Mage Pro Release Notes

Highlights

• Autopilot self-healing: When enabled for a project, Autopilot investigates failed pipeline runs and attempts autonomous repairs. Run lists show healing and retry progress, and repair details let teams inspect the changes and outcome.
• Mage Agent for Claude Desktop: Install a standalone Claude Desktop extension on macOS or Windows to use Mage Agent tools and skills. It includes the agent executable and browser sign-in, without requiring a separate Python or Claude Code installation.
• Trigger day calendar: See upcoming starts and execution status for enabled recurring triggers in a day view, with timezone, search, and filtering controls.
• Analytics charts across pipeline outputs: Choose a fixed output partition, the latest completed partition, the latest N partitions, or all completed partitions when refreshing a chart. Existing charts keep their saved source behavior until you change it.
• Pipeline run actions: Cancel selected active runs across pipelines. In Standard pipelines, retry a generated dynamic child and the downstream work that depends on it while preserving unaffected siblings.
• AI Sidekick: Investigate dependency-related pipeline failures with Fix with AI, review a proposed requirements change before it is applied, and continue chatting while an active turn is running. Chat history, artifacts, and failure messages are easier to use.
• Find work faster: Applied filters are visible and removable on pipeline, run, and trigger lists, and persist when you navigate away and return during the same browser session. Custom UTC ranges can be entered directly for run and monitoring views.
• Navigation and appearance: Grouped navigation makes dashboard, monitoring, library, and tool destinations easier to find. A System theme follows your device appearance, alongside refreshed light-mode contrast.

New capabilities

Autopilot self-healing

• Enable self-healing for a project: When Autopilot is available in the deployment, enable it for the project and configure the Self-healing routine to investigate pipeline failures and attempt repairs automatically. A project is enabled only after its registration succeeds; deployment-wide availability alone does not enable it.
• Follow repair progress: Global, pipeline, and trigger run lists show Healing while a repair is active and Retrying while the associated run retries. A partially healed state distinguishes a repaired block from a pipeline that has not completed successfully.
• Inspect what happened: Open a run's Autopilot status to review repair explanations, captured changes, attempt history, and a recovery-run link when one exists. Healed appears only after the pipeline run completes with a recorded repair.

Analytics charts and data exploration

• Select pipeline output partitions for charts: Keep a chart on one saved partition, follow the latest completed partition, combine the latest N completed partitions, or include all completed partitions. Mage reevaluates dynamic selections when you explicitly refresh a chart or dashboard; reloading the page keeps the saved chart snapshot.
• Keep existing charts stable: Charts already saved retain their source selection until you change it. Partition selection reports missing outputs and safety-limit failures instead of silently charting an incomplete set.
• Manage connections from Data Explorer: Open connection management from the Data Explorer toolbar in a new tab while keeping your exploration page available.

Learn more: Explore Analytics dashboards | Read the analytics docs

Mage Agent for Claude Desktop

• Install a standalone extension: Download the Claude Desktop extension for macOS (Apple silicon or Intel) or Windows. The package includes the Mage Agent executable, so a separate Python or Claude Code installation is not required.
• Connect to Mage: Sign in through the browser, then use Mage Agent tools and skills from Claude Desktop.

Learn more: Read the Mage Agent docs

Trigger day calendar

• See planned work: Open Calendar from Triggers to view upcoming starts and execution status for currently enabled recurring time triggers, organized by day and timezone.
• Find and inspect starts: Search and filter by status or tag, inspect overlapping starts and their details, and open related runs. The calendar refreshes today's view without changing trigger scheduling.

Learn more: Read the trigger calendar docs

Pipeline runs and troubleshooting

• Cancel only the runs you select: From the project-wide Pipeline runs page, select Ready or Running runs across pipelines and cancel those runs without cancelling other active runs.
• Retry dynamic work precisely: In Standard pipelines, retry a generated dynamic child and its dependent downstream work while preserving unaffected siblings. Consumers of an affected reduced collection or shared output rerun when they depend on it.
• Inspect logs with clearer controls: Typing filters the logs already displayed; Search or Enter searches matching logs across the full selected time range. Custom UTC ranges require a valid start and end before applying.
• Keep results during refresh errors: If a later logs refresh fails, the last loaded results remain visible with a retry action.

Learn more: Read about pipeline run actions and filters | Read about retrying block runs | Read the monitoring docs

AI Sidekick

• Investigate dependency failures: When AI Sidekick is enabled, Fix with AI can examine relevant failed-run evidence and suggest a change to requirements.txt. Review the proposed diff before an authorized edit; confirmation follows the environment's action policy. Installation, verification, and manual retry remain separate steps.
• Continue a conversation during a running turn: Send a follow-up while Sidekick is busy. Queued messages wait for backend acceptance, with retry handling when the session is temporarily busy.
• Navigate and recover chats: Resize Sidekick without losing artifact controls, see a reason when a turn fails, page through active chat history, keep archived chats archived, and preserve renamed titles during live updates.

Learn more: Explore AI Sidekick | Read the AI Sidekick docs | Read the dependency management docs

Filters, dates, and navigation

• Return to your filtered lists: Status, tag, type, and other applied filters on Pipelines, Pipeline runs, and Triggers stay in place when you leave and return during the same browser session. A filter band shows active selections and lets you remove individual values or clear them together.
• Enter exact UTC ranges: Type start and end dates and times directly or choose them from the picker in run and monitoring date filters. Incomplete or invalid ranges cannot be applied.
• Discover recent features: Browse curated product highlights from within the workspace.
• Use grouped navigation: Find related dashboard destinations through grouped flyouts, with keyboard and responsive layouts supported.
• Follow your device theme: Choose System, Light, or Dark from the account menu or personalization settings. System follows browser or operating-system appearance changes; explicit selections persist after reload.

Learn more: Read the workspace personalization docs

Improvements

• Dashboard charts: Chart refreshes show clearer progress, renamed charts retain the right heading, and date-axis labels are easier to read.
• Monitoring focus: Global Monitoring now emphasizes run health and active runs. Average duration and Duration Trend are no longer shown there; runtime trends remain available on individual pipeline monitoring pages.
• Alerts settings: Edit and save valid alert settings even when another notification channel has a configuration error. Invalid channels are shown in the UI, and a malformed webhook URL does not prevent Email alerts from working.
• Workspace access: Users with workspace-specific roles but no global role can find their assigned workspaces in the management directory under the supported authentication configuration, without gaining workspace-management privileges.
• Mage Pro API authentication: Clusters can opt into service authentication using a customer-owned API key for Mage Pro API requests that have no user identity. The setting is disabled by default; requests with a user identity continue to use user authentication, and failed service-token exchanges do not fall back to a user identity.
• AI usage attribution: Sidekick AI usage is attributed from trusted server-side context instead of accepting an unverified client-supplied user identity.
• Executor memory guard: Configure the local Python executor's maximum memory-usage threshold for pipeline runs with MEMORY_USAGE_MAXIMUM; the default remains 0.95.

Learn more: Read the workspace access docs | Read about Mage Pro environment variables

Reliability and fixes

• Integration setup and execution: Make relative-date expressions work in integration configurations, retain edits when source templates first load, and place newly inserted transformers before exporters.
• Dynamic stream runs: Preserve named child UUID suffixes and upstream metadata in dynamic stream runs and the run graph.
• Autopilot jobs: Use the correct project storage and inherited database service for cross-namespace repair jobs.
• Backfill log retention: Prevent automatic cleanup from deleting logs that still belong to active or recent backfill pipeline runs.
• Navigation and sign-in: Preserve cross-workspace destination paths, OAuth callback query values, and selected workspace roles across navigation. Improve sign-in notice and block-menu contrast.
• Workspace role dropdowns: Show roles for every workspace on the current user-editor page, not only the first.
• Filters and editor: Preserve repeated query values in resource filters, keep Coder file paths and version-history URLs through refreshes, and stabilize template browsing controls.
• Workspace recovery: Keep editor connections stable across tabs and reconnects, and provide a retry action when Home cannot load the workspace.
• Browser telemetry: Reduce captured browser event payloads, including request headers and bodies.
• Account and visual polish: Improve light-mode colors, filter-menu visibility on small screens, and account controls for pipeline collaborators.

Release metadata

• Image tag: 20261023-1000-py3.12
• Release ID: 20261023-1000
• Repository: mage-ai/mage-pro
• Date: October 23, 2026
• Start commit: 3734a1b61829
• End commit: 4a46fde19a95

Questions or feedback? Check workspace Settings for help links, or contact Mage support.