What’s new
- Agents Agent pages now use one shared layout across custom, policy, fraud, Seller Risk, and support agents.
- Agents Agent jobs now open in a shared job rail and job view across every agent type.
- Agents Agent updates now edit configuration in place on the Configuration tab.
- Agents Agent lists now show run and failure activity over the last 24 hours and 7 days.
- Agents Google Sheets and Docs sources now get distinct tool names per source.
- Integrations Connection setups can now be created, edited, previewed, and reused from the app.
- Integrations Agents can now use ready connection setups as context sources and update them by revision.
- Integrations Connections now show what depends on them before setup changes, archive, disconnect, or delete.
- Integrations Connection API testers now run confirmed write actions, not just read tests.
- Integrations Google Drive, Docs, Sheets, payment, risk, SQL, storage, and SharePoint operations now appear in the connection catalog.
- SOPs Rule changes raised on older rulebook versions can now publish when the current rule still matches.
- QA Assistant links now use the right workspace path, and agent search is more precise.
- QA Evaluators can now suggest fixes on the trace steps where a job went wrong.
- QA Golden CI now fails when an agent’s answers change from the baseline.
- QA Experiments with many evaluators now keep job rows visible and scrollable.
- QA Datasets can now opt in or out of CI backtests from Agent Evaluations.
- Dashboard Knowledge metrics now have a monitoring page with filters and drill-down into jobs and proposals.
- API Collection metrics are now available through built-in knowledge tools and MCP.
- API CPSC recall tools now come from Atlas with one Roe key and no extra setup.
- Models GPT-6.1 Terra is now the default agent model, with GPT-6 Luna for internal steps.
- Models Evaluators can now run on any supported model available in your tenant.
- Security Security updates cleared open dependency alerts across Roe services.