Changelog

Improved

• Workflow execution timing now separates Execution Duration, Wait time, and Elapsed time for human approval steps, including a live waiting counter and clearer logs and notifications.
• Latency alerts and execution dashboards use Active duration, so time spent waiting for human approval no longer inflates execution latency.

New

• Create, configure, and publish responsive Apps from a redesigned builder with chat, canvas, palette, and inspector tools, plus access controls, version activation, rollback, and unpublishing.
• A unified chat experience adds streaming responses, visible reasoning and tool activity, approval steps, model selection, and a full-screen workspace resource panel.
• Workspace Docs now support content search, folders, bulk move and delete, drag-and-drop organization, edit provenance, user and workflow mentions, and share links for individual documents.
• The organization home now adapts to each team's maturity and needs with setup guidance, actionable alerts, searchable and pinnable workspaces, and a keyboard command palette.
• Authorization settings now include a role editor, access explorer, access inspector, change-impact previews, and filterable activity history.
• Organization-scoped SSO can now be configured by administrators and automatically provisions successful sign-ins into the correct organization and workspace.
• Workflow Value & Cost planning adds a public estimator plus observed execution cost, ROI, budget controls, and recommendations inside the workflow panel.
• Google Calendar polling triggers are now available in the workflow builder and native trigger discovery.
• Export generated workflow clients with portable metadata and a clearer handoff into full chat.
• A new public security practices page explains how SteelEngine approaches platform and data protection.
• The public Tables API can atomically claim and update one matching row, allowing concurrent workers to use tables as safe work queues.

Improved

• The workflow canvas now previews output schemas and last-run values, visualizes which connections carried data, and surfaces broken references before execution.
• AI and MCP workflow builders prefer native integration, data, and control blocks and provide clearer guidance when a workflow falls back to raw API or code blocks.
• Chat now preserves model choices across navigation, labels the model used for each response, supports a wider model roster, and provides reasoning-effort controls.
• Empty states provide clearer next steps, while chat, workflow, and document drag-and-drop follows the displayed order and preserves cross-folder drop positions.
• Integration failures provide more actionable messages, and document, Notion, and other URL outputs are easier to open directly from workflow results.
• Workflow execution is more durable across long runs, worker restarts, and queue recovery, with consistent hosted and bring-your-own model credentials and safer OAuth token refreshes.

Fixed and secured

• Environment secrets are redacted from block logs, and database connections are pinned to validated addresses to reduce DNS-rebinding risk.
• Human approval steps in child workflows now pause and resume the parent workflow with the correct execution and trigger context.
• Condition blocks evaluate only referenced outputs, preventing unrelated large payloads from exceeding evaluation limits.
• Workflow validation no longer flags valid references in several supported configurations and reports genuinely broken references more consistently.
• Existing organization members receive requested workspace access immediately, and default workspace creation is protected against duplicate concurrent setup.
• Sandbox and workflow failures are isolated to the affected execution instead of destabilizing other queued work.

Improved

• Queue recovery now reconciles workflow jobs safely across multiple workers, reducing duplicate execution and stalled runs during worker restarts or lease recovery.

Fixed

• Condition blocks now evaluate only the outputs they reference, so workflows with large upstream or child-workflow payloads no longer fail because unrelated data exceeds the evaluation request limit.

New

• The public Tables API can now atomically claim and update one matching row. Concurrent workers can safely use a table as a work queue without processing the same pending row twice.

Improved

• Workflow execution now contains runtime failures, coordinates dispatch across workers, and scales workers with queue pressure for more predictable performance under load.
• Workflow state and attempts are recorded more durably, improving recovery for long-running and interrupted executions.
• SteelEngine logos, favicons, documentation branding, and collapsed-sidebar presentation are now consistent across the product.

Fixed

• Sandbox and workflow failures are isolated to the affected execution instead of destabilizing other queued work.

Improved

• Documentation link previews now use the current SteelEngine branding.

Fixed

• Human approval steps inside child workflows now pause the parent workflow instead of allowing it to continue prematurely.
• Child workflows preserve the execution identity and trigger context needed to resume successfully after approval.
• Resuming a paused workflow through MCP now starts the execution immediately and supports structured approval input.

Improved

• AI and MCP workflow builders now search for native integration, data, and control blocks before falling back to raw API or code blocks.
• MCP input schemas are compatible with Claude Code, preventing SteelEngine domain tools from being dropped by the client.
• Atlassian OAuth now uses the correct provider-specific production configuration for more reliable Jira and Confluence connections.

Fixed

• Realtime workflow runners now include the shared runtime utilities required to start successfully.

New

• A public Workflow Investment Calculator can estimate cost per run, monthly investment, potential time savings, major cost drivers, and optimization opportunities from a workflow description and supporting documents.

Improved

• Integration failures now return clearer, actionable messages across Gmail, Slack, Atlassian, Confluence, databases, file storage, document parsers, and Microsoft and Google services.

Fixed

• Inviting an existing organization member to a workspace now grants the requested workspace access immediately.
• Default workspace creation is serialized to prevent duplicate setup when multiple organization requests arrive at the same time.