Skip to content
← Back to release notes
v0.111.0 stable major release

Release v0.111.0

Sep 09, 2026

Work aimed at an offline device now waits for it instead of failing, running scripts and automations have a real Stop button, and every customer gets an organization record page with contacts, sites, devices, tickets, contracts, and billing in one place. Devices you cannot install an agent on are first class, configuration policies can inherit from a partner-wide baseline, and a 62-screen pre-release browser sweep with agent service installs on real Windows and Linux hosts went in before the tag.

Offline Work Queue

  • Work aimed at a device that is asleep or offline now waits for it instead of failing. Patch jobs, automation script and command actions, and scan or rollback work are held with a delivery deadline and picked up the moment the device checks in — so a nightly run across a fleet of laptops no longer shows a wall of red the next morning.
  • A waiting step reads 'Queued — device offline' instead of Failed. Queued work is listed on the device page where you can cancel anything you have changed your mind about, and it expires on its own if the device never comes back: a week for ordinary work, a day for inventory and power actions.
  • Every automation action gets an 'If the device is offline' control, so you choose Queue or Skip per step. Steps that are only meaningful against a live machine can opt out.
  • This is on by default, and existing automations queue until you set a step to Skip — see the upgrade notes.

Scripts & Automation

  • Stop a running script from the UI. Execution history and the run detail page each get a Stop button, Force stop is there for a script that is ignoring the polite request, and Cancel run stops an automation across every device at once.
  • Cancellation is real, not cosmetic: the agent terminates the script and everything it spawned, on Windows, macOS, and Linux. If a stop lands after the script already finished, Breeze says so plainly instead of pretending it worked.
  • A fully cancelled automation run now reports Cancelled rather than Failed, so a deliberate stop no longer reads as an outage in your history.
  • An individual script can be acknowledged past the strictest security pattern checks, so a routine configuration script is no longer hard-blocked — while genuinely dangerous patterns stay blocked for everyone.

Customer Records & Service Desk

  • Every customer now has an organization record page: Overview, Contacts, Sites, Devices, Tickets, Contracts & Billing, and Activity in one place, with tickets, invoices, and quotes linking straight back to it.
  • A partner-level Service Management mode decides whether Breeze's own service desk is on, off, or handled by an external PSA. Switching it off hides the Service Desk section and stops ticket creation everywhere — the web app, the customer portal, email-to-ticket, and the AI assistant. External mode is API-only in this release.

Devices Without an Agent

  • Manual assets are a first-class device class. Add anything you track but cannot install an agent on by hand, look up its warranty, and see it in inventory reporting alongside your agent devices.
  • Add network devices yourself instead of waiting for discovery to find them — including a website or URL as a target, with a one-click hand-off to set up an HTTP monitor against it.
  • The unified device list is on by default, with All, Agent, Network, and Manual segments, and switches, firewalls, printers, and NAS get their own network device page.

Configuration Policies

  • A configuration policy can inherit from a partner-wide baseline, with Override and Revert per feature tab and banners showing what it inherits from and how many policies inherit from it.
  • Every reader now resolves inherited settings properly — agent delivery, automations, patch scheduling, event log, and monitoring. A policy that was marked as inheriting but silently delivering nothing will start delivering the real settings, so check your inheriting policies before you upgrade.

AI & Automation

  • Creating an AI agent is now a four-step guided flow with a review card, and the free-text tool allowlist is replaced by a picker of 15 capabilities that spell out exactly what the agent may do on its own.
  • Pick the specific scripts an agent is allowed to run unattended, and see how a new agent's autonomy projects against your partner ceiling before you create it.
  • AI Operator, off by default: delegate a service-recovery incident from a device or an alert to an operator task that keeps working after you close the browser, with its own task page, a device feed tab, and approvals through the existing inbox.
  • AI assistant runs report their script outcomes truthfully, an approved-by-worker result is no longer shown as a failure, and command approvals explain the impact in terms of the command type.

Migrating & Custom Fields

  • Import custom fields from another RMM — Datto, NinjaOne, ConnectWise Automate, N-central, or a CSV — bringing across both the field definitions and the per-device values, including warranty data, with a preview before anything is written.
  • Custom fields are stored properly now rather than as a blob, with one namespace per device, so a field defined for your whole partner account can no longer be silently shadowed by an organization-level field of the same name.

Device Lifecycle

  • Removing a device now asks about the agent and uninstalls it by default. Previously a wiring bug meant the uninstall was effectively never sent, so agents kept running on devices you had removed.
  • Bulk restore for removed devices, bulk permanent delete with a progress indicator, an uninstall-state badge on removed devices, and an optional retention policy that purges removed devices after a number of days you set.
  • The device list gets an Agent Version column, coloured by how each device compares to the version you have pinned.

Billing & QuickBooks

  • Payments you record in Breeze against an invoice already in QuickBooks are now created in QuickBooks automatically, and removed there if you void or fully refund the Breeze payment. This switches on for connected realms at upgrade — read the upgrade notes first if you are not ready for Breeze to write into a customer's books.
  • Voiding a paid invoice no longer fails against QuickBooks, and voiding a synced invoice that has payments applied stops with a clear message instead of leaving the two systems out of step.
  • Account mappings sync themselves after a decision, the invoice card shows a live sync state once you issue, and Integrations stays on the Accounting tab where it should.

Security & Compliance

  • Maintenance windows with 'Suppress scripts' are honoured by every script path now, not just manual runs — the AI assistant, automations, fleet remediation, and edition auto-migration all skip a device inside a window. A skip is recorded as a skip, so the run stays green and later steps still run. A window that cannot be evaluated still fails loudly.
  • Network monitor alerts go through the full alert pipeline, so ICMP, TCP, HTTP, and DNS monitors send notifications, start escalations, and fire automations instead of only appearing in the inbox. Expect a step change in volume — review those rules and channels before you upgrade.
  • Site-restricted users and roles without alert read permission no longer see organization-wide alert, software, or endpoint-protection summaries. Some people will correctly see less data than before.
  • Delegated API keys with site restrictions can no longer bypass those restrictions when writing custom fields or distributing agent binaries.
  • Privileged access elevation is auditable off-box on Windows now, writing the whole request-and-approval lifecycle — including who asked and who approved — to the Windows Event Log and a local audit file.
  • Partner-wide configuration was reachable from the wrong place and unreachable from the right one across identity contracts, notifications, software security, automation and alert rules, AI, and custom field definitions. Both sides are closed.
  • High-assurance approver devices can now be verified against the platform itself on Android and iOS, so an approval is tied to real hardware rather than a claim.

Sign-in & Mobile

  • Actions on your own account no longer sign you out. Disabling your own MFA, adding or removing a passkey, and confirming a phone number all still end your other sessions but keep the one you are working in.
  • An administrator MFA reset clears every factor including passkeys, in one transaction, on all four reset paths — and a policy change that would lock out already-enrolled users is refused, with a count of who would be affected.
  • Recovery-code rotation can no longer save an empty set, and bulk alert actions now apply to everything you matched rather than the first hundred rows on screen.
  • Mobile device detail shows IP address, logged-in user, OS version, and open alert and ticket counts; the Systems tab issue count includes fleet findings; the new-ticket screen gets an organization selector and assignee picker; and the Home empty state shows a fleet status strip.
  • Around twenty mobile paper cuts fixed — streaming chat rendering, keyboard behaviour on ticket screens, photo upload, toast overlap, MFA auto-submit, and timesheet durations among them.
  • Emails and PDFs are generated in the recipient's language, and the web app picks the right language on the very first page load.

Remote Access & Agent

  • Remote desktop can capture the Windows logon, UAC, and secure desktop screens again — those were coming through blank on some machines.
  • Installing the agent service on a host that is already enrolled restarts it instead of leaving it stopped, on Windows, Linux, and macOS. The old behaviour stranded a machine Offline until someone started it by hand, and on Windows it reported 'running' when it was not and failed outright over an existing install. Both were verified on real Windows Server 2022 and Ubuntu hosts before the tag.
  • The viewer no longer sticks at 0 FPS on the watchdog fallback path, session teardown says why it ended so a mid-session capture failure surfaces, and Remote Tools and Overview pick up remote-access permission changes without a refresh.
  • Reboot deferral shows one dialog per Windows session instead of repeating itself, and macOS package leftovers are cleaned up on uninstall.

Reliability & Performance

  • New indexes for the per-organization process rollup, the abuse-signals sweep, and four recurring jobs that had no covering index — all built without blocking writes.
  • Metric anomaly detection no longer piles up overlapping runs against the metrics tables, which was a real incident on hosted. The trade-off: a skipped check no longer catches itself up, so a gap needs a manual backfill.
  • Dynamic device groups are re-evaluated in the background when a device changes rather than on the request, and the agent is lighter about command capture, shipped log batches, and process counts.
  • Expected database contention reports honestly instead of as a server error: software inventory answers with a retry and does it for you, and a filter preview that takes too long says so.
  • A real readiness endpoint, separate from the liveness check, so a load balancer can hold traffic until the API's dependencies are genuinely up.
  • Integration fixes: UniFi device and client lists page through fully and match MAC addresses consistently, Cisco Umbrella moved to the current activity API, and Windows service names containing spaces are accepted.
  • With a pinned agent version and auto-promote off, the download route serves the pinned version, so pinned devices no longer sit in 'Updating' forever.

Self-Hosting

  • Standard upgrade: bump `BREEZE_VERSION`, then `docker compose pull api web portal && docker compose up -d binaries-init api web portal`. Remember `portal` — it is its own container.
  • 35 database migrations apply automatically on first boot. Take a snapshot first. Six indexes on tables that can be large build without blocking writes, and backfills log their row counts as database warnings worth reading once.
  • Two migrations will deliberately stop your deploy if they find pre-existing duplicate or conflicting custom-field definitions, rather than silently pick a winner among rows that must be unique. Preflight scripts for both ship with the release — run them before you pull. Both hosted regions were clean.
  • No breaking API changes and no new required environment variables, but five defaults change on upgrade: offline work queues instead of failing, 'Remove device' uninstalls the agent, network monitor alerts notify and automate, QuickBooks connections start pushing payments into connected realms, and on Linux and macOS stopping the agent service is temporary while the watchdog is installed.
  • Forced MFA for Partner Admins is reconciled but not enforced this release, so nobody is locked into enrolment at upgrade time. A version with a grace window, banner, and deadline ships next.
  • Everything AI stays off until you turn it on — the AI agent subsystem including the new builder, and both AI Operator flags. The removed-device purge policy ships off too.

Self-hosting? Install v0.111.1 instead of this version — it contains everything below plus a fix for an upgrade-blocking migration issue.

Breeze v0.111.0 is mostly about work that used to just fail. Point a patch job or a nightly automation at a fleet of laptops and, until now, every machine that happened to be asleep came back Failed. Now that work waits: it is held with a delivery deadline, the step reads “Queued — device offline”, and the agent picks it up the moment it checks in. You can see what is queued on the device page and cancel anything you have changed your mind about, and each automation action has a Queue-or-Skip switch for steps that only make sense against a live machine. The other half of “things you could not do before” is stopping. Scripts and automation runs have a Stop button now — a real one, that terminates the script and everything it spawned, with Force stop for a script that is ignoring the polite request, and an honest answer when the stop lands after the script already finished.

Every customer also has a page now: contacts, sites, devices, tickets, contracts and billing, and activity in one organization record, cross-linked from tickets and invoices, with a Service Desk section you can switch on, off, or point at an external PSA for your whole partner account. Alongside it, devices you cannot put an agent on are first class — hand-entered assets, manually added network devices, and websites you want monitored, all in one device list. Configuration policies can inherit from a partner-wide baseline, so a change to your standard belongs in one place instead of thirty. And the AI agent builder is a guided flow that spells out exactly what an agent is allowed to do on its own before you create it. A 62-screen browser sweep before the tag found and fixed five functional defects and four paper cuts, and the agent service install was checked on real Windows Server and Ubuntu hosts.

For self-hosters this is a standard upgrade with a caveat worth five minutes: 35 migrations, no breaking API changes, and no new required environment variables — but five defaults change on upgrade, and two migrations will deliberately stop your deploy if they find conflicting custom-field definitions rather than guess which row wins. Preflight scripts ship for both; run them before you pull. Full operator detail, including the behaviour changes one at a time, is in the GitHub Release notes.