side.