github LanternOps/breeze v0.130.0

34 minutes ago

Breeze RMM v0.130.0: a guided ticket import wizard, reattaching a reinstalled computer to its device record, Microsoft Entra SCIM for the customer portal, a round of remote desktop fixes, Try again and browser notifications in AI chat, and hardening across AI tools, sign-in and the agent.

No breaking changes. Read Behaviour changes under Self-Hosting / Upgrade Notes before you upgrade: several defaults changed, and some fixes need the 0.130.0 agent.

Summary

  • Ticket import wizard (Zammad) (#8313: #11949, #11951, #11756, #11730). Partner-wide admins can run a whole cutover at Tickets → Import tickets (/tickets/import): connect a source, run preflight, map organizations, groups, priorities and states, do a dry run, open a migration window, then run the bulk import and a delta import.
    • Imported tickets carry an Imported chip, and the ticket list gets an Imported only filter.
    • Merged source tickets become "duplicate of" links when both tickets land in the same organization.
    • Create missing requester contacts is a per-source option, off by default. Contacts it creates can be listed and removed as a set.
    • A very large ticket now fails on its own ("ticket too large") instead of exhausting memory.
    • v0.129.0 shipped the import API only, with no screen. This release adds the screen.
  • Reattach a reinstalled computer to its existing device record (#8318: #11762, #11886).
    • When Breeze flags a newly enrolled device as a probable reinstall, the Unassigned Devices assign dialog offers Reattach to , and the possible-replacement banner offers Reattach.
    • Both show the new install and the existing record side by side. They require a typed hostname and MFA step-up, then keep the old record's history and retire the new one.
    • The record shows a Reinstalled on badge, and the activity feed lists reinstall events.
    • TPM-backed identity is groundwork in this release and off by default; see Added.
  • Customer portal: Microsoft Entra SCIM provisioning (#11377: #11983, #11884, #11917). This completes the "directory-driven provisioning" item listed as still to come in v0.129.0.
    • An MSP can create a SCIM token under Organization → Settings → Portal → Sign-in & registration. Entra then creates, updates and removes portal accounts through SCIM 2.0.
    • Directory deprovisioning (people disabled or deleted in Entra or Google Workspace) adds a panel with a preview, per-user badges, and Restore / Remove sign-in actions. It is report-only in this release: runs are recorded and nobody is disabled until a server sets PORTAL_DIRECTORY_DEPROVISION_ENFORCE=true.
  • Remote desktop fixes (#12194, #11325, #11916, #12006, #12074, #12075, #12086, #12209, #12191, #12151). Most of these need the 0.130.0 agent and the new Breeze Viewer.
    • The stream follows a display resolution change mid-session on Windows, macOS and Linux X11, and clicks land where they should.
    • The clipboard pastes whichever side changed last. A remote change that happens outside 2 seconds of your own input is held behind a Remote clipboard changed badge instead of overwriting yours.
    • Session history says why each session ended (ended by technician, peer disconnected, capture failure and others). A viewer hang-up is recorded as ended by the operator.
    • A network that blocks the connection no longer causes endless retries. The viewer stops after one retry and says which side's network is likely blocking it.
    • Sessions no longer drop after about 20 seconds on devices whose clock runs fast, or outlive their limit on devices whose clock runs slow.
    • The built-in H.264 encoder sharpens again after a burst of change on a static screen (#11639, listed as a known issue in v0.129.0).
    • Connect opens one window when Breeze Viewer is already running, and closing the viewer window always ends the session.
  • AI chat: Try again and browser notifications (#9428: #11985, #9429: #12050, #11912).
    • A failed turn (timeout, dropped connection, provider error) offers Try again in the same session; a timeout says "This response timed out".
    • The bell in the chat header turns on browser notifications for when a reply is ready, fails or needs your approval while you are on another tab.
    • Pressing Stop during a streaming reply ends the turn cleanly.

Added

  • Wake-on-LAN management, first release, read-only (#10838: #11996, #12163, #12219, #12251). A new Wake-on-LAN tab in Configuration Policies can limit Wake buttons to devices verified ready, and a status panel shows wake readiness across devices. Dell PCs report their BIOS Wake-on-LAN and deep-sleep settings. Nothing changes BIOS or NIC settings yet; applying them is groundwork in this release, compiled off (#12223). Readiness reporting needs the 0.130.0 agent. Devices with no Wake-on-LAN policy are unchanged.
  • Network filter and certificate blocks on devices (#8868: #12051, #12077, #12090). When a web filter, captive portal or untrusted TLS interceptor answers instead of the Breeze server, the agent reports what answered. The device Info tab shows a notice, and the device list shows a Network filter or Certificate not trusted pill. After 3 hours without a report the notice reads "last reported"; it disappears after 30 days. Needs the 0.130.0 agent.
  • Connection health: sync staleness (#9127: #12049, #12069). EDR, QuickBooks / Xero and ticket mailbox connections that silently stop syncing show as Sync stalled. The TD SYNNEX nightly price file is watched too.
  • Devices list (#11121, #12068). There is an optional Disk Usage column with an All disks / OS disk only toggle that also applies to sorting and CSV export (the OS-disk view on Windows and Linux needs the 0.130.0 agent; older agents show a dash). There is also an optional, sortable Model column, and long serial numbers are truncated with the full value on hover.
  • Maintenance windows open on the day you choose (#9574: #12185). Weekly windows can open on any weekday, and monthly windows on any day of the month (short months use their last day). Existing windows keep Sunday and the 1st.
  • Tickets: Recently updated sort (#12067).
  • Wisp analysis sessions (Breeze Cloud only, #11844: #11984, #12000, #12039, #12076, #12101, #12105, #12110). Wisp can run a sandboxed analysis across a whole site or organization (up to 1,000 devices by default), return its findings and files to the conversation, and continue from an earlier run. Export permissions are frozen at launch. Chains stop at 8 continuations, 12 runs or $6.00 by default, and the grant's daily budget usually stops first. Analysis can export reliability, boot performance, hardware event, service check, alert, patch and change-log data. The run page shows the chain, goal and device scope.
  • Groundwork, off by default:
    • TPM-backed device identity (#8318: #11956, #11957). With TPM_ATTESTATION_ENABLED=true, agents with a TPM 2.0 prove their endorsement key once per process start, and the server records the baseline and audits any later mismatch. Off by default; nothing uses it yet.
    • Network isolation: offline release token and dead-man timer (#10088: #11991, #12001). Both are built inert. Leave NETWORK_ISOLATION_TOKEN_SIGNING_KEY unset. The Containment policy tab shows a dead-man select (default off).
    • Vendor updates and HP warranty: read-only observe steps for Surface, HP and Lenovo (#11065: #12227, #12296, #12301; #5511: #12250). They report what a device would be offered, or whether warranty-lookup prerequisites are present, without installing, flashing or rebooting. They do nothing unless a server sets BREEZE_VENDOR_CANARY_PARTNER_IDS. The vendor-updates observe step needs the 0.131.0 agent, so it does not run on 0.130.0 agents. Dell gates stay closed (#12262, #12261).
    • Restore point history: an "Interrupted" status (#11981). The status exists on the server; the agent starts reporting it in a later release.
  • Metrics: breeze_email_send_total, breeze_email_provider_monthly_used and breeze_email_provider_monthly_used_updated_timestamp_seconds for outbound email (#11556). The production Caddyfile exposes the worker role's metrics at /metrics/worker, behind the same scrape token as the API (#11558).

Improved

  • Report PDFs (#12087, #12066, #12073, #12063, #12072, #12088, #12146). Report PDFs share one page size, date format and letterhead, taken from partner settings, and Branding warns about low-resolution logos.
    • Security Posture figures agree with each other, and a device with no data counts as not assessed.
    • The Executive Summary accounts for every device and scores devices seen in the last 7 days.
    • Vulnerability Management lists the affected software for each CVE.
    • Threat Detection Review coverage tiles add up and count SentinelOne and EDR as well as Huntress.
    • Content flows onto new pages instead of being cut off or printing over the footer.
  • Document page size by region (#12159). Partners created from now on print quotes, invoices, certificates and reports on their region's page size (Letter for the US and Canada, A4 elsewhere) until they choose one. Billing → Documents adds a Region default option. Existing partners keep their stored size.
  • Cmd+K search (#11934, #12237) covers every customer you can reach again, labels each result with its customer and ranks the header-selected customer first. A This customer only toggle restores the filter. It also finds sidebar pages by name or synonym ("helpdesk" finds Tickets) and jumps to Partner Settings sections.
  • Quick Support (#11887). The session code stays visible while the session is pending, including after a page reload, and New code replaces an expired one.
  • Windows Add Device (#10329). The install command is a short one-liner. It downloads a server-generated PowerShell installer (GET /agents/install.ps1) that verifies the agent binary's signature and checksum before enrolling. Old saved commands keep working.
  • Settings → Plan (#12055) shows whether device enrollment is paused and why.
  • Navigation (#12056, #12057). The collapsed sidebar rail is navigable: section icons open a flyout, and tablets get an expand toggle ([). The mobile nav drawer is a keyboard-accessible modal, and phones show the current organization's name under the header.
  • Software (#12070, #12071). The winget picker shows when its index was last updated and warns after 48 hours. The Software Library gains a Source filter. The Software Library and Package Catalog pages say what they do and do not control.
  • AI assistant and MCP tools (#12195, #12030, #11874, #12181).
    • Org documents, deliverable templates, the vulnerability report, user risk detail and custom fields page their results, so large tenants get complete answers.
    • get_ai_agent_run returns each executed step's bounded result.
    • A context window learned from a provider rejection is kept as a floor.
    • A gateway model whose verification record cannot be read says so instead of looking unverified.
  • Wisp (Breeze Cloud only, #11476, #11982, #11963). The docked pane header is two rows. Specialist step approvals appear in the chat, and the bell rings for them. The Wisp page composer can flag a conversation.
  • Expired-snapshot cleanup (#12170) runs on its own bounded queue and resumes where it stopped, so long cleanups no longer delay backup dispatch.
  • Breeze Assist (#12162, #11933). The Windows installer's desktop shortcut, Start entry and Uninstall entry, and the in-app titles and tray tooltip, say Breeze Assist. Install paths and registry keys are unchanged.
  • Agent: idle agents use about 515 KiB less memory (#12144).
  • Multi-instance deployments (#12202). Changes to partner IP allowlists, MCP token scope policy and six other cached settings apply on every API and worker instance at once.
  • Postgres 17 and 18 are supported for self-hosted installs (#11416: #12224, #11897). See Self-Hosting / Upgrade Notes.
  • Mobile: the store version is 1.0.6 (#12315).

Fixed

  • Backup:
    • Integrity verification and test restore of large snapshots no longer end partial or time out at 30 minutes. The budget scales with snapshot size (2 to 12 hours) and is counted from delivery. When the storage session ends, unreached files count as unchecked instead of failing (#12259, #12260, #11869; 0.130.0 agent).
    • Vault sync copies every object an incremental snapshot needs, and reads the destination the snapshot was actually written to. A sync that cannot copy everything, times out or is never delivered now fails instead of staying pending (#11927, #11904, #11925; 0.130.0 agent).
    • Windows file backups skip OneDrive online-only files (named in the job warning) instead of downloading them, and back up fully local OneDrive and Known Folder Move files (#12204; 0.130.0 agent).
    • A queued run is re-sent to a new helper session when the old one ends, Windows devices stay awake during a backup, and a late keepalive no longer closes helper sessions after a device sleeps (#12205, #12192; 0.130.0 agent).
    • Backups that were running during a server or database outage are no longer failed as stale. Stale provider devices no longer show green days or "Completed"; they show No recent backup (#11900, #12065).
    • Windows restores of sparse files recreate their holes (#12190; 0.130.0 agent). Snapshot index counts are no longer doubled (#11899). Platform S3 clients tolerate an unparseable abort-date header (#11910). The Dropsuite Australian API host is accepted (#11866).
  • Agent:
    • The Windows agent uninstall runs the MSI uninstall and reports completion only after a verified teardown (#11349).
    • On macOS, the desktop helper restarts itself once when a granted permission still reads as denied (#11868).
    • Disk usage on immutable Linux (Silverblue, CoreOS, bootc and similar) measures real storage (#11892).
    • An intermittent crash after the video encoder loaded is fixed (#12120).
    • Remote desktop no longer opens a multicast (mDNS) socket (#12247).
    • Event log lists show real record counts and keep their log name when truncated (#12029, #12154).
    • macOS IOKit loader noise no longer counts as hardware errors (#11876).
    • All of these need the 0.130.0 agent.
  • Compliance (#12028, #12184, #12239). Configuration-policy rules that check registry or config values now collect those values. A rule the device is not probed for is Not assessed instead of non-compliant. Removed or renamed rules no longer show on the device tab.
  • Commands (#11906, #11946, #11948). Single, bulk and auto-update commands are committed before they are sent, so fast agent results are no longer lost.
  • Sign-in and SSO (#11893, #12281, #11867). Invited and SSO users no longer hit "MFA required" when the policy does not require MFA. Organization SSO sessions see partner-wide policies, backup profiles and catalog items straight away. Promoting agent versions no longer fails with an invalid-signature 409 for partner- or org-scoped sessions.
  • Microsoft 365 (#12182, #12167, #12206). Admin actions connections can reach Active when the permission inventory read is denied, and Retest reports a degraded result honestly. Delegated mail consent works again after the MSAL 7 upgrade.
  • Tickets and email (#12203, #12212, #12248, #12245, #11962, #11959).
    • A requester who writes again about a new issue gets an autoresponse.
    • Microsoft 365 autoresponses keep their template subject.
    • A reply to an email linked as an internal note no longer posts publicly.
    • Mailgun mail reaches cc recipients.
    • j/k on the ticket list and Confirm in dialogs no longer drop the first key or click.
  • Trust Center (#12078, #12080, #12081, #12089, #12104). A newly published page no longer briefly answers 404. "MFA enforced (Microsoft 365)" needs a live connection. The editor detects changes made while you were editing. Publishing says when the public page takes up to a minute to update.
  • AI approvals and runs (#12035, #12083, #12096, #12099, #12147, #12179, #12037, #12038, #12003, #11907, #12036). Script-proposal approvals name the script and device, and a timed-out card says expired. An approval granted just before a deadline is honoured. Concurrent tool calls in one chat no longer mix results. Agent run sessions show their tokens, cost and transcript.
  • Integrations (#12108, #12062). The Test buttons no longer save settings. Discord sends a real test message, and Slack and Teams no longer report a fake success.
  • Disaster recovery: plans built in the editor no longer halt with "unsupported command type" (#11894).
  • Devices and views: partner users with several organizations can save a device view again (#11891). Software presence monitor alerts show the application name instead of {{name}} (#12253).
  • Logs: Log Search reopens saved searches (#11909).
  • Vulnerabilities: the severity filter matches feed-imported rows (#11937).
  • Discovery: alert settings entered at profile creation are saved (#11936).
  • Extensions: extension pages can call their own API again (#11938).
  • Administration: administration panels follow the dark theme (#11908). Platform-admin partner tools, tenant export and tenant erasure work across partners for a platform admin whose login also belongs to a partner (#12303).
  • Customer portal: the network overview excludes dismissed assets (#11890). The self-registration email no longer mentions Microsoft sign-in when it is not in use (#11855).
  • Reports: an included-by-rule recipient badge matches who is mailed (#11061).
  • Network proxy tunnels: a burst of requests queues briefly instead of being refused (#11872).
  • Read-only database episodes: logs are throttled, startup writes wait until the database is writable, and the GitHub binary sync retries (#11903, #11935, #11871).
  • Database-only restore: startup recreates missing audit-admin and search roles (#11940).
  • Postgres 18: "in use" refusals show the friendly error (#11897).
  • Wisp (Breeze Cloud only): #11675 / #11918, #11673 / #11914, #11513 / #11913, #11612 / #11911, #11967 / #12026, #11973 / #12024, #11978 / #12027, #12137 / #12158, #11678 / #12186.

Security

This release tightens checks across AI tools and MCP, sign-in and MFA policy handling, device notes, report and device-page access scoping, operator alerts, the Linux agent's helper channel and the API image. Self-hosters are encouraged to upgrade. The changes integrators may notice are listed under Behaviour changes. (#11774, #11995, #11993, #11930, #12152, #12188, #12153, #12214, #12168, #11729, #11779, #12196, #11865, #12180, #12040, #12164, #12211, #11717)

  • AI tool approvals: the highest-risk tool tier is refused instead of being offered for approval, and the approval method is recorded on every call (#11995). Every AI agent session runs with a fixed permission mode (#11993).
  • Device notes written over MCP, through unattended grants or by flow steps are bounded: 200 characters, 30-day expiry, 8 active per device (#12152, #12188).
  • Linux agent: helpers verify they are connected to the root agent, and helper connections are size-capped before authentication (#12168).
  • Breeze Assist: security prompts (consent, session banner, caller verification) no longer take partner brand colours (#11779).
  • API image ships production dependencies only: about 250 fewer packages and about 240 MB smaller (#12196).
  • Dependencies: patched helper crates and one replaced date/time library (#11865, #12180, #11747).
  • Partner trust and abuse handling hardening (Breeze Cloud; #12040, #12164, #12211, #11717).

Self-Hosting / Upgrade Notes

Upgrade command

Check how your install pins its images: grep '^BREEZE_API_IMAGE_REF=' .env.

  • Digest-pinned (the value ends in @sha256:…, which is every guided-setup.sh install from v0.112.0): fetch the current guided-setup.sh from main, then run bash guided-setup.sh --upgrade 0.130.0. It verifies the signed image inventory, updates BREEZE_VERSION and all four image digests together, and copies the verified release manifest pair into the binaries volume (#11896, #11915). Then it pulls and restarts.
  • Tag-form (the value ends in :${BREEZE_VERSION}): set BREEZE_VERSION=0.130.0, then run docker compose pull api web portal && docker compose up -d (and pnpm install if you build from source). If you run the optional worker container, pull and restart it too.
  • Do not edit only BREEZE_VERSION on a digest-pinned install. That re-pulls the images you already run.
  • Update the API, web, portal and worker containers together. Every API instance must run 0.130.0 before an MSP disables or reactivates a portal user. After the migration, an old-build API fails those two actions (#11884). The ticket import wizard, the SCIM panel, the reattach dialog and the Wake-on-LAN tab call routes the old API does not have.
  • Breeze Viewer and agents. Several remote desktop fixes need the new Breeze Viewer as well as the 0.130.0 agent (#12151, #12074, #12006, #12191, #12194). Update agents in stages. With AGENT_AUTO_PROMOTE=false, a new agent release is registered but not promoted.
  • --upgrade does not change your docker-compose.yml. To pick up this release's Compose changes (see Environment variables and PostgreSQL), take the v0.130.0 docker-compose.yml / deploy/docker-compose.prod.yml, or add the lines to the api service's environment: block yourself. A value that is only in .env does not reach the container.
  • BINARY_SOURCE=local without guided-setup.sh (for example an air-gapped install): stage release-artifact-manifest.json and its .ed25519 signature at the root of the binaries volume. The API now names the missing file and links the docs when it is absent (#11896). This was already required; 0.130.0 adds the error message and the docs, not the requirement.
  • See the Upgrade Guide.

Breaking changes

  • No breaking changes. Nothing is removed on an install with default settings. Read Behaviour changes for the defaults that changed.

Database

  • 25 migrations. They are idempotent and apply automatically on first API boot via autoMigrate, unless AUTO_MIGRATE=false. Expect [auto-migrate] Applied 25 migration(s) from v0.129.0. No shipped migration was modified, renamed or removed. None of them needs superuser or extra privileges.
  • First boot: six indexes are built concurrently, so on a large install the first boot can take minutes and needs free disk for the new indexes. Do not restart the API mid-migration. If a later start stops with INVALID index(es): <name>, run DROP INDEX CONCURRENTLY public.<name>; and start again.
  • Large-table work:
    • Index builds with CREATE INDEX CONCURRENTLY (reads and writes continue; each build reads its table twice): agent_health_observations and software_inventory_observations (2027-01-25-110000-observation-retention-indexes, #11744); device_boot_metrics, device_hardware_events and spc_results (2027-01-27-140000-export-dataset-reliability-indexes, #12076); alerts (2027-01-27-150000-export-dataset-alert-index, #12101). On an install with a long history, the first boot can take from seconds to a few minutes per table. If a build is interrupted, the next boot stops and names the INVALID index: run DROP INDEX CONCURRENTLY public.<name> and start the API again.
    • Plain index builds on a possibly large table: 2027-01-27-130000-ai-analysis-run-lineage builds two partial indexes on ai_agent_runs without CONCURRENTLY and holds an exclusive lock on ai_agent_runs for the whole file, with no lock timeout. On a large ai_agent_runs table, upgrade at a quiet time (#12039).
    • Rows rewritten: 2027-01-24-230000-vulnerabilities-severity-lowercase lowercases severity on the global CVE catalog in one set-based UPDATE and logs the row count (#11937). 2027-01-24-200000-portal-directory-deprovisioning sets disabled_by = 'msp' on already-disabled portal users (#11884).
    • Constraint validation scans (do not block writes): remote_sessions end-reason CHECK (2027-01-24-220000, #11916), and four CHECKs on device_hardware (2027-01-28-190000-device-hardware-tpm-baseline, #11956).
    • Short locks under a 5-second lock timeout: device_disks (2027-01-24-190000) and partners (2027-01-27-120000). If a long transaction holds a conflicting lock, the migration fails, the API exits, and the file runs cleanly on the next start.
  • New tables: portal_scim_tokens (#11983), portal_directory_runs and portal_directory_actions (#11884), and partner_billing_fraud_flags (#12211).
  • Other changes: nullable columns, columns with a constant default, CHECK changes and one enum value (config_feature_type gains wake_on_lan). They touch device_disks, portal_users, support_sessions, wisp_check_ins, ticket_links, device_restore_point_attempts, partners, device_network_isolations, devices, ai_agent_runs, ai_run_artifacts, m365 action connections, config_policy_maintenance_settings, device_hardware and agent_versions. partners.document_page_size becomes nullable with no default (#12159).
  • Rollback. Take a backup, or note a PITR point, before you upgrade. Rollback to v0.129.0 was not tested.
    • Known break, per #11884's PR (not tested): a v0.129.0 API on the upgraded database fails MSP portal-user disable and reactivate.
    • Rollback beyond that break was not tested. Rolling forward is the supported path; restore the pre-upgrade backup if you must go back.
  • Action needed in one case: an install already on Postgres 18 with the upstream volume layout must keep its own PGDATA, or Postgres refuses to start. See PostgreSQL.

PostgreSQL

  • PostgreSQL 17 and 18 are now tested and supported; the default stays 16 (#11416: #12224).
    • The postgres service in docker-compose.yml now sets PGDATA: /var/lib/postgresql/data, which is the 16/17 default. Existing volumes stay where they are, and the container restarts once.
    • It also gains an optional second data volume, postgres_data_alt, selected by POSTGRES_DATA_VOLUME. Only postgres_data (the default) and postgres_data_alt are accepted. The selector is used only by the new PostgreSQL 16 to 18 runbook.
    • A new existing-data guard refuses to start Postgres on an empty PGDATA when a database sits next to it. It logs the PGDATA to set instead of creating an empty database.
  • Already running 18 with your volume mounted at /var/lib/postgresql: check SHOW data_directory and keep PGDATA on /var/lib/postgresql/18/docker. If you take the release's postgres block with that mount, Postgres refuses to start and names the PGDATA to set. Merging only the PGDATA line gives you no such check.
  • Do not change the major version by editing POSTGRES_IMAGE_REF on an existing install. Follow the runbook, which restores into the second volume and keeps the old one as the rollback. A new install may set a reviewed pg17 or pg18 image before its first start.

Environment variables

  • Required: none. No new required environment variables. The config validator refuses to boot only when one of the optional variables below is set to a malformed value (TPM_ATTESTATION_ENABLED, NETWORK_ISOLATION_TOKEN_SIGNING_KEY, BREEZE_VENDOR_CANARY_PARTNER_IDS, BREEZE_VENDOR_CANARY_DISABLED, PARTNER_TRUST_FRAUD_FLAGS_ENABLED).
  • New, optional, mapped in both Compose files:
    • PORTAL_DIRECTORY_DEPROVISION_ENFORCE (unset = report-only): set to true to let directory deprovisioning disable portal sign-ins. Any other value runs report-only and warns once at boot (#11884).
    • TPM_ATTESTATION_ENABLED (false): agent TPM endorsement-key attestation. Leave it off; nothing uses it yet (#11956).
    • NETWORK_ISOLATION_TOKEN_SIGNING_KEY (unset): not available yet; leave it unset (#11991).
    • BREEZE_VENDOR_CANARY_PARTNER_IDS (empty = off; at most 3 partner UUIDs) and BREEZE_VENDOR_CANARY_DISABLED (a feature:vendor kill switch): who may run the read-only vendor observe steps. Leave them unset unless you are asked to canary (#12227).
    • PARTNER_TRUST_FRAUD_FLAGS_ENABLED (false): a Breeze Cloud billing integration. Self-hosters leave it unset (#12211).
    • Three IP_CLASSIFY_*_URL overrides (unset = built-in defaults): partner trust checks use them, and those checks are off on a default self-hosted install. Leave them unset (#11717). The variables are IP_CLASSIFY_TOR_LIST_URL, IP_CLASSIFY_HOSTING_ASN_URL and IP_CLASSIFY_ASN_DB_URL.
    • TICKET_IMPORT_MAX_TICKET_ARTICLES (10000) and TICKET_IMPORT_MAX_TICKET_READ_BYTES (256 MiB): per-ticket read limits for ticket import (#11730).
    • BACKUP_SNAPSHOT_CLEANUP_CRON (27 2,8,14,20 * * * UTC; off stops it) and BACKUP_SNAPSHOT_CLEANUP_TIME_BUDGET_MS (4 h): expired-snapshot cleanup schedule and budget (#12170).
    • POSTGRES_DATA_VOLUME (unset = postgres_data): docker-compose.yml only. See PostgreSQL (#12224).
    • Retention, report-only: AGENT_HEALTH_OBSERVATION_RETENTION_* and SOFTWARE_INVENTORY_OBSERVATION_RETENTION_*, each with _DAYS, _BATCH_SIZE, _MAX_BATCHES and _ENFORCE (default false; an hourly job, #11744). There are also ten report-only _DAYS windows with no _ENFORCE (nothing is deleted in any configuration; #11928): SSO_SESSION_, C2C_CONSENT_SESSION_, TUNNEL_SESSION_, UNIFI_SYNC_RUN_, FLEET_REMEDIATION_RUN_, TICKET_IMPORT_RUN_, TOPOLOGY_DIAGNOSTIC_RUN_, AUDIT_BASELINE_RESULT_, PERIPHERAL_POLICY_DELIVERY_EVENT_ and ML_FEEDBACK_EVENT_RETENTION_DAYS. A new docs page covers reclaiming disk with pg_repack once a job is enforced (#11955).
  • Now mapped, so a value in .env takes effect: BACKUP_GC_LIVE_SET_MAX_BYTES and BACKUP_GC_MAX_DELETES_PER_RUN. Before this release, setting them in .env had no effect (#11924).
  • Compose: docker-compose.yml and deploy/docker-compose.prod.yml add the mappings above to the shared api environment block. docker-compose.yml also changes the postgres service (PGDATA, entrypoint guard, volume selector) and adds the postgres_data_alt volume. No mapping was removed. docker/Caddyfile.prod adds /metrics/worker (#11558).

Behaviour changes

  • Cmd+K searches every customer by default and ranks the header-selected customer first. This customer only restores the old filter (#11934).
  • AI tools refuse arguments they do not define, naming the unknown key, instead of ignoring them. An MCP or API integration that sends extra or misspelled keys gets an error (#11875).
  • MCP and API-key callers with ENABLE_2FA on get MFA_REQUIRED for MFA-gated actions, and fully gated tools leave tools/list (#11774). The highest-risk AI tool tier is refused (#11995).
  • Bounded AI device notes and run-mode check-ins no longer raise approval cards. Actions a check-in proposes still need approval (#11987).
  • Integration Test stores nothing, and tests are rate limited (429 with Retry-After past 5 per minute per user or 10 per minute per organization). Discord accepts only discord.com webhook URLs (#12108, #12062).
  • Reports:
    • The Executive Summary's headline health score is the share of devices seen in the last 7 days (#12073).
    • A Security Posture GOOD or STRONG score is held at FAIR while critical patches or vulnerabilities are open or a device is unprotected (#12066).
    • Report PDFs follow the partner's document page size and date format (#12087).
    • Devices with no data are "not assessed".
  • Tickets:
    • The autoresponse cap is 5 per sender per partner per hour; it was 1 per sender per day, shared across partners (#12203).
    • Reply-header threading matches only public links from inbound mail and the Outlook add-in (#12248).
    • With EMAIL_PROVIDER=mailgun, cc recipients start to receive mail (#12245).
  • Compliance:
    • Registry and config checks the device is not probed for store not_collected (Not assessed) and raise no alert or remediation (#12184).
    • The heartbeat probe list now includes active configuration-policy compliance rules (#12028).
  • Backup:
    • Online-only OneDrive placeholders are skipped with a job warning (#12204).
    • Verification results carry filesUnchecked and a storage_session_lost reason code (#12260).
    • Backup status views add an overdue bucket (#12065).
  • Windows agent uninstall reports its result after the teardown finishes, not when it starts (#11349).
  • macOS agents below a release's minimum macOS (currently 12.0) stay on their current agent, and the pkg refuses to install below the floor. From v0.131.0, macOS agent assets without a signed minimum-OS value are refused at sync. The self-host signing template already includes it (#12258).
  • Network proxy tunnels queue for up to 5 seconds (96 waiters), then return 429 with Retry-After (#11872).
  • BINARY_SOURCE=github: boot sync retries transient network failures in the background for about 15.5 minutes, then logs the manual sync-github command. During a read-only database episode, /health/ready reports binarySyncPending: true until the sync can run (#11871, #11935).
  • Database-only restores: boot creates breeze_audit_admin and breeze_search if missing, and restores their membership when AUDIT_ADMIN_DATABASE_URL is unset. On managed Postgres without the privilege, it logs the GRANT to run (#11940).
  • Portal:
    • Reactivating a portal user who has no usable credential answers 409 PORTAL_ACCOUNT_RESTORE_SIGN_IN_FIRST (#11884).
    • Network overview figures drop for organizations with dismissed assets (#11890).
  • Vulnerabilities: stored severities are lowercase (#11937).
  • New partners get their region's document page size (#12159).
  • API changes to check if you integrate:
    • GET /tickets accepts imported=true and sort updated (#11949, #12067).
    • POST /devices/:id/reattach and GET /devices/:id/reinstall-candidates; device responses carry lastReattachedAt (#11762).
    • SCIM 2.0 endpoints for the portal (#11983). /organizations/:id/portal-directory routes (#11884).
    • POST /remote/support-sessions/:id/code. Pending support sessions return code and landingUrl to callers who may create Quick Support sessions (#11887).
    • GET /agents/install.ps1 (public) (#10329).
    • POST /agents/:id/attestation/challenge and /complete answer 403 TPM_ATTESTATION_DISABLED while off (#11956).
    • POST /backup/vault/:id/sync answers 422 when the snapshot's destination does not resolve (#11904).
    • Event-log responses omit recordCount when it is unknown (#12029).
    • POST /orgs/:orgId/trust-page/publish and /unpublish return serving (#12104).
    • GET /software/package-search returns indexUpdatedAt (#12070).
    • GET /reports/:id/recipients adds ruleContactIds (#11061).
    • Chat error events carry a code, and done carries stopped (#11985, #11912).
    • The AI and MCP tools in #12195 return paged results.
    • Audit details gain delivery on auto-update and bulk command entries (#12160).
  • Wisp (Breeze Cloud only, BREEZE_WISP_ENABLED): no default changed.
  • No flag in this release has to be turned on.

Known gaps

These were not verified on real hardware or accounts before this release:

  • macOS: the Apple Silicon upgrade-in-place leg was not run (Intel passed with no permission prompt). Remote desktop double-click (#11670, #11549), Breeze Assist labels in Portuguese (#11204), the HiDPI half of mid-session resolution changes (#12226, needs a 4K display), and the helper's self-restart on a stale Screen Recording grant (#11868, needs a manual revoke). The Local Network prompt on update is still open (#11311, #12225); #12247 removes one trigger.
  • Windows: self-uninstall (#11349), sign-in screen capture (#9583), YARA rule-cache exclusion (#11728), cloud placeholder files with a Microsoft account (#12204), and the agent half of several remote desktop and Breeze Viewer fixes (#12151, #12074, #12006). Viewer clipboard (#11325) passed most cases; one case is intermittent.
  • Hardware vendors: vendor updates on HP, Lenovo and Surface (#12250, #12251, #12227). Vendor-updates observe needs the 0.131.0 agent, although 0.130.0 agents carry the handler (OEM_OBSERVE_MIN_AGENT_VERSION).
  • Linux: Wayland capture (#12207, #12189, #12168) and read-only root filesystems (#11892).
  • Built-in H.264 encoder fallback (#11639): #12209 landed; picture quality after a burst on a static screen is unconfirmed.
  • External accounts: Entra SCIM (#11983), directory deprovisioning against live Entra or Google directories (#11884), Zammad import against a live instance (#11949), TD SYNNEX price-file staleness (#12069) and Microsoft 365 Admin actions (#12182).
  • Portal Sign in with Google (#11384) ships switched off.
  • Postgres 16 → 18 runbook (#12224) was not run on a real self-hosted volume.

What's Changed

New Contributors

Full Changelog: v0.129.0...v0.130.0

Don't miss a new breeze release

NewReleases is sending notifications on new releases.