Skip to main content
June 27, 2026 This release improves how teams configure, review, and scale agents in Roe. Highlights include the new policy WYSIWYG editor, reviewer comments on agent outputs, configurable pricing rules, and a broad set of security hardening updates.

What’s new

  • Agents Agent result pages now support threaded comments, highlights, replies, and resolved notes.
  • Agents Agent jobs now support batch-level metadata at run time and after launch.
  • Agents Agent job pages now show total job counts, including filtered counts.
  • Agents Agent lists and detail pages now load job counts without full-history recounts.
  • Agents Stale agent jobs now reconcile against closed workflow states before timing out.
  • Agents Merchant Risk agents now expose add-on toggles in agent configuration.
  • Agents AML agents now include social scraping during investigations.
  • Agents Fraud Investigation agents now support IP, email, phone, and distance enrichment tools.
  • Agents Interactive web agents now support custom output schemas and flat per-step billing.
  • Platform Agent configuration sliders now drag smoothly and stop resetting while typing.
  • SOPs Policy instructions now open in a WYSIWYG editor when the content is supported.
  • SOPs Knowledge bases can now be linked to agents by Roe knowledge base ID.
  • SOPs Knowledge bases now support unlinking without deleting the underlying Atlas lens.
  • SOPs Knowledge base pages now load from the correct routes and return larger lists.
  • Dashboard Non-policy agents now use the standard metrics view again.
  • API Agent job APIs now return customer-safe cost breakdowns for completed jobs.
  • API Agent job comments are now available through backend APIs.
  • API Agent pricing now resolves from configurable organization and agent-level rules.
  • Security Dataset lists and project APIs now enforce validated organization scope server-side.
  • Security Dataset uploads now block active web files and force unsafe previews to download.
  • Security Webhook callbacks now block internal and non-public destinations.
  • Security SQL editor queries now block cross-database access and database enumeration.
  • Security SQL query errors now return sanitized messages instead of raw database details.
  • Security SQL query helper requests now enforce stricter per-user and per-organization rate limits.
  • Security Ingress TLS policies now disable weak TLS 1.2 CBC cipher suites.
  • Workflows Large workflow loops now run in shards of 25 items.
  • Workflows Workflow live trace streaming now caps at 250 events with a truncation marker.
  • Workflows Workflow scheduler admission now reserves separate capacity for orchestrator parents.