OpenFrame v1.0.4 - First GA Release: No Access Codes, Free Until Aug 15

OpenFrame is now generally available. Access codes have been fully removed from the registration flow - anyone can sign up directly without a code. The platform is free of charge for all users until August 15th, giving MSPs a full window to onboard, explore, and deploy without any billing friction.

MSPs can now see their projected billing costs directly within product settings, giving full cost visibility before any charges apply. This is live alongside the free-until-August-15th period so teams can plan ahead.

OpenFrame now ships its own native, open-source RMM agent built directly into the platform. The agent owns the full endpoint stack - monitoring, script execution, patching, remote access hooks, and reporting - with no third-party dependency. MSPs can audit, fork, and self-host the agent. TacticalRMM has been fully decommissioned across all environments.

A guided, step-by-step onboarding flow walks MSPs from first login to first client, first agent deployed, and first ticket worked. The checklist is surfaced directly in the UI, tracks progress in real time, shows what's done and what's next, and covers workspace setup, client creation, agent deployment, ticket flow, and Mingo. Steps are skippable for experienced users.

Fae now supports two independent levels of AI assistant customization: one at the MSP level (applied globally across the MSP's workspace) and one at the individual customer level (overrides or extends the MSP-level settings per client). This gives MSPs fine-grained control over how Fae presents and behaves for each customer relationship.

A dedicated page for viewing and managing archived devices is now available. Previously there was no UI to browse archived devices. The device status filter has also been fixed and aligned with actual statuses (online, offline, archived).

A comprehensive /settings/billing page now covers all subscription states (start, OK, 90% limits, out of limits, overdue, AI-only limit, device-only limit), plan selection with annual/monthly toggle, pay-as-you-go mode, device limit modals, subscription guards, invoice history pulled from Stripe, and a cancellation flow.

Scripts can now be executed natively through the OpenFrame RMM agent across Windows and macOS, with proper session management, timeout handling, and result processing. Running scripts from the devices page now correctly prompts for required variable input before execution, matching the behavior from the scripts page.

A reusable calendar-based date filter component with date range sorting has been added and applied across all relevant tables platform-wide, including organizations, script schedules, and logs.

Scripts requiring variables would execute immediately without prompting for input when triggered from /devices, causing failures. Now correctly opens the variable input prompt before running, consistent with the /scripts page behavior.

Some customers displayed a dash instead of a count despite having active devices, and others showed inflated counts that included archived devices. Device counts now reflect only active, non-archived devices.

Standard paste and other editing shortcuts (Undo, Redo, Cut, Copy, Select All) were broken in the desktop client on macOS because the custom app menu had replaced Tauri's default menu. A native macOS Edit menu has been restored.

External links in Mingo guide mode were either not opening, navigating inside the app instead of the browser, or breaking the active guide session. Links now correctly open in the user's default browser without interrupting the guide.

On pages reachable from multiple sources, the back button always redirected to a hardcoded default rather than the actual page the user came from. Back navigation now correctly returns to the previous page, with edge cases for deep links and refreshed pages handled.

The header user menu dropdown (Profile Settings / Log Out) rendered behind the open Mingo AI chat drawer, making it invisible and unclickable. The dropdown z-index has been raised above the drawer.

The Mingo chat drawer's mobile breakpoint did not match the design system's md breakpoint (800px), causing layout inconsistencies. Now aligned to take full-screen width below 800px.

The 'Sign in' button on the unauthorized overlay pointed to the auth root, landing users on the Sign Up tab. It now correctly points to /auth/login/ so users with existing accounts reach the Login tab directly.

When domain registration returned a 409 conflict due to zero available cluster capacity, users saw a generic or unclear error. A clear, actionable error message is now shown explaining the capacity issue and suggesting next steps.

The logs page details buttons had incorrect padding and a misplaced icon. Fixed padding, removed the erroneous icon, and added a 'more actions' menu with 'copy log details' and 'device details' options. Log details sidebar spacing and stroke were also corrected.

The device name was resolved from different sources (hostname, display name, various API fields) depending on the page, causing the same device to appear under different names across the dashboard, device details, device lists, and scripts. Now unified to a single canonical source.

Fixed incorrect image tag health checks at the Wait for Management Init Containers step in the saas-tenant repos, preventing tenant bootstrap failures.

The 'AI-Assistant Appearance' block on /customers/edit and the 'Custom AI Assistant' tab on customer details showed a full-page loader (icon + 'Loading' text) instead of skeletons. Replaced with skeleton loading states. 'Edit Default Appearance' now deep-links directly to /settings/ai-settings with the Customer AI Assistant tab in edit mode. 'Use default appearance?' confirmation now immediately deletes the custom appearance with pending state and toast feedback.

The Mingo AI Chat tab icon has been replaced with MingoMonochromeIcon (takes accent color when active). The full-page loader on /settings/ai-settings page enter and tab switch has been removed; the tab bar now renders immediately with skeleton content while data loads.

NotificationDrawer, TimeTrackerPanel, EmbeddableChat, and MobileBurgerMenu had inconsistent backdrop styles. Unified to a single shared dimmed background without blur. Mingo Chat now renders above the mobile burger menu. The two are now mutually exclusive - opening one closes the other.

All access code functionality has been removed from the system - UI elements, API endpoints, database fields, configuration references, and related code paths. Registration is now open without any access code requirement.

TacticalRMM servers, scheduled jobs, Helm charts, Terraform resources, DNS records, and monitoring dashboards have been removed across all environments. All TacticalRMM paths on the shared gateway are now explicitly blocked (returning 410 Gone). Customer scripts from TacticalRMM have been restored into the native OpenFrame RMM scripts system.

The category system in the Scripts section has been replaced with tags across the list view, detail page, create/edit forms, and filters. The scripts list table has been redesigned with a chevron-based action pattern, 'Run Script', 'Schedule Script', and 'Delete Script' in the actions dropdown, and description moved below the script name. The Script Details page now uses a three-tab layout: Script Details, Script Schedules, and Execution History.

The Scripts page now uses server-side pagination, search, and filter controls to handle growing datasets without client-side performance degradation.

Script arguments and environment variable fields now show one empty input row by default with an 'Add' button for additional rows. The syntax editor on the script editing page now expands to fill all available vertical space.

Script execution output from the native OpenFrame RMM is now truncated to a reasonable cap, with the beginning and end of output preserved. This prevents oversized payloads from degrading Mingo AI response quality or hitting token limits.

All native RMM AI tool executions (script run, device actions, etc.) are now wrapped with a stoppable executor, allowing long-running or stuck operations to be cancelled gracefully so Mingo-driven workflows don't hang indefinitely.

The agent now refreshes its authentication token preemptively based on TTL rather than reactively after receiving a 401. This results in more stable WebSocket connectivity, fewer 401 errors, and smoother Fae interactions.

Tenant deployment has migrated from cluster-per-tenant to namespace-per-tenant on shared Kubernetes clusters. New manifests provide isolated tenant environments with RBAC, NetworkPolicy, ResourceQuota, LimitRange, and ServiceAccount primitives. Terraform modules for multi-tenant cluster and namespace creation are now in v1.

Logs from all tenant microservices (Java services, MeshCentral, Fleet) and tenant devices are now forwarded to the shared Loki instance and available in Grafana for troubleshooting and analysis, with capacity validated for scaling to thousands of tenants and millions of devices.

Fleet's vulnerability database directory (/tmp/vuln) is now backed by a PersistentVolumeClaim, preventing the full ~800MB re-download of NVD, CPE, OSV, OVAL, MSRC, and other feeds on every pod replacement. Only incremental delta updates are fetched after the initial load.

Orphaned devices stuck in pending status from the earlier provisioning period have been identified and removed across all tenants, improving dashboard accuracy and reducing clutter.

Mingo AI can no longer create or suggest scripts targeting Linux, since Linux is not a supported platform. Only Windows and macOS are available as target OS options for AI-generated scripts.

The MSP branding bar (company image and name) is now always visible on the main AI Assistant chat screen, ensuring end-users always see which MSP is managing their IT.

Approval blocks in Fae chats now have two distinct versions: an end-client view that hides the raw command and shows only the description and approval buttons, and an admin view that uses the same approval block component as Mingo chat.

Messages sent on a previous day now display the full date alongside the time in message timestamps. Messages from today continue to show time only.

The chat input field and send button are now disabled while a command or batch of commands is executing, with a visual in-progress indicator. Input is re-enabled once all commands complete or fail, preventing conflicts from concurrent input.

On /tickets (board view), the Customer and Assignee autocomplete filters are now hidden on mobile and replaced with a filter button that opens a 'Filter Tickets' modal with Reset/Apply actions. ModalV2 panel width is capped to the viewport on narrow screens.

Admins and technicians can now leave internal notes on tickets. Notes are visible only to admins and technicians - not to end clients.

Backend notification templates now generate and serve correctly formatted description strings for all notification types, including new customer, customer details updated, customer archived, new device deployed, and other system events.

A dry-run billing mode has been implemented that shows users estimated costs without real Stripe charges, invoice generation, or screen locks. Usage tracking for devices and AI tokens operates independently from Stripe in this mode.

The onboarding experience has been redesigned as a dedicated standalone page (replacing the modal/overlay approach) with three sections: Device Management & Customers Setup, AI Experience & Tickets, and Additional Setup covering Company/Team and Knowledge Base.

Login and sign-up page layouts, typography, spacing, and component styles have been updated to match the current design system. Auth page skeletons have also been updated to reflect the current form field layout and button placement.

Field status messages now wrap up to two lines. Paired form fields are stacked into a single column on all breakpoints. /auth/login now opens in an empty state, showing SSO providers only after Continue. SSO provider buttons default to 'Continue with ...' labeling.

Draggable scrollbar thumb (elevator-style scroll) has been implemented across all scrollable areas platform-wide - tables, lists, sidebars, modals, and panels - via a single shared component for consistent behavior and styling.

The dashboard card layout and styling has been updated to match the current design direction. The gradient effect on tabs has been corrected. A preview mode for the EmbeddableChat component now lets MSPs see how their chat widget will look before deploying.

Tenant Helm values have been refactored to use the openframe-saas.image.tag key instead of the global.image.tag key across saas-tenant repos, improving configuration clarity and reducing misconfiguration risk.

Tenant bootstrapping now uses an ApplicationSet approach, enabling more scalable and consistent automated provisioning of new tenant environments.

Tenant storage has been migrated from per-tenant buckets to a shared bucket model, reducing infrastructure overhead and simplifying storage management at scale.

A Grafana dashboard for client engineers is now available to monitor tenant health and work with logs, with filtering by tenant domain, machine ID, and log level, and 30-second auto-refresh.

The CLI codebase has been reviewed, refactored for quality, and planned for future enhancements to improve maintainability and performance.