Skip to content
Live from GitHub Releases

Release notes

Every release of RivetIT, RivetMSP and the Android apps, straight from GitHub. This page updates itself when a new release is published.

RivetIT 26.10.88

v26.10.88
Release notes

Onboarding/offboarding as tickets (switch off by default; portal New hire / Leaver items), worksheet on ticket templates, Clone for ticket/workflow/worksheet templates and automation rules, Workflows list page, New from workflow on Tickets, Employee Workflow editor redesign, blank-page fix when adding a preset. Needs database update 2.6.236 (Administration > Update; set a backup passphrase first). See CHANGELOG.md.

RivetIT 26.10.87

v26.10.87
Release notes

Rolls up 26.10.86 and 26.10.87: ticket list/status redesign, Counts-as-closed ticket status flag, ticket time zone fix and DB repair (2.6.234/235), credentials TOTP widget, asset details card fix, new asset types (Server Room, Rack, UPS/Power, Storage/NAS), Employee Workflow editor redesign. Database updates run from Administration > Update. See CHANGELOG.md.

RivetIT 26.10.85

v26.10.85
Release notes

Requires RivetCore 1.0.0-rc.60 (CSS/JS and agent fixes; its migration 0049, the "managed by another RMM" installer column, is not used by RivetIT yet and is applied by the database step that adopts the feature).

  • No underline on hover, anywhere. One last rule in the shared design CSS overrides Tabler's hover underline and every per-page copy of it (asset names, interface links, report, training and webhook links, kanban task titles).
  • Assets list: no sliver above the sticky table header. Collapsed borders painted the sticky header about a pixel low, so scrolled rows showed through above it.
  • RMM Assets: advanced filters. Search, integration, status, department (including "none assigned"), last seen (within an hour to 30 days, or not seen for 7 or 30+ days), operating system, manufacturer and model, remote agent, asset record and rows per page; health checks (needs reboot, patches pending, maintenance mode, disk, CPU or memory 85% or more); removable filter chips, sortable columns, paging, status pills, disk-usage bars and reboot/patch badges, tiles for reboot and patches.
  • Fix: "N of N could not be assigned" on RMM Assets. Assigning a department (single or bulk) failed with "No integration configured" when no default RMM integration was set; it needs none. The dialog now also says why a bulk assignment failed.
  • Hypervisor host page redesigned (Proxmox/Hyper-V): at-a-glance tiles, a clear banner when the collector reports a problem or the host is stale, aligned cluster/storage/VM cards with usage bars, the VM filter in the card header. A Proxmox host with ceph-common but no Ceph cluster no longer shows "ceph: /usr/bin/ceph exited with status 1" (server drops it from reports of older agents; new agents do not send it).
  • API: POST /api/v1/tickets accepts asset_id (must belong to client_id).
  • Alerts: the Critical + error open tile is highlighted only when both its filters are active. Asset edit modal: the Details grid shows only while its tab is active.
  • Tickets: a ticket in a Resolved or Closed status is no longer counted or listed as unresolved. The open/closed split also looks at the status name, so tickets moved to Resolved by the API, the Kanban board or an older version leave the Unresolved tile, the Open count and the open list.
  • Ticket status pills use a light tint of the status colour with a darker shade for the text (deep tint, light text in dark mode), so neon, near-black and pastel status colours stay readable.
  • Empty ticket lists say why. The empty state names the applied filters (including a saved default view) and offers "Show all open tickets", "Include closed" or "Clear filters".

RivetIT 26.10.84

v26.10.84
Release notes
  • Production branch renamed. The production release channel now follows the git branch production (it was main). New installs and the setup script use it. A server already on the Production channel needs one step after this release: git fetch origin && git branch -m main production && git branch -u origin/production production. Servers on Beta are not affected.
  • Settings. Tiles for Compliance framework, Security and trust, and Reporting / Power BI on the settings hub.
  • Telemetry page removed. RivetIT sends nothing anywhere; the Settings > Telemetry page and its installation-ID display are gone (the ID is still used internally to sign calendar-feed links).
  • Funding. .github/FUNDING.yml (GitHub Sponsors).

RivetIT 26.10.83

v26.10.83
Release notes

Release candidate 2 on top of 26.10.81: issue #29 leftovers, #31/#33 lifecycle follow-ups, #36 Teams/Slack buttons, #30 risk score display, plus test, accessibility and API-doc work. Database 2.6.231 to 2.6.233 (2 additive, idempotent steps; take a database dump first, then Administration > Update > Update Database).

Reversed decisions: in docs/FINDINGS-DECISIONS.md, item 11 (company-wide vendors stay Financial-only) is superseded by item 13, and item 12's "no UI moves a person to another department" row is superseded by item 14 (Move department).

Added

  • Move department (issue #29 leftovers): People row menu. Moves a person to another active department (needs Departments at Modify and access to both). Their devices and licence seats in the old department are unassigned by default (a box keeps them), the location is kept only if the new department can use it, the primary-contact flag is cleared, the move is audited, and employee.transferred is raised through the existing lifecycle detection when Employee workflows auto-start is on. Web only, no API equivalent.
  • Users export has an Include archived users box; archived agents are then listed with the status Archived.
  • HR requests from the employee portal (issue #31): managers and department administrators can ask HR to start an onboarding, offboarding or role-change workflow (portal menu HR requests, off until Administration > Settings > Employee portal > "Allow HR requests from the portal" is ticked, DB 2.6.232). The request waits for an HR agent (Service desk > HR requests, also GET/POST /api/v1/hr_requests); nothing starts and no record changes until it is approved, and rejecting or withdrawing starts nothing. Department scoped (a manager can name only their reports), audited, in-app notification to eligible HR agents. Approving picks the template and starts the run through the normal workflow code.
  • Okta workflow actions (issue #31): create a user (staged, no password, no activation email), suspend a user and clear their sessions (never delete), and add a user to groups, as workflow tasks. Off by default (Administration > Employee Workflow Templates > Okta actions; domain, encrypted API token and the "Allow workflows to change Okta" switch, DB 2.6.233). Department check, idempotent, token never in messages, logs or audit. Tested against a mock only; not verified against a real Okta org.
  • Legacy ticket rules in the Event rules tab (issue #33): every older ticket rule is listed read-only with its state; a rule that has an event-rule equivalent has Convert to event rule (a switched-off copy, the legacy rule keeps running) and then Switch over (copy on, legacy off, one confirmed step). Rules with no equivalent stay as they are, with the reason.
  • Workflow approval tasks reach approvers in Slack / Teams through the existing chat layer (ChatNotifier::approvalRequested); covered by a new end-to-end test. No new chat code.
  • Teams ticket buttons via a relay (issue #36). A Microsoft Teams webhook destination with an interactive-action token (Edit webhook > Generate) posts Acknowledge and Assign to me (Action.Submit) next to Open ticket. A relay flow you own forwards each click to /teams_interactive.php, authenticated with that per-destination token (constant-time compare, never in the card), a 5 minute timestamp and single use; the clicker is matched to exactly one active agent by email with the same role and department rights as Slack. No migration. Loopback-tested only, never a real Teams tenant. See docs/SLACK_TEAMS_SETUP.md.
  • Service catalog risk score is visible (issue #30). The admin-set risk score shows as a banded Risk n (Low / Medium / High) badge in the admin catalog list, the agent Request Something tiles, the ticket's Request details block, and the agent and portal approval inboxes. No migration.
  • One-click catalog request in chat (issue #36). With "actions from chat" on, catalog results from the Slack and Teams virtual agent show a Request this / Request button that raises the request for the linked user (same rules as request <id>: required questions go to the portal, approval items wait for approval). Tested with signed Slack simulators and a mocked Teams connector only.

Changed

  • Company-wide vendors (reverses the earlier decision that they stay Financial-only): a role with Tickets, assets & docs can list and open them (Read) and add, edit, archive and restore them (Modify). The account number, Delete and the vendor CSV stay Financial-only; without Financial the account number is hidden from the list, search, details and forms and is never changed by that role's edits.
  • Deleting an RMM check policy now asks the RMM to remove the checks it already pushed to agents. If any cannot be removed the policy is kept, those deployments are marked remove_failed, and Delete anyway (which writes the device names and check ids left on agents to the audit log) is offered.
  • The Waiting on Employee dashboard tile also counts open tickets in any status flagged "waiting on someone else" (pauses the SLA clock), not only the two names.
  • Mobile API: a catalog table answer now accepts JSON null for an empty cell (it was a 422). docs/MOBILE_API.md documents answering table questions with an example and the limits, and the new /hr_requests endpoint.

Fixed

  • Users CSV listed Department Portal logins; it is now agents only (what the Users list shows).
  • Bulk department email mailed the contacts of an archived department when it was selected.
  • Licence seat links: a forged asset or person id from another department was linked on add or edit; only the licence's own department can be linked now.
  • Departments date range: a custom range with only a From (or only a To) date showed ... - Dec 31, 2099 (or 1970) after the page reloaded; it now reads From Jan 15, 2026 / Up to ....
  • deploy/restore.sh refused a backup whose inner manifest is encrypted although it already requires --passphrase-file; it now decrypts the manifest with that file.

Tests, accessibility and API docs

  • API: api/v1/openapi.yaml now documents the endpoint agent's agent_changes, agent_logs, agent_transfer, agent_software and agent_linux_file, the local build host protocol /build_host/{action}, and the RivetPanel device API (panel_enroll, panel_me, panel_categories, panel_tickets, panel_ticket_chat), with the new panelAuth and buildHostAuth security schemes.
  • scripts/check_openapi_drift.sh forces LC_ALL=C so comm sees byte-sorted input (it reported false drift for hyphenated resource names under a UTF-8 locale); it now exits 0.
  • Accessibility: portal and announcement pages no longer skip heading levels; the amber "not verified" badge, amber device status text on portal cards on Chat settings, outline buttons in dark mode and urgent portal announcements in dark mode now meet WCAG AA contrast.
  • Tests: the browser suites take their scratch database from RIVETIT_TEST_DB_NAME / RIVETIT_TEST_DB_USER; the portal seed wipes safely on an install that already holds demo data; tests/reporting_views.php sets a UTC company first and can load time zone tables into a private throwaway MariaDB without Docker; tests/a11y_new_pages.js is a repeatable accessibility probe; tests/browser/lib (the CDP driver and harness the browser suite needs) is tracked again. Status of every top-level test is in docs/TEST_STATUS.md.

Notes

  • New tests: tests/i29_leftovers.php, tests/browser/i29_leftovers_seed.php, tests/browser/i29_leftovers_smoke.mjs. Findings status tables are final in docs/user-guide/tools/app-findings.md and docs/FINDINGS-STATUS.md; decisions 13 and 14 in docs/FINDINGS-DECISIONS.md supersede 11 and 12.
  • Database: 2.6.232 (hr_workflow_requests, settings.config_portal_hr_requests) and 2.6.233 (settings.config_lifecycle_okta_*). Everything new is off by default.

RivetIT 26.10.82

v26.10.82
Release notes
  • Every-page UI pass. Fixes and one user-facing improvement per area: collapsible Ticket Views rail, "Their other tickets" on the ticket page, similar-ticket hints in New Ticket, hidden-tab validation that opens the offending tab (about 95 dialogs), Health column on Departments and expiry chips on domains/certificates/software, project health chips, quote status pipeline, answer-first headlines on 12 reports, a device-status band on the asset page, training overdue banner and compliance matrix, admin jump bars on long settings pages (one shared look), grouped Ticket settings, status badges for SSO/mail/identity, retry-all on the job queue, and a 24-hour health banner on webhook deliveries.
  • Command palette. Ctrl/Cmd+K jumps to any page or record (RivetCore rc.58).
  • Portal. "How can we help?" box suggests knowledge-base articles and catalog requests before a ticket is raised; clearer no-access banner, ticket-detail summary, status tabs; tables stack as cards on phones.
  • Fixes. Table badges and buttons no longer break mid-word, mailbox and credentials tables fit at 1366 wide, readable amber text, org chart Location column, panel_deploy 500 without a department.
  • Requires RivetCore 1.0.0-rc.58 (CSS/JS only; no migration).

RivetIT 26.10.81

v26.10.81
Release notes

Integration release of the issue #30-#36, #43, #52 / #55 work, #37 (docs) and #29 (docs and tests). Database 2.6.211 to 2.6.231 (21 additive, idempotent steps; take a database dump first, then Administration > Update > Update Database, or php scripts/update_cli.php --update_db). Nothing risky is switched on by this release; several parts are verified against local mocks only and are marked UNVERIFIED LIVE.

  • Service catalog, issue #30: categories, computed risk and routing, parallel approvals, delegation and timeouts, repeating tables, catalog report (database 2.6.211 and 2.6.212). Everything is additive and off by default; existing catalog items behave exactly as before.

    • Category tiles. New Categories list on Administration > Service Catalog (name, description, icon, order); pick one per item. Request Something (agent and portal) shows a row of category tiles with request counts above the item tiles and filters when one is clicked. Items without a category are listed under Other.
    • Computed risk score and risk-based routing. An item's score is now a base plus risk rules (an answer equals / is one of / is answered / is at least / is at most a value: plus or minus points, kept between 0 and 100). The computed score and its breakdown are stored on the request and shown on the ticket page; it decides the auto-approve test and which approval steps apply (every step has a risk band, default 0 to 100), so a high-risk request can need an extra approver.
    • Parallel approval steps. Tick Parallel on a step to decide at the same time as the step above it; the next stage starts when every parallel step is satisfied. One rejection ends the request; one person in two parallel steps decides once.
    • Out of office, reminders, escalation, timeouts. Agents can hand their approvals to a colleague for a period (Approvals page). Each item can set a reminder interval, an escalation time (the active administrators are added and can decide the stage) and a timeout (the request is rejected and its ticket closed), in hours (0 = off). The new worker cron/catalog_approvals.php (required from cron/cron.php) does the work and stays off until an administrator ticks Approval worker on the Service Catalog admin page (settings.config_catalog_approval_cron).
    • Repeating table questions and multi-condition show/hide. A new Table question type (columns Heading|text|number|date, up to 8 columns and 25 rows, every cell validated on the server, shown as a table on the ticket and approval pages) and up to three conditions per question combined with all/any.
    • Report. Reports > Service Catalog Requests: request volume, approved / rejected / waiting, rejection rate and approval turnaround per item, category and day, department scoped, CSV export, saveable and schedulable.
    • API. GET /api/v1/service_catalog.php now also returns categories, each item's category_id, a table question's columns and compound show_if rules; POST is unchanged for existing clients and now runs through ServiceCatalogService::createRequest(), the entry point for chat integrations (issue #36). An older mobile app does not know the table question type. Deploy notes: take a database dump first, then run the database update. Two additive, idempotent steps (2.6.210 -> 2.6.211 -> 2.6.212). Reminders, escalation and timeouts only run after the Approval worker is switched on.
  • Employee lifecycle automation (issue #31)

    • Role-change and department-transfer workflow templates, with employee.role_changed and employee.transferred events and start rules.
    • The Microsoft and Google directory syncs now raise hire / departure / role-change / transfer events (only when "Start workflows automatically" is on).
    • Onboarding form (questions asked when a workflow starts, usable as {{var.name}}) and conditional tasks (department, job title, employee type, form answer); the dry run shows what is left out.
    • Delayed tasks ("archive the record 30 days after the end date"): scheduled status, run now / skip, released by cron/workflow_cron.php.
    • New actions: assign / collect devices, archive the record, Microsoft 365 licences, mailbox automatic reply, Odoo create / archive (off by default, own setting), identity-system connectors (stubs).
    • Bulk start for many people (email list / CSV, and from the People import); built-in presets and starter templates.
    • Database 2.6.213 and 2.6.214.
    • Unverified on real systems: Entra licence / mailbox actions, Odoo actions, identity connectors, directory-sync events (see docs/EMPLOYEE-LIFECYCLE-WORKFLOWS.md).
  • Employee portal (issue #32): announcements (new Settings > Portal announcements page; all departments or one; schedule, expiry, pin, plain text) shown at the top of the employee home and in a new notification feed with an unread badge.

  • Help articles (popular this month, new and updated) on the employee home; portal article reads are counted.

  • My devices shows a card per device with OS, last seen, warranty and restart-needed.

  • Employee API: /api/v1/portal_auth, portal_home, portal_devices, portal_checklist, portal_announcements, portal_notifications with their own hashed tokens (not usable on the staff API). Not yet verified against a live install.

  • The employee home no longer shows the stat tiles (department administrators keep the original page). Two new home sections, announcements and kb, are added to existing installs that had chosen their sections.

  • Database 2.6.215-2.6.218 (new tables portal_announcements, kb_article_views, portal_api_tokens, portal_feed_state; settings column default).

  • Automation engine, issue #33: more triggers, more actions, an API and a retryable run log (database 2.6.219 - 2.6.222). Nothing risky is switched on by this change.

    • New events for event rules: employee.created, employee.disabled (needs Employee workflows auto-start, and only raised while a rule or webhook names them), employee.offboarded (an offboarding workflow completed), asset.assigned (every assignment path), and, raised by cron only for events something listens to and once each, asset.offline, asset.online, certificate.expiring (at the configured number of days, then 7, 1, 0), credential.rotation_due (never carries the username, password, OTP seed, note or URL) and schedule.hourly|daily|weekly|monthly clock ticks (conditions on hour, weekday, day_of_month make "every Monday at 8").
    • Conditions beyond payload keys: rules can now test department, location_name, asset_type, asset_status, user_group, user_role, job_title, employment_status and more; they are looked up from the ids in the event before matching (never overwriting what the event carries) and also fill {placeholders}.
    • New rule actions: Run an RMM script (OFF until Allow rules to run RMM scripts is switched on; the job is created as a named technician through the RMM module, so permissions, platform checks and second-person approval apply; the script and parameters are fixed in the rule), Entra: disable the person's account and Entra: add the person to groups (the workflow actions, reused: write switch OFF by default, department check, never deletes), Send a chat message (calls a pluggable ChatNotifier; no Slack/Teams sender ships here, issue #36 plugs in) and Create an Odoo user (switched OFF and no Odoo write connection ships: contract and tests only; production Odoo is never written).
    • Run log: each try of the job queue shows as try N of 3 (and gave up); a failed run keeps its event so an administrator can Run again from the History drawer or the Run log tab (refused when it already succeeded on a retry, so nothing is done twice; audited).
    • Legacy ticket rules: Automation > Ticket rules lists which of them have an event-rule equivalent (asset online/offline + notify) and imports a switched-off copy on request; the old rules are untouched and keep running. The rest are listed with the reason they stay a separate engine.
    • API: GET/POST /automation_rules, GET/PUT/PATCH/DELETE /automation_rules/{id}, GET /automation_rules/runs[/{id}], POST /automation_rules/retry/{run_id}, GET /automation_rules/meta (administrators, user token only; documented in openapi.yaml and docs/API.md).
    • Not proven live: running a script on a real device, the Entra actions against a real tenant, chat delivery and anything in Odoo; see docs/AUTOMATION.md, "What is not proven live". Deploy notes: take a database dump first, then run the database update. Four additive, idempotent steps: automation_rules (five new action_type values, source_legacy_id unique), automation_rule_runs (context_json, attempt, trigger_source, retry_of_run_id), new table automation_emit_marks, and three settings.config_automation_* columns (both risky-action switches default 0).
  • Reporting and dashboards, completed (issue #34; database 2.6.223 and 2.6.224). Everything below is additive; existing reports, saved views and schedules behave as before. Nothing is switched on by this change.

    • PDF reports. Every catalogued report now has Download PDF in its toolbar (?export=pdf) next to Export CSV and Print. The PDF is made on the server with the TCPDF already bundled in plugins/TCPDF (no browser, no extra program) from the same tables as the CSV, so the three always agree; it states the date range and the filters used. A new schedule delivery, PDF attached to the email, sends the PDF as a real attachment: the mail queue can now carry attachments (email_queue_attachments, removed as soon as the message is sent or its queue row is deleted).
    • Quarterly schedules. A new Quarterly frequency (due again 3 months after the last send) for every schedule type, including the original headline summaries.
    • Reports hub headline row. Open Tickets, Mean Resolution Time (30 days), SLA Compliance (30 days), Tickets This Month and User Satisfaction (30 days, when CSAT is on), then Unassigned and Opened Today. All respect the viewer's department restriction.
    • Universal report filters. Service Desk and SLA, Tickets, Tickets: Day by Day, Time by Technician, Technician Performance and Customer Satisfaction (and the new SLA and backlog reports) now filter by department, location, technician, category, priority, status and asset. The Filters button in the report toolbar opens the panel; the filters are saved in saved views, replayed by schedules and shown on printed and PDF output. Values are validated against fixed lists or whole numbers, and combine with (never widen) a user's department restriction.
    • New reports. SLA Breach and Aging (response and resolution breaches in a date range, compliance, breaches by priority, technician and department, how long open tickets have been overdue, the breached tickets), Ticket Backlog (open tickets by age, status, priority, technician and department, a weekly backlog trend, the oldest open tickets) and Asset Lifecycle and Warranty (warranty expired or expiring, device age, replacement candidates, by type and department). Each is in the hub, the side menu, saved views, schedules, CSV and PDF.
    • Recipient groups and per-group views. On Scheduled Reports, create recipient groups (for example Executives, Service desk) and let one schedule send a different saved view to each group. Addresses must be active agents or administrator-approved recipients, checked when the group is saved and again each time cron sends. A failing group is named in the schedule's status without resending the groups that succeeded.
    • Tests. Six new test files under tests/i34_*: migrations (fresh vs upgraded SHOW CREATE TABLE), filters and PDF, report queries on fixture data, the existing report helpers with filters on, the scheduler with recorders, and a real SMTP hand-off to a throwaway local listener proving the PDF arrives byte for byte. Deploy notes: take a database dump first, then run the database update. Two additive, idempotent steps: 2.6.223 (email_queue_attachments) and 2.6.224 (report_recipient_groups, report_schedule_targets). No new settings, cron entries or permissions. Verify on the live mail setup that one PDF schedule arrives with its attachment (a real mailbox is the only thing the tests could not reach).
  • Entra ID / Intune sync, scheduled and incremental (issue #35, DB 2.6.225 to 2.6.226) Everything new is off by default and read-only towards Microsoft. Mock-tested only (tests/mock/graph_mock.php); UNVERIFIED LIVE, see the live checklist in docs/ENTRA_INTUNE_SETUP.md.

    • Scheduled Entra directory sync from cron (opt-in, every 1 h to daily) beside the Intune device sync.
    • Incremental sync with Graph delta queries. The first run saves a delta token; later runs read only changed users; users deleted in Entra are archived like disabled ones; an invalid token (410) falls back to a full read by itself. New microsoft_directory_users link table (Entra user id to contact): a renamed mailbox now updates its contact instead of creating a second one.
    • Previews (dry run) for the directory sync, Intune device sync, groups, licenses and Intune details: the real counts, inside a rolled-back transaction, logged separately and ignored by the sync watch.
    • Entra groups and members, license SKUs (purchased/consumed, who holds what) and Intune compliance policy states and detected apps per device (shown on the asset page).
    • Client-secret expiry: date field plus an optional lookup in Entra; alerts at 30 days, 7 days and on expiry (event integration.secret_expiring).
    • Stalled / failing sync watch now covers the scheduled directory sync and the new syncs; Admin > Update > Checks and Server status show the secret expiry and sync health.
    • Entra / Intune sync status page (/admin/entra_status.php): per sync last run, counts, classified error, next due, run and preview buttons; license, group and Intune-detail summaries.
    • Test Connection now probes groups and licenses when they are switched on. New migrations: 2.6.225 (columns on microsoft_integrations, sync logs, microsoft_directory_users), 2.6.226 (entra_sync_log, entra_groups, entra_group_members, entra_licenses, entra_user_licenses, intune_device_compliance, intune_device_apps, asset_intune_links.details_synced_at).
  • Slack and Microsoft Teams chat (issue #36), all off by default: Slack Events API (two-way ticket threads, direct messages and mentions to the virtual agent), /ticket slash command, ticket modal and message shortcut, status / priority / reply and approve / deny actions, explicit identity links with an optional email fallback, KB and service-catalog virtual agent, ChatActions / ChatNotifier services for other modules, Slack app manifest and Teams app package generators (Administration > Settings > Chat), and a Teams Bot Framework endpoint with JWT validation. Database 2.6.227 to 2.6.229. Proven against local mocks only; Teams and all Slack parts are unverified live (docs/CHAT_LIVE_VERIFICATION.md).

  • Quality module: the integration with the separate quality management system (issues #52 and #55; database 2.6.230 and 2.6.231). Everything is off until an administrator turns it on. RivetIT does not become a QMS: it shows each person their own items from the QMS, sends it changes, and signs people in to it.

    • Administration > Quality (new): the module switch (also under Settings > Modules), the QMS address (https only, checked against SSRF when saved and before every call, connection pinned to the vetted address, no redirects), client id and secret (sealed, write-only), Test connection, two signing secrets (shown once), the departments and the features that are enabled, timeouts and sync intervals, Sync health (ok / degraded / failing, with reasons), Run sync now, Queue full resync, the outbound and inbound event lists with Replay for dead-lettered events, and the integration error log (no secrets or personal data; the same correlation id the QMS logs).
    • Department portal: a Quality tab with My tasks, Corrective actions, Documents, Training, Notifications and History, plus a Department view for managers (overview, corrective actions, overdue tasks, training compliance, pending acknowledgments). Items are read live for the signed-in person and shown with the time they were read; when the QMS cannot be reached the page says so with a reference and shows nothing rather than stale data. Acknowledging a document or answering a corrective action asks for the password again unless the sign-in is fresh. Kiosk: counts, required training and document acknowledgment with the PIN; corrective actions never appear on a shared screen.
    • QMS client with a per-user context: OAuth 2.0 client credentials; every request for a person's items also carries a 60 second signed context (ES256, the SSO provider's key) naming that person, their department and role; the client refuses a person's request without one, and the QMS must authorize per person. Bounded retries with backoff, timeouts, a 1 MiB response cap and correlation ids.
    • Events: signed (HMAC-SHA256, 5 minute window), versioned, deterministic ids. RivetIT queues lms.assignment.*, lms.completion.*, employee.updated, employee.deactivated and department.updated (identifiers only, only for the enabled departments) with retry, dead letter, replay, a daily reconciliation window and a manual full resync. New inbound endpoint POST /api/v1/integrations/qms/events (scope quality:events, also signed, replay-protected) stores the QMS's own ids for people and departments (never matched by name) and lets the QMS ask for a resync.
    • Single sign-on (#55): the Quality module switch and the department list now gate the OpenID Connect provider (module_disabled, department_not_enabled); the portal "Open Quality system" button and the administrator menu entry follow. The seed row of oidc_provider_settings is now in db.sql (a fresh install could not switch the provider on).
    • Docs and tests: docs/QUALITY_CONTRACT.md is the exact contract the QMS must implement; docs/integrations/qms-configuration.md and qms-troubleshooting.md are new; OpenAPI and docs/API.md list the new endpoint. tests/quality_module.php (mock QMS in tests/support/mock_qms.php) and tests/quality_ssrf.php. Deploy notes: take a database dump first, then Update Database. Two additive, idempotent steps (2.6.229 -> 2.6.230 -> 2.6.231): settings.config_module_enable_quality (0), quality_settings, quality_features, quality_departments, quality_error_log, quality_outbox, quality_inbound_events, quality_sync_state, quality_id_map. Nothing is switched on. New cron entry cron/quality_worker.php (every 5 minutes, written by deploy/install.sh; exits at once while the module is off); on an existing server add it to the cron file or re-run the installer's cron step. Not verified against the real QMS (Datum): see the go-live checklist in docs/integrations/qms-configuration.md.
  • Training kiosk: pick tablets, one Web Clip profile per iPad (issue #43)

    • Fleet links > Per-device URLs and profiles now lists the department's tablets with a usable serial. Tick the ones you want, then download a CSV (device name, serial, type, start URL) or a ZIP with one .mobileconfig Web Clip per iPad (its own serial in the URL, Full Screen, not removable, kiosk icon embedded, stable PayloadUUIDs per device). Limit 300 tablets per ZIP; foreign-department, archived, non-device and unusable-serial assets are ignored. Needs the Training kiosk permission at level 3 plus Assets view, and the once-shown link token.
    • New API action fleet_device_profiles (Training kiosk admin API); each download is audited (training.kiosk_fleet_profiles).
    • No database change (the fleet-link schema is DB 2.6.151). Docs: docs/training-kiosk-setup.md. Tests: tests/kiosk_fleet_links.php.

Docs: added COMPETITOR_RESEARCH_FRESHSERVICE_JSM.md (Freshservice and Jira Service Management feature matrix, sourced 2026-11-01, prioritized gaps) and refreshed ROADMAP section 3.6 and open question 4. No code or database changes.

  • Security and permissions (issue #29)
    • New installs now give the built-in Technician role Knowledge Base Modify access (browser installer and setup_cli.php). Reporting stays an explicit administrator grant. Existing installs are unchanged; no database change.
    • Re-verified the 22 security/permission items from the user-guide findings against current code; most were already fixed. docs/user-guide/tools/app-findings.md now carries a status table and tests/security_findings_i29.php pins the guards.

RivetIT 26.10.80

v26.10.80
Release notes
  • Administration > Update: the check for generated vendor/composer files that differ from the repository is now a Note, not a warning (those files are regenerated by composer install). The RMM data store tile shows the short PostgreSQL version; the full version string is on hover.

  • Project tasks, phase 1: priority, description, labels, colour, and a task detail pop-up (database 2.6.207). Every project task can now carry a priority (none, low, medium, high, urgent), a description, labels (global ones, or ones that belong to one project) and a card colour. The Kanban card shows the priority as a coloured left edge plus a badge with its name (never colour alone), the labels as chips, an icon when there is a description, an Overdue due date, a progress bar and the assignee, and is more compact so more cards fit on a phone. The board has a Sort (manual order or priority) and filters by priority, label and assignee; these only change what you see, never the saved order. Click a card title or the expand icon for the task detail pop-up with every field editable; New Task and Edit Task have the new fields too, and the project's task list shows the priority marker and label chips.

  • Task actions: Delete (asks first; needs the same Service Desk level as before, removes the task's labels links), Duplicate (copies fields and labels, not completion) and Move to another project are in the detail pop-up, and a small per-card menu (Edit, Create ticket, Mark complete, Duplicate, Delete). The old Delete link in the task list now also applies the client-access check to project-only tasks (it only checked ticket tasks before).

  • Kanban drag fix. The hamburger drag handle (which overlapped the ticket and edit icons on phones) is gone. Grab a card anywhere to drag it: with a mouse it drags at once; on a touch screen press and hold for about a quarter of a second, then drag, so a quick swipe still scrolls the board. Moving cards between lanes and the saved order work exactly as before.

  • API: GET/POST /tasks, GET/PATCH/PUT/DELETE /tasks/{id} (new; scopes api:tasks:read / write / delete, role level 1 / 2 / 3, department scope), and priority, priority_name, description, color, labels on every task (including in GET /projects/{id}). Existing fields and clients are unchanged. See docs/API.md and api/v1/openapi.yaml. Deploy notes: take a database dump first, then Update Database (Administration > Update, or php scripts/update_cli.php --update_db). One additive, idempotent step (2.6.206 -> 2.6.207): tasks.task_priority (default 0), tasks.task_description, tasks.task_color, and the new task_labels / task_label_links tables. Nothing is back-filled. New code on a database that has not been updated yet keeps working (the new fields simply do not show).

  • Project tasks, phase 2: checklists, subtasks and comments (database 2.6.208). The task pop-up has a Checklist (add, tick, edit, delete, move up / down; the card shows 3/5), Subtasks (add, tick to complete, make a regular task again; the card shows 2/3; a subtask is a full task in the same project, one level deep, and shows in its parent's pop-up rather than as a card of its own) and Comments (add, edit your own, delete your own; administrators can delete any; shows author and time; the card shows a count). These save as you click, without leaving the pop-up. Deleting a task deletes its checklist and comments but never its subtasks: they stay in the project as ordinary tasks. Duplicate copies the checklist (unticked) and the subtasks. Moving a parent moves its subtasks; a subtask cannot be moved on its own.

  • Project tasks, phase 3: several assignees, files, reminders, repeating tasks (database 2.6.209). Several people can be on a task (the "Assigned To" person stays the primary, so older screens and apps keep working; the card shows avatars and the assignee filter matches anyone on the task; existing assignees are carried over). Files can be attached in the pop-up (scripts, web pages and installers are refused; downloads need the same department access as the task and are always sent as a download). A task can have a reminder (a date and time, or relative to the due date); the cron job (cron/task_reminders.php, run by the main cron) sends an in-app / push notification to everyone on the task once (administrators when nobody is assigned; no email). A task can repeat every N days / weeks / months / years: completing it (drag to Done, Mark complete, the pop-up, the API) creates the next occurrence once, with the due and start dates moved on (a task finished late gets a due date of today or later, month ends are clamped, not skipped), checklist unticked, same people, labels and reminder offset; comments and files are not copied.

  • API: every task gains parent_id, assignees, checklist, subtasks, comment_count, attachment_count, reminder_at, reminder_sent_at, repeat; parent_id, assignee_ids, reminder_at, repeat are writable; new sub-resources /tasks/{id}/checklist, /subtasks, /comments, /attachments. Existing fields and clients are unchanged. See docs/API.md and api/v1/openapi.yaml. Deploy notes (phases 2 and 3): take a database dump first, then Update Database. Two additive, idempotent steps: 2.6.208 (tasks.task_parent_id, task_checklist_items, task_comments) and 2.6.209 (tasks.task_reminder_at / task_reminder_sent_at / task_repeat_interval / task_repeat_unit / task_repeat_spawned_at, task_assignees with a one-time back-fill from tasks.task_assigned_to, task_attachments). Attachments are stored in uploads/task_attachments/ (created on first upload; the web server user must be able to write to uploads/; this folder is under /uploads/, so keep the nginx rule that stops scripts running there). New code on a database that has not been updated yet keeps working (the new panels and fields simply do not show).

  • RivetCore parity pass (Core 1.0.0-rc.55; database 2.6.210). Everything RivetCore's edition prompts ask for that RivetIT did not have yet. Nothing is switched on by this change.

    • Compliance frameworks (RivetCore #92 phase 4). New page Administration > Compliance framework: select a built-in or custom framework, copy one, edit it (every save is a new version, a stale edit is refused, restore any older version), archive, delete (only a never-selected one), export and import (JSON, 64 KiB, hostile files change nothing), and record a restore test. Once a framework is selected its requirements (backup age, off-site copy, encryption, restore-test age, retention minimums, log presets) are checked on the Compliance status page and recorded in snapshots, its retention minimums are refused below in the RMM settings and raised in the audit-trail cleanup, and every change is audited. With no framework selected nothing differs from before.
    • Backup facts. Each backup run now records which stores it really contained, whether it was encrypted and how long its writer keeps sets (backup_runs.run_stores, run_encrypted, run_retention_days; filled by the in-app backup and by deploy/backup.sh). The compliance checks judge those facts and the backup file's header, not configuration. The in-app backup now includes the RMM PostgreSQL store (rmm-postgres.dump, through RivetCore's BackupHook, password through a 0600 PGPASSFILE only) when that store is in use; a dump that is expected and fails fails the whole run (no zip is written, the run is recorded as failed).
    • Reporting / Power BI (RivetCore #91 phase 4). New page Administration > Reporting / Power BI and RivetIT's own views rpt_clients, rpt_sites, rpt_assets, rpt_tickets, rpt_alert_tickets, rpt_sla (allow-listed columns, no free text, client_id on every row, frozen by a fixture). Create, rotate and drop a dedicated read-only database account (MariaDB/MySQL, and a PostgreSQL role when the RMM store is PostgreSQL) with a generated password shown once and an administrator login that is used once and never kept; test the connection as that account; copy Power Query for this install; computed warnings. Timestamps in the RivetIT views are converted to UTC from the company time zone by zone name (daylight saving time proven on MariaDB 10.11); on a server without time zone tables the views are created unconverted and the page says so.
    • Security baseline. docs/security/EDITION_BASELINE.json states honestly, per RivetCore control, what RivetIT enforces (3 enforced, 15 partial, 2 gap), checked by scripts/edition-baseline-check.php (new CI job security baseline). New Security and trust page, public at /trust.php and an administrator copy with per-control detail; it uses only RivetCore's approved wording and a test fails the build if it ever says "SOC 2 compliant" and the like. The staff password minimum can no longer be set or stored below RivetCore's floor of 12.
    • RMM. An asset that moves department (asset edit, inline edit) now takes its RMM device with it through DeviceService::transfer(): jobs, open alerts and the stored log/change history follow, and a PostgreSQL store that is down repeats the history move later. The device-type rule of EndpointAssets::createForDevice now applies RivetCore's pre-rc.29 OS rule when a caller sends no type (the Core conformance kit pins it).
    • Tests. Three stale conformance tests repaired (API keys became integration clients, https-only webhooks, the asset types RivetIT really has); new tests for backup facts (Core kit plus the edition rules), the in-app PostgreSQL backup, frameworks, reporting views/page/PostgreSQL role, the trust page and the baseline check. Deploy notes: take a database dump first, then run the database update (Administration > Update, or php scripts/update_cli.php --update_db). One additive, idempotent step (2.6.209 -> 2.6.210): three nullable columns on backup_runs and the six rpt_ views. The reporting account is NOT created by the update. If the database server's time zone tables are not loaded (mariadb-tzinfo-to-sql), the RivetIT reporting views are created without UTC conversion; load them and press "Rebuild the RivetIT views".
Release notes

Deploy notes: back up the database, pull this tag, composer install --no-dev (RivetCore 1.0.0-rc.19). No database change (DB stays 2.6.172; rc.19 has no Core migration) and no cron change. After the update: publish the agent 0.7.1 from the Agent page and rebuild the MSI template with it, then install the starter pack's update (pack 1.0.1) from the Starter pack page.

  • Alerting tab: thresholds for policy checks. The threshold endpoints (GET/PUT endpoint_devices/{id}/checks/{key}/thresholds) answered 404 for every check a policy delivers (the whole starter pack), so the asset page Alerting tab showed no editor for them and an override could not be saved. Core now also looks in the device's effective policy; the editor renders and an override saves for a policy-defined check.
  • Starter pack 1.0.1. battery-report, defender-status, defender-update-signatures and disk-health print unknown / n/a / a note instead of Windows sentinel values (71582788 minutes, 65535 days, a bare /).
  • Agent 0.7.1 (Windows), found by running the agent on a real Windows 11 machine: a junction inside an allowed file-transfer folder could reach files outside it (fixed; update agents 0.6.0 and older if file transfer is on, it is off by default); scripts that call ping, whoami, w32tm and similar failed or reported success with empty sections (the job environment lacked PATHEXT, COMPUTERNAME, ComSpec); PowerShell error output arrived as #< CLIXML XML; proc_kill reported success before the process ended; service checks failed with "Access is denied" on Dnscache and mpssvc; the av check now accepts a third-party antivirus Windows Security Center lists as on. Full evidence: RivetCore docs/rmm/WINDOWS_VERIFICATION.md.
  • Tests: tests/rmm_ui_p23_c_alerting.php (a policy with a thresholds check assigned to a device through the real handler: the thresholds API answers, the Alerting tab renders the editor, an override saves, shows and clears; a check no list or policy defines, and a device the policy does not reach, are still refused). The test fails on rc.18 and passes on rc.19.
Release notes

[26.10.48] RivetIT — the Remote button finds the device (MeshCentral node ids)

Deploy notes: back up the database, pull this tag, composer install --no-dev (RivetCore 1.0.0-rc.18), then run the database update (Administration > Updates or php scripts/update_cli.php --update_db). DB 2.6.172: Core migration 0027_rmm_mesh_node_ids rewrites the stored MeshCentral node ids to one form (data only, no schema change, idempotent). No cron change.

  • Bug fixed: Remote opened the MeshCentral home page instead of the device. MeshCentral finds a device by its 64-character id (...?viewmode=11&gotonode=<64 characters>), but the agent reports the same 48 bytes as 96 hex characters and the link carried node//<hex>. The link is now <mesh>/?login=<token>&gotonode=<64 characters>&viewmode=11&hide=31; every stored form (hex, 64 characters, node// either) is converted when the link is built, so it works even before the database update.
  • Paste whatever you have. The MeshCentral node field (Remote card, legacy device page) accepts the address bar of the device in MeshCentral, node//..., the 64 characters after gotonode=, or 96 hex characters, and shows node// plus 64 characters. Help text: "In MeshCentral open the device and copy the address bar, or paste the node id."
  • Same node, one mapping. An agent that reports the node in hex and a hand-typed 64-character id are the same node: conflict detection compares them as such.
  • Not verified on a live MeshCentral with this exact link; the shape is the one MeshCentral shows in its own address bar. See RivetCore's docs/rmm/REMOTE_AND_LOGS.md.
  • Tests: tests/rmm_ui_p68_remote.php (the launch URL carries the 64-character form for a device whose stored id was hex, on every view; a pasted address is saved canonical), tests/rmm_ui.php (field help), tests/endpoint_agent_migration.php (DB 2.6.172 step), tests/browser/rmm_rc17_remote_smoke.mjs extended.
Release notes

Deploy notes: back up the database, pull this tag, composer install --no-dev (RivetCore 1.0.0-rc.17), then run the database update (Administration > Updates or php scripts/update_cli.php --update_db). DB 2.6.171: Core migration 0026_rmm_mesh_hide adds two columns to endpoint_agent_mesh_settings (mesh_hide default 31, mesh_hide_tools empty); nothing else is touched. No cron change.

  • No MeshCentral screens around the session. The Remote button used to land on MeshCentral's own interface (header, device list, tabs). The URL now carries MeshCentral's hide parameter, as Tactical RMM's "Take control" does: the default hide=31 hides the masthead, top bar, footer, tab titles and the left bar with the device list, so only the device view is left. From MeshCentral's page source: bit 1 masthead, 2 top bar, 4 footer, 8 titles, 16 left bar, 32 back button.
  • It opens in a NEW tab, reliably. Browsers block a tab opened after an asynchronous request, so the tab could silently fail to open. The click now opens a blank tab immediately and sends it to the session when the launch call answers. If the launch fails the blank tab is closed and the reason is shown; if the browser refuses the tab an Open remote session link appears instead. A tab from an earlier launch is never reused (each launch has its own login token). Same for the Remote tab (Desktop, Terminal, Files), the legacy device page and the Tactical asset page.
  • Settings. Administration > Endpoint agent > MeshCentral: Hide MeshCentral interface elements (0 to 255, default 31; 31 = only the remote screen, 0 = the full MeshCentral and the old URL) and an optional value for Terminal and Files (empty = same; 0 allowed). Only shown once the database update has run.
  • MeshCentral does not connect by itself. From its source, nothing in the URL starts the connection: the new tab shows the Desktop view with MeshCentral's Connect button; click it once. The card says so.
  • Not verified on a live MeshCentral. The hide bits come from MeshCentral's page source (default.handlebars); the smoke runs against a mock Mesh server. See RivetCore's docs/rmm/REMOTE_AND_LOGS.md, "The Remote button", for the test to run on yours.
  • Tests: tests/rmm_ui_p68_remote.php (default hide on every view, the form, per-view values, 0 equals the old URL, refusals, a technician cannot change it, the helper is linked before the page scripts and nothing opens a tab after the async answer), new browser smoke tests/browser/rmm_rc17_remote_smoke.mjs (10 checks: a tab exists before the launch answers, the final URL has login, gotonode, viewmode=11 and hide=31, the tab cannot reach its opener, every click opens its own tab, the popup-blocked link, a failed launch closes the tab, hide 0, Terminal and Files, the admin card).
Release notes

DB 2.6.170 (idempotent, additive: ADD COLUMN IF NOT EXISTS on the integration's own tables only; the nearly full settings table is untouched; every default keeps an existing connection behaving as before: cloud edition, no asset creation, no retirement). Builds on the cloud-only ManageEngine MDM sync of 26.10.45 and earlier; there is still one inventory (Infrastructure > Assets). Setup, field mapping and every docs-only assumption: docs/MANAGEENGINE_MDM.md. The on-premises edition, MSP mode and the asset writes of this release have not been run against a real ManageEngine server (local mock only); a first live run of the cloud edition happened before (see the is_removed fix below) and corrected one response shape.

  • On-premises edition: server address plus API key in the Authorization header only (stored encrypted, never shown or logged), an expiry date the administrator enters with a notification 30 days before and when expired, Test connection (shows the device count). The server address is vetted by RivetCore's URL policy like a webhook URL: https only, public addresses, plus the networks listed under Administration > Webhooks > Allowed internal networks for a server on a private LAN; loopback, link-local and metadata addresses never; the connection is pinned to the vetted address. Optional TLS verification off for a trusted self-signed certificate. Plain http is accepted only to loopback behind an explicit test constant.
  • MSP edition (either edition): customers are listed and each one's devices fetched with the X-CUSTOMER header; each customer name goes through the shared Client mapping queue (saved mapping, exact department name, ignore, or wait as needs mapping). A default department catches the rest.
  • Client: MdmClient interface + ManageEngineMdmClient (cloud OAuth and on-premises key), response size cap (8 MB), pagination with a page cap, 429 / 5xx backoff, 401 handling per edition (no refresh loop for a key), no logging.
  • Assets: match order existing link, serial, IMEI, UDID, then a unique same-named asset in the device's department; a device matching nothing creates a Phone or Tablet asset (Laptop / Display / Other by platform) when Create and maintain assets is on (new connections; upgraded ones keep it off, and with it off a sync writes nothing on an asset exactly as before). With it on, make, model and OS go through asset_sync_state (a human edit is never overwritten) and the user becomes the contact when exactly one contact of the department has the e-mail (the Intune rule).
  • Stale devices (off by default): archive an asset whose MDM device is gone or removed for N days (never a merely quiet device, never an RMM-managed asset), into the existing review queue; restored automatically when the device returns.
  • Recovery sync watch: the MDM sync is registered (source mdm) with its own expected interval (the connection's sync interval); failed runs, including a refused server address, always write a mdm_sync_log row. Mute / interval override / Forget history work.
  • UI: the Endpoints menu group is now Managed devices (label only; ids, URLs and permissions unchanged) with Mobile devices: compliance, department, platform, last-contact-older-than-N-days filters, search, review list with the needs-mapping reason; the asset card shows owner type, enrolled date, customer and a console link; a settings form with an edition switch. Permission: module_client reused (documented), no new module.
  • Security: api_key_enc added to the rewrap registry; new tests/crypto_registry_scan.php fails when any *_enc column in db.sql or the migrated schema is neither registered nor explicitly excluded.
  • Tests: new tests/manageengine_mdm_onprem.php (93 checks), tests/crypto_registry_scan.php, tests/browser/mdm_smoke.mjs; _web.php +46, mock extended (on-prem key, MSP customers, malformed and oversized bodies). Fixed a baseline bug in a client test (array_unique key order).
  • Not verified: no live on-premises or MSP server; the MSP customer-list path, the on-premises 401 behaviour, key permission names, per-device console URL and an API version field are not documented (see the doc's list).

Full notes, including the merged mapping options (DB 2.6.169) and the is_removed fix: CHANGELOG.md. Setup and every docs-only assumption: docs/MANAGEENGINE_MDM.md.

Not deployed anywhere by this release.

Release notes

No database step. Fixes the asset page's Run script dialog, which listed only "PowerShell text" (the RivetCore script library was reachable only from a separate card on the Jobs tab, and Linux devices had no Run script at all), and merges the two script menus.

  • One dialog for everything. What to run lists, in this order: Script library (N) (only when the scripts switch is on, the viewer may run saved scripts and the library has scripts for this device's platform), A saved script (the older rmm_scripts list, Windows, only if any, grouped "Saved here" and "Synced from "), and PowerShell (type or paste) on Windows or Shell (type or paste) on Linux (Core job types powershell and shell, ability rmm.job.run_script; the option is hidden for a viewer without it). A single choice is shown as text, never as a one-option dropdown. The old wording "PowerShell text" is gone.
  • Library pane. A search box filters the list live (name, description, tags; the count is announced), the list is grouped Read only / Changes the system / Disruptive / your own scripts, and the chosen script shows its risk badge, tags, description and a typed parameter form (numbers with min/max, bounded text, choices, secrets as password fields, required marks and hints). A script's dry_run parameter is the checkbox Dry run (nothing is changed), ticked by default; it always posts an explicit true or false. A script that needs approval says the request goes to the approvals queue (and does); a destructive script has a required confirm box. The form posts to the same handler as the Jobs-tab card (agent/post/rmm_automation_scr.php, action run_script): CSRF, client scoping and Core's ability checks are unchanged, the person lands on the Jobs tab with "Job queued". Empty library: the dialog says so and links the Starter pack (administrators) and Scripts.
  • Linux devices have the Run script button and the dialog (Linux library scripts, free-form shell text).
  • One Scripts entry. The menu had Script library and Scripts. Now there is one Scripts entry (active on both pages). Its page has a Library tab (the RivetCore library, starter pack included) and a Synced from tab (the older list; the tab exists only when that list has rows or a vendor RMM integration is configured). rmm_scripts.php with nothing of its own to show redirects to the Library tab (query kept); rmm_script_library.php, rmm_script_run.php and every deep link keep working. With the scripts switch off the menu keeps the single older entry. The older list is also usable on a phone now (table scrolls inside its card).
  • Import into the library. Administrators with the script-write level get a one-time card on the older list: the scripts written by hand (no vendor script id) become signed library scripts through the technician API (language from the stored type, tags from the category plus risk:<class>, class chosen by the administrator, default changes_system; a disruptive class sets approval and destructive like the starter pack; note "imported from legacy Scripts"). Idempotent (same name and text is skipped), audited, CMD scripts are reported, the older rows stay and say Imported.
  • Tests. New tests/rmm_ui_p23_d_runscript.php (dialog markup for Windows, Linux, restricted roles, empty states; real runs through the handler with the job decoded for library_script_id, version and params; dry run on and off; approvals; destructive confirm; CSRF, viewer and department refusals; switch and module off with zero queries; menu, tabs, old URLs, import idempotence) and tests/browser/rmm_runscript_smoke.mjs (13 checks in Chrome, desktop and 390 px, light and dark). tests/rmm_ui.php and rmm_smoke.mjs updated for Linux's enabled button.
  • Not verified: nothing was deployed; browser run is Chrome only.
Release notes

Service Desk ticket filters now take more than one value, and a saved view can be made the default. DB 2.6.166: one idempotent step gated on 2.6.165 that creates one new table, ticket_view_defaults (CREATE TABLE IF NOT EXISTS, nothing existing altered; db.sql carries the same table, fresh import and migrated schema compared over all 408 tables with 0 differences). Nothing is switched on by the update; until it runs everything works as before and only the default-view stars and boxes are hidden.

  • Multi-select filters. Assigned to (Me, Unassigned, any agents, in any combination), Priority (Low / Medium / High), Board, Category, Where (On-site / Remote; both is the same as no filter), Due (Overdue and Due today are independent: either) and the specific statuses are checkbox groups, each with Select all, Clear and a count badge. Inside a group it is OR, across groups AND (High or Medium, and assigned to me or unassigned). The saved-view dialogs (Save Current View, Edit) show them as labelled fieldsets; the filter bar at the top of Tickets shows them as buttons that open the same checkbox list and apply when it closes (Esc closes; keyboard and 390 px checked, light and dark). The Status button keeps All open / All closed / Open + Closed and lets you tick specific statuses instead (the two cancel each other). Assigned to moved from More Filters into the bar; Where and Due are new in More Filters.
  • Old links and saved views keep working. assigned=me, priority=High, board=5, category=9, onsite=1, overdue=1 are read as one-item lists; the stored form of a view is now lists (assigned[]=me&assigned[]=7&priority[]=High&priority[]=Medium). A comma list in one value is read the same way. Two small fixes ride along: the dashboard's tickets.php?assigned=0 (Unassigned tickets) used to ignore the filter and now shows unassigned tickets, and assigned_to=<id> (the "My tickets" link) is read as assigned.
  • One reader, no raw input in SQL. includes/ticket_filter_sets.php turns the request into validated sets (ids are digits only, priorities come from a fixed list, nested arrays, objects and anything else are dropped, 50 values per group at most) and builds the IN (...) SQL; the Tickets list, the kanban board, saved views, the active-view highlight and the API all use it. The kanban board now honours board, priority, on-site, due and tag filters too (it only applied status, assigned and category before). Pagination and sort links carry the lists.
  • Readable descriptions. Open tickets · Priority: High, Medium · Assigned to: me, Unassigned, Ted Tech · Board: Helpdesk, Infrastructure · Where: On-site · Due: Overdue, Due today; tags are described too.
  • Default view. A star next to each view in the list makes it your default (click again to clear); the Save and Edit dialogs have Make this my default view. Opening Tickets, or the list or kanban link, with no filter of its own loads it, with a banner ("Showing your default view ... Show all open tickets"). Any filter in the URL (or ?nodefault=1) wins, Reset to my default (the Reset button) comes back to it, and Show all open tickets is one click. An administrator with the support module at level 3 can also make a view shared with all agents the default for everyone who has no personal default (box in the dialogs, or the view's menu). One personal default per person and one shared default, enforced by a unique key (default_key = u:<user id> or shared) and replaced in one statement. A deleted, archived or no-longer-shared default is ignored: the person falls back to the shared default, then the built-in open list. Setting and clearing are written to the log and the audit trail (ticket_view.default_set, ticket_view.default_cleared).
  • Permissions. The Edit dialog used to load any saved view by id; it now only opens views the person may edit (their own, or shared ones for an administrator) and needs the support module at level 2, like the Save dialog. The default stars need the support module (a read-only agent can pick a personal default; only an administrator can touch the shared one).
  • API (no new routes; check_openapi_drift.sh output is identical to the base). GET /ticket-views adds shared, is_default, default_scope, and a view with several values gives comma lists in params (priority high,medium, assigned me,unassigned, board, category_id; assigned to only the caller keeps mine = 1, so older app versions read it as before). GET /tickets accepts the same comma lists for priority, assigned, board, category_id and onsite; category_id (which the app sent but the endpoint ignored) now filters, and an unknown priority still matches nothing.
  • Not changed. Export still exports every ticket (it never followed the filters); there were no saved-view count badges in the sidebar, so none were added; the per-ticket live stream is not a list and has no filters.
  • Tests: new tests/ticket_views.php (197 checks: parsing and validation of old and new forms, hostile and oversized values, the SQL run against fixture tickets for every combination, form to stored query to description round trips, default storage rules and resolution order, the page, kanban, the post handlers, both dialogs, permissions and the API with real HTTP), new tests/browser/ticket_views_smoke.mjs + ticket_views_seed.php (15 browser checks), tests/endpoint_agent_migration.php +8 (2.6.166).
  • Not verified: nothing was deployed; the browser run was Chrome only.
Release notes

Fixes false-positive and repeated integration-sync alerts from the recovery watcher (src/Recovery/SyncWatch.php, run from cron/cron.php). DB 2.6.165: one idempotent step gated on 2.6.164 that creates one new table, recovery_watch_overrides (CREATE TABLE IF NOT EXISTS, nothing existing altered; db.sql carries the same table). Nothing is switched on or off by the update; until it runs everything works and only the new Mute / Expected interval / Forget history buttons are disabled. Two behaviour changes apply at once, with no setup: a sync whose newest run is over 14 days old is dormant and no longer mailed, and the same problem is mailed at most once per 24 hours.

What went wrong on a live install (about 12 emails a day): (1) "UniFi sync has stopped: no run for 94 days (expected every 17 min)": the watcher inferred the interval from old scheduled runs of the in-app UniFi sync, which now runs from an external script that does not write unifi_sync_log. (2) "Sophos Firewalls: the last 3 runs all ended in an error": an external script copied the old "sync every enabled integration" loop, hit the rivetit_agent integration (no vendor API) and logged its "Endpoint agent devices are managed from the device page" refusal as a failure.

  • Dormant rule. If an integration's newest run is older than Sync dormant after (days) (new recovery setting, default 14, 0 = off; never sooner than twice its stale limit, so a monthly job is allowed to be quiet) it is dormant: no stale or failing mail, shown as "dormant since ". An integration that never ran in that long is dormant too. One "no longer running" notice goes out only if it was healthy before and has just become dormant (within 30 days of turning dormant), at most every 30 days; one that never ran, whose last run failed, or that has been dormant for longer is silent.
  • Per-integration controls (table recovery_watch_overrides: source, integration id, muted, expected interval in seconds, forget-before timestamp, note, who and when). On Admin > Backup > Integration sync watch: Mute / Unmute (no mail at all; the panel still shows what the state would be), Expected interval (1 minute to 90 days, minutes / hours / days, with a note; wins over the per-source setting and over an inferred interval, so an external daily script is declared as every 24 h and goes stale after three missed days), Forget history (every run so far is ignored; nothing is deleted; the integration shows as dormant until it runs again). Admin only, CSRF-checked, the integration must be one the watcher watches (never rivetit_agent), every change is written to the log (Settings / Edit) and the audit trail (recovery.sync_watch_changed, with the acting user).
  • Repeat cadence. The same problem is mailed at most once per 24 h (new setting Repeat a problem alert after (hours), 1 to 720; it was a fixed 6 h). The first occurrence is still immediate. A recovered notice ("sync has recovered") goes out once when an integration that had alerted becomes healthy again, then the state is cleared so the next failure is immediate again. The window applies to every recovery alert (backup failed / stale, restore drill stopped, audit trail chain) as they share the same alert table.
  • Failure rule. "The last N runs failed" counts only runs of that integration whose status is failed AND whose error text is real: blank text, [], null and the rivetit_agent sentinel ("... agent devices are managed from the device page") are not errors, and a sentinel or empty-text run in the last N breaks the chain. A run stuck "running" counts only if it carries error text (it had none before, so this rule is now effectively off). The rivetit_agent type is excluded everywhere (watcher, panel, handler).
  • State per integration on the panel: healthy, failing, stale, dormant or muted, with the last run (and its status), the expected interval and where it came from (set by an admin, the setting, inferred), why it alerted (the detail text and when the last alert went out) and the note. The inferred interval now uses scheduled runs only (triggered_by = 0) once there are three.
  • Docs. docs/RECOVERY_RUNBOOK.md: new "Integration sync watch and external sync scripts" section: an external sync script should write to rmm_sync_log / unifi_sync_log (own integration id, a failed row only with real error text) or the admin should set its expected interval; any script that copies the old "sync all enabled integrations" loop must skip type rivetit_agent. The Sync alerts row and the 24 h repeat are updated.
  • Tests: new tests/recovery_syncwatch.php (100 checks): the exact live cases (UniFi, 94-day-old runs with a 17 min inferred interval, is dormant with no email; Sophos with mixed success / failed / sentinel rows is not failing, even six sentinel failures in a row; a genuinely failing integration with three real errors sends one email and none for 24 h, then one more; the expected-interval override (20 h healthy, 80 h stale); mute; the recovered notice once and a new failure immediate afterwards), the dormant boundaries and notice rules, forget history, the repeat setting, only-own-runs and the endpoint-agent exclusion, panel rendering (all five states, escaping, CSRF, no inline handlers), the handler (mute, interval, forget, refusals, audit and log lines) and a database without the new table. tests/recovery_drill.php updated for the 24 h window and the real-error rule; tests/endpoint_agent_migration.php +8 (2.6.165: gated, db.sql equals the migrated schema, replays keep data, a dropped table is recreated).
  • Not verified: the mail was exercised through the queue stub and the in-app notification stub, not a real SMTP server. Nothing was deployed.

All RivetIT releases on GitHub

RivetMSP 26.10.53

v26.10.53
Release notes

No database version change (DB stays 2.6.137). Update Database now also catches up pending RivetCore migrations (0048/0049) on every run.

Fixed

  • Update App refused with "the app files are not owned by the web server user". 26.10.52 only trusted a checkout owned by the web user; the usual shared layout (checkout owned by the administrator, web user in its group) works again. The page only refuses when its own write checks fail.
  • Update page: a "Take a backup now" button (opens Backups) and a last-backup line.
  • Update Database applies RivetCore migrations a half-finished earlier update skipped (it could mark the edition database 2.6.137 while the files on disk still had RivetCore rc.55).

RivetMSP 26.10.52

v26.10.52 Pre-release
Release notes

Database 2.6.137 (RivetCore migrations 0048 audit hash chain tables and 0049 installer managed_externally; both additive, applied by Update Database). Do not deploy before the mobile apps are updated (see "Mobile API changes").

Do not press Update App (or run deploy/update.sh) on a server with mobile-app users until the mobile apps have shipped their update. The GitHub "pre-release" flag gates nothing: Update App pulls the branch of the configured release channel (production = master, beta = beta) and does not look at releases, and both branches contain these changes. Updating also runs the database migrations (Update Database).

Security

  • Client access: archive/restore/delete client, notes on clients/contacts/assets, edit_transfer, edit_project and linking tickets to projects, worksheets, the client and expense CSV exports, the vendor-contact import/export and RMM assign_client now check the user's client restriction on the client stored on the record (not the one in the request). Credentials under an asset or service stay within the asset's client; a credential can only be linked to an asset of the same client. Closed worksheets (signed by the customer) can no longer be edited or deleted.
  • Modals: agent/modals/* now need the module their folder belongs to (financial, sales, client, credential, KB, support); an unmapped folder fails closed to module_support. Ticket modals check the ticket's client.
  • RMM: reboot needs module_rmm_scripts 2, free-form run command 3; link, unlink, sync, assign and sync scripts need Write (they accepted Read).
  • Sign-in: the mobile API password login refuses an account that must have two-factor but has not set it up (and is out of grace). Changing your email, turning two-factor on or off, and adding or removing a passkey now need your current password and a live session (disabling two-factor is a form, not a link).
  • Vault: every place that shows decrypted credential values (CSV export, client PDF, global search, asset/contact/client pages and modals, edit form, share link, TOTP, API) goes through one gate: the reveal limit (30 per 10 minutes by default), a password step-up for exports and PDFs, and an activity-log line. Viewing an asset or contact page counts as one reveal.
  • Notifications: a notification about a client goes only to administrators, users with no client restriction and users whose list includes that client (also in the API). Open: per-notification-type module permission (e.g. invoice amounts) is not filtered yet.
  • Portal and guest ticket views show only Public and Client replies (the guest sign-off page had no filter).
  • Money: saved-card, portal and autopay charges use the remaining balance, never the invoice total; an invoice or quote can only be marked sent from Draft (no re-invoicing, no charge-in-full of a Partial invoice); deleting a payment needs level 3. Pax8: re-mapping or un-mapping a company stops billing the old client, billing uses one deterministic map row and only companies with billing on, and the subscription refresh runs under the billing lock in one transaction. Open: the double-charge race between portal, agent and autopay (design in docs/design/stripe-double-charge-INJ-26b.md), unbounded quote/recurring discounts.
  • Core TEN-1..12 (tenant isolation in Core's RMM actions): no change expected for RivetMSP, administrators are unrestricted by design.

Mobile API changes (apps must ship first)

  • PUT /api/v1/me with device_public_key_pem now requires current_password (and totp_code or a recovery code when MFA is on); already-enrolled devices keep working.
  • GET /api/v1/credentials/{id} shares the web reveal limit and answers 429 with Retry-After when exceeded.
  • POST /api/v1/auth/login answers 403 "Two-factor authentication is required..." for accounts that must have MFA and have none.

Changed

  • RivetCore 1.0.0-rc.55 to rc.60: shared UI assets updated; includes the rc.59 CSP/XSS fixes in the webhook wizard and the migrations above. The audit chain is installed but not switched on (no AuditChainWriter wired yet).
  • Add device page uses the RivetIT layout (one card, platform buttons, device type, Advanced section, "what this server can build"); the Mesh wizard and Agent fleet pages got the RivetIT-style cleanups (calm empty state, KPI tiles, one button style).
  • Update page: the web server user no longer runs git on files it does not own. With the new root helper (deploy/install-web-update.sh, one sudo rule) Update App takes a backup, updates as the file owner, runs composer and the migrations and reloads PHP-FPM (php<full version>-fpm first); without it, a checkout the web user owns updates in the page as before, otherwise the page tells you how to set up the helper. deploy/update.sh no longer hard-codes php8.4-fpm and fails loudly when no PHP-FPM unit is active.

Fixed

  • The .deb install command makes the package readable by apt's _apt user (no more "unsandboxed as root" warning).

RivetMSP 26.10.51

v26.10.51
Release notes

No database change (DB stays 2.6.136).

Added

  • Remote access (MeshCentral) setup: a collapsible "Don't have a MeshCentral server yet?" helper in the wizard (what MeshCentral is, what to have ready, first-account and sign-up hardening, reverse-proxy and WebSocket notes, the device paths that must never sit behind a CAPTCHA, the Linux / BSD agent choice). Steps 2 and 5 now cover Linux servers, not only Windows PCs.
  • Dependabot configuration (Composer and GitHub Actions, weekly).

Fixed

  • Browser setup wizard: the Company step no longer ends in a 500 (the vault master key was minted in the User step but saved in the Company step, a separate request). The Company step also no longer leaves the invoice, quote and ticket prefixes and counters empty.
  • RMM pages: Linux servers show the right OS icon on the asset page; flash errors no longer raise toastr.danger is not a function; the patch compliance client column and the endpoint agent device table no longer break words on phones; the "Waiting for approval" row no longer prints a stray comma.
  • Security: Pax8 billing sync only reuses the synced client's invoice and runs under a named lock; copying a quote derives the client from the source quote and checks access first.

Changed

  • Dependencies: aws/aws-sdk-php 3.403.1, predis/predis 3.7.1 and their transitive packages (composer audit: no advisories). a11y_auto.js escapes label lookups with CSS.escape.

RivetMSP 26.10.50

v26.10.50
Release notes

No database change (DB stays 2.6.136).

Fixed

  • Notifications: the hourly "Automatic backup was NOT taken" alert is now sent at most once per 24 hours instead of every cron run, so the bell no longer fills with duplicates. The "See all Notifications" button in the notifications modal is visible and readable again.
  • Full UI sweep (client portal, agent portal, administration, RMM, billing, docs): about a hundred layout, alignment, contrast, dark-mode, mobile-width and accessibility defects fixed across pages, modals and forms, plus several functional bugs found by clicking through every flow.
  • Administration > Settings: icons and rows in the settings directory line up and there is room above the footer; the Remote MCP entry uses a link icon.

Changed

  • The administration footer no longer links to the upstream ITFlow docs, forum and services sites; it shows the product name, version and a link to the in-app Docs (Knowledge Base).

RivetMSP 26.10.49

v26.10.49
Release notes

Per-user widget dashboard with a 'Waiting on you' money band, Lead pipeline board, tickets page aligned with RivetIT, scan fixes (accessibility, contrast, portal CSP, KB upload, vendor links). DB 2.6.136.

Release notes

Deploy notes: back up the database, pull this tag. RivetCore is now 1.0.0-rc.19 (the tracked vendor/rivet/rivet-core is the tag's git archive, so no composer install is needed). No database change (DB stays 2.6.95), no cron change. Then, for the Windows fixes: publish agent 0.7.1 (or newer) to your rings, rebuild the MSI template with it, and install the starter pack's update (Starter pack page, "updated" items).

  • Bug fixed: the Alerting tab of the asset page could not show or override the thresholds of a check delivered by a policy. The threshold endpoints answered 404 for every check defined only by a policy (the whole starter pack is delivered that way), so the tab said "This check declares no thresholds" and the override form was missing. They now answer for those checks; the editor renders and the override saves like any other.
  • Security (agent): agent 0.7.1 closes a file-transfer hole on Windows where a junction inside an allowed folder reached files outside it (file transfer is off by default and needs an administrator-listed folder). Anyone running agent 0.6.0 or older with file transfer on should update the agent.
  • Agent fixes from the first run on a real Windows 11 machine: PowerShell jobs get a proper environment (ping, whoami, w32tm and others are found; nine starter scripts no longer report success with empty sections), PowerShell error output is plain text instead of CLIXML, proc_kill waits for the process to end, service checks read services with a restrictive descriptor, and the antivirus check accepts a third-party product Windows Security Center lists as on.
  • Starter pack 1.0.1: battery-report, defender-status, defender-update-signatures and disk-health print unknown or a note instead of Windows' sentinel values. Policies you edited and scripts you edited are never overwritten.
  • Tests: tests/rmm_ui_p23_c_alerting.php gains a regression block (a check defined only by a policy: GET thresholds answers, the asset page Alerting tab renders the editor, the override saves and is removed); tests/rmm_ui_p1213_starter.php reads the pack version from the pack instead of assuming 1.0.0; tests/wave1_integration.php checks the rc.19 pin.
Release notes

[26.10.24] RivetMSP — the Remote button finds the device (MeshCentral node ids)

Deploy notes: back up the database, pull this tag, run the database update (Administration > Updates or php scripts/update_cli.php --update_db). RivetCore is now 1.0.0-rc.18 (the tracked vendor/rivet/rivet-core is the tag's git archive, so no composer install is needed). DB 2.6.95: Core migration 0027_rmm_mesh_node_ids rewrites the stored MeshCentral node ids to one form (data only, no schema change, idempotent). No cron change.

  • Bug fixed: Remote opened the MeshCentral home page instead of the device. MeshCentral finds a device by its 64-character id (...?viewmode=11&gotonode=<64 characters>), but the agent reports the same 48 bytes as 96 hex characters and the link carried node//<hex>. The link is now <mesh>/?login=<token>&gotonode=<64 characters>&viewmode=11&hide=31; every stored form is converted when the link is built, so it works even before the database update.
  • Paste whatever you have. The MeshCentral node field (Remote card, legacy device page) accepts the address bar of the device in MeshCentral, node//..., the 64 characters after gotonode=, or 96 hex characters, and shows node// plus 64 characters. Help text: "In MeshCentral open the device and copy the address bar, or paste the node id."
  • Same node, one mapping. An agent that reports the node in hex and a hand-typed 64-character id are the same node: conflict detection compares them as such.
  • Not verified on a live MeshCentral with this exact link; the shape is the one MeshCentral shows in its own address bar.
  • Tests: tests/rmm_ui_p68_remote.php (the launch URL carries the 64-character form for a device stored with the hex id, on every view; a pasted address is saved canonical), tests/wave1_integration.php (the 2.6.95 step, the rc.18 pin), tests/browser/rmm_rc17_remote_smoke.mjs extended.
Release notes

Deploy notes: back up the database, pull this tag, run the database update (Administration > Updates or php scripts/update_cli.php --update_db). RivetCore is now 1.0.0-rc.17 (the tracked vendor/rivet/rivet-core is the tag's git archive, so no composer install is needed). DB 2.6.94: Core migration 0026_rmm_mesh_hide adds two columns to endpoint_agent_mesh_settings (mesh_hide default 31, mesh_hide_tools empty); nothing else is touched. No cron change.

  • No MeshCentral screens around the session. The Remote button used to land on MeshCentral's own interface (header, device list, tabs). The URL now carries MeshCentral's hide parameter, as Tactical RMM's "Take control" does: the default hide=31 hides the masthead, top bar, footer, tab titles and the left bar with the device list, so only the device view is left. From MeshCentral's page source: bit 1 masthead, 2 top bar, 4 footer, 8 titles, 16 left bar, 32 back button.
  • It opens in a NEW tab, reliably. Browsers block a tab opened after an asynchronous request, so the tab could silently fail to open. The click now opens a blank tab immediately and sends it to the session when the launch call answers. If the launch fails the blank tab is closed and the reason is shown; if the browser refuses the tab an Open remote session link appears instead. A tab from an earlier launch is never reused (each launch has its own login token). Same for the Remote tab (Desktop, Terminal, Files), the legacy device page and the Tactical asset page.
  • Settings. Administration > Endpoint agent > MeshCentral: Hide MeshCentral interface elements (0 to 255, default 31; 31 = only the remote screen, 0 = the full MeshCentral and the old URL) and an optional value for Terminal and Files (empty = same; 0 allowed). Only shown once the database update has run.
  • MeshCentral does not connect by itself. From its source, nothing in the URL starts the connection: the new tab shows the Desktop view with MeshCentral's Connect button; click it once. The card says so.
  • Not verified on a live MeshCentral. The hide bits come from MeshCentral's page source (default.handlebars); the smoke runs against a mock Mesh server. See RivetCore's docs/rmm/REMOTE_AND_LOGS.md, "The Remote button", for the test to run on yours.
  • Tests: tests/rmm_ui_p68_remote.php (default hide on every view, the form, per-view values, 0 equals the old URL, refusals, a technician cannot change it, the helper is linked before the page scripts and nothing opens a tab after the async answer), new browser smoke tests/browser/rmm_rc17_remote_smoke.mjs (10 checks: a tab exists before the launch answers, the final URL has login, gotonode, viewmode=11 and hide=31, the tab cannot reach its opener, every click opens its own tab, the popup-blocked link, a failed launch closes the tab, hide 0, Terminal and Files, the admin card).

RivetMSP 26.10.22

v26.10.22
Release notes

[26.10.22] RivetMSP — the Run script dialog offers the script library, and one Scripts area

The bug. On an asset page the top Run script button opened a dialog whose "What to run" list held only "PowerShell text" (and "A saved script" when the older table had rows). The RivetCore script library (the Starter pack and your own scripts) was not in it, and Linux devices had no Run script button at all.

Deploy notes: back up the database, pull this tag. No database change (DB stays 2.6.93), no new cron line, no setting. The older addresses keep working.

  • One dialog, every source. "What to run" is now: Script library (N) (when the Scripts switch is on, you may run saved scripts and the library has scripts for the device's platform), A saved script (the older list, Windows, only when it has rows), Synced from (N) (older scripts that can run through the vendor RMM the asset is linked to, only with the older scripts grant) and PowerShell (type or paste) on Windows or Shell (type or paste) on Linux, only where Core supports that free-form job for your role. A single choice is shown as text, not as a one-option dropdown. Linux devices get the button.
  • Script library source. A search box (name, tag, description), the scripts grouped by risk class (Read only, Changes the system, Disruptive, Not classified), the chosen script's description, tags and risk badge, and its parameter form built from the script's schema (typed inputs, ranges and length hints, required fields, secrets as password fields, choices). Dry run (nothing is changed) is a tick box that starts ticked for a script that changes things (and follows the script's own default for a read-only one); unticking it really sends false. A script that needs approval says so and the button reads Request approval; a destructive script needs its confirmation box. It submits through the same handler as the Run a library script card of the Jobs tab (CSRF, client scoping, Core's checks), and the answer is a message with a link to the Jobs tab (or to Approvals). Empty states explain and link the Starter pack (administrators) or Scripts.
  • One Scripts area. The Endpoints menu had both Script library and Scripts. There is now ONE Scripts entry and one page with tabs: Library (the RivetCore library, the default) and Synced from (the older list, labelled with the vendor RMM's real name; shown when older rows exist or a vendor RMM that provides scripts is on). rmm_script_library.php redirects to the Library tab with its query string, rmm_scripts.php?category= opens the synced tab, rmm_script_run.php is unchanged. The older list is responsive now.
  • Import into the library. A one-time, idempotent action for the hand-made older scripts (no vendor script id): creates library scripts through Core's API as you (platform from the stored type, the old category as a tag, the risk class you pick, default Changes the system, approval only for Disruptive, note "imported from legacy Scripts"), skips what is already there with the same name and text, copies a name clash to " (legacy)", reports CMD and empty scripts as skipped, audits it, and leaves the older rows alone ("Imported" mark). Needs write access to scripts; Core checks you again for each script.
  • Wording. "Scripts" in the menu, titles and help text.

Tests: new tests/rmm_ui_p23_d_scripts_hub.php (dialog sources by platform and role, grouping, forms, dry-run default, approvals, destructive confirmation, client scoping, switch and module off at zero queries, synced scripts, one menu entry, tabs, older addresses, import and its idempotence) and tests/browser/rmm_run_dialog_smoke.mjs (desktop and 390 px, light and dark, keyboard); the existing RMM suites and smokes are updated for the single menu entry.

Release notes

The Service Desk ticket filters took one value each (one priority, one agent, one board). Every filter is now a group of checkboxes, so a saved view can be "High + Medium priority, assigned to Alex or Bo, on boards Network and Email", and an agent or an administrator can choose the view Tickets opens with.

Deploy notes: back up the database, pull this tag, php scripts/update_cli.php --update_db (2.6.92 to 2.6.93: ONE new table, ticket_view_defaults, CREATE TABLE IF NOT EXISTS, nothing existing altered; the code works before it runs, only choosing a default needs the table). No new cron line, no nginx change, no new composer package. Existing saved views and bookmarked URLs keep working unchanged.

  • Checkbox filters, OR inside a group and AND across groups. Assigned to (me / unassigned / any agents, any combination), Priority (Low / Medium / High), Board, Category, On-site / Remote (both ticked = no filter), Due (Overdue and Due today are independent; both ticked matches either) and Status (All open / All closed / Everything, or specific statuses). Each group is a fieldset with a legend, a count badge ("Any" when nothing is ticked) and Select all / Clear; every box has a label and works from the keyboard. The Save View form and the live quick-filter panel on the Tickets page use the same groups (the toolbar buttons open a checkbox menu; closing it or pressing Apply runs the filter; "More Filters" holds Assigned to, On-site / Remote and Due).
  • Old and new URLs. assigned=me, priority=High, board=5, onsite=1, status=Closed, status=7, overdue=1 and the dashboard's assigned=0 (nobody) still work and mean the same as the one-item lists. The canonical form is assigned[]=me&assigned[]=7&priority[]=High&priority[]=Medium. A stored view of either form opens as the canonical URL, and the highlighted view in the sidebar is decided by what the filters mean, not how the URL spells them. The viewer's own id and me are the same choice. status=All (the "All Tickets" box) now really shows every ticket.
  • One validator. ticketFiltersNormalize() turns any input into integers, enum members and fixed tokens (hostile values, nested arrays and oversized values are dropped; a group keeps at most 50 values), ticketFiltersSql() builds the conditions from that, and the list, the pagination and sort links, the kanban board, the client ticket lists and the API all read it. Kanban now honours Priority, Board, On-site, Due and Tags too (it used to ignore them). ticketViewDescribe() reads naturally ("Open tickets · Assigned to Alex, Bo or me · Medium or High priority · Board: Network or Email · Overdue or Due today").
  • Default view. Add / Edit view: Make this my default view. A star beside every view in the Ticket Views list sets or clears it in one click. Opening Tickets with no filter in the URL loads the default (as a redirect to the same page with the view's filters in the address bar, keeping the list / kanban display mode). Any filter in the URL, or a view link, wins. The list offers Reset to my default (?default=1) and Show all open (?status=Open). One personal default per agent and one shared default (set by an agent with support level 3, Make shared default in the view's menu, or the extra checkbox in the form) for agents who have no personal default; the order is personal, then shared, then the old behaviour. A default that was deleted, un-shared or can no longer be used is skipped gracefully; deleting a view forgets the defaults that point at it. Setting, replacing and clearing a default is audit-logged (type "Ticket View"). A client's own ticket list never gets the default.
  • Permissions tightened on the way. The Save View and Edit View modals now require write access to the support module, and the edit modal only opens for a view the agent owns (or a shared view for support level 3); /api/v1/ticket-views requires support read access and refuses a key locked to one client. Client portal logins never reach any of it (agent pages sign them out; API tokens are agents only).
  • API. GET /api/v1/tickets takes several values for priority, onsite, assigned (me / unassigned / ids), status_id, category_id, board_id and tag_id (comma list or repeated name[]; OR inside a filter, AND between filters; mine=1 and priority=high work as before; overdue + due_today together match either). GET /api/v1/ticket-views translates a multi-value view into those comma lists (all values still strings) and marks the agent's default with is_default. api/v1/openapi.yaml and docs/API.md updated.
  • Not changed. The CSV export (an unfiltered export of all tickets, as before), the sidebar counters and the stat boxes (whole-desk numbers), and the per-ticket live stream (sse_ticket_stream.php, which follows one ticket, not a list) do not take list filters.

DB and tests: DB 2.6.93; db.sql gains ticket_view_defaults (a fresh import and the 2.6.92-to-2.6.93 update produce the same schema, 301 tables compared, no uca1400; the step is idempotent). Tests: new tests/ticket_views.php (171 checks: old / new forms and the canonical round trip, hostile and capped values, every filter combination against an independent PHP filter on 24 seeded tickets, the form validator, descriptions, default resolution / set / clear / replace / archived / un-shared / deleted fallbacks, the real Tickets page, pagination and sort links, kanban, the redirect and its escapes, handler permissions, audit lines, CSRF, the modals, the API, client portal and client-locked key refusals) and a browser smoke tests/browser/ticket_views_smoke.mjs (14 checks: two priorities and two agents by checkbox, save as default, Tickets opens with it, an explicit filter overrides, Reset to my default, Show all open, kanban, the star toggle, Select all / Clear, keyboard, phone width, dark theme). tests/wave1_integration.php and tests/endpoint_agent_module.php follow the new step.

Release notes

Fixes the false-positive and repeating "sync stopped / sync failing" emails the Wave 1 recovery watcher produced on a live install (about twelve a day): UniFi sync has stopped: no run for 94 days (expected every 17 min) (UniFi now syncs from an external daily script that does not write unifi_sync_log; the old in-app scheduled runs were still used to infer the interval) and Sophos Firewalls: the last 3 runs all ended in an error (an external per-minute script looped over every enabled integration, including the built-in rivetit_agent type, and wrote that type's refusal, "Endpoint agent devices are managed from the device page, not through an RMM client.", against the previous integration's row). No RivetCore change: the pin stays at 1.0.0-rc.16.

Deploy notes: back up the database, pull this tag, php scripts/update_cli.php --update_db (2.6.91 to 2.6.92: ONE new table, recovery_watch_overrides, CREATE TABLE IF NOT EXISTS, nothing existing altered; the code works before it runs, only saving a mute or an expected interval needs the table). No new cron line, no nginx change, no new composer package.

  • Dormant integrations. An integration whose newest run is older than Dormant after (recovery setting sync_dormant_days, default 14 days, 0 = never) is dormant: no email, shown as "dormant since " with Mute and Forget history actions. At most ONE "no longer running" notice per 30 days, and only when its last run was good and it has just gone quiet (within a week of crossing the line): an integration that went quiet long ago, such as the 94-day UniFi case, produces nothing. A never-run integration older than the setting is dormant too.
  • Per-integration overrides (table recovery_watch_overrides: source, integration id, muted, expected interval in seconds, note, updated by/at). Admin > Backup > Integration sync watcher: Mute / Unmute, Edit (expected to run every N hours, a note, muted), Forget history. An external daily script can be declared "expected every 24 h"; a declared interval is judged at 3x and opts out of dormancy (a declared script that stops is stale). Every change is CSRF-protected and written to the audit log with who, which integration and the new values; Forget history needs its confirmation box and removes only that integration's sync-log rows and alert state.
  • Repeat cadence. The same problem is mailed at most once per 24 hours (new setting alert_repeat_hours, default 24, was a fixed 6; applies to every recovery alert, the e-mail footer shows the live value), the first occurrence is immediate, and a recovery ("sync is working again" / "running again") is announced once, only when the problem had actually been mailed. Stale and failing no longer both keep state for one integration (so no double recovery notice).
  • Failure rule. "The last N runs failed" counts only that integration's runs with a real error: a failed status with a message that is not the "Endpoint agent devices are managed from the device page" sentinel. Sentinel rows are not the integration's runs at all (they neither fail it nor keep it looking fresh). Integration types without a vendor API (rivetit_agent, SyncWatch::UNWATCHED_RMM_TYPES) are not watched.
  • Panel. A per-integration table: state (healthy / failing / stale / dormant / muted), last run (UTC), expected interval (and whether declared), the reason, actions. Names and notes are escaped; no inline handlers.
  • Docs. docs/RECOVERY_RUNBOOK.md section 2a: the watcher, and the rule for external sync scripts (write rmm_sync_log / unifi_sync_log, or declare the interval or mute; any script that copies the "sync all enabled integrations" loop must skip type rivetit_agent).
  • Other. SyncWatch::snapshot() (pure read, used by the panel) and SyncWatch::evaluate() (pure verdict); RecoveryAlerts::recovered(); RecoveryAlerts::raise() takes an optional repeat window (the dormant notice uses 30 days). New: src/Recovery/WatchOverrides.php, tests/recovery_watch.php, tests/browser/recovery_watch_smoke.mjs.

DB and tests: DB 2.6.92; db.sql gains recovery_watch_overrides (a fresh import and the 2.6.91-to-2.6.92 update produce the same schema, no uca1400; the step is idempotent). Tests: new tests/recovery_watch.php (80 checks: the exact live cases, UniFi with 94-day-old runs, Sophos with sentinel rows mixed in, a genuinely failing integration mailed once then not for 24 h, the 23 h / 24 h / 48 h boundaries, recovered notice once, a declared 24 h interval, mute, the 30-day dormant notice, forget history, the panel's rendering and escaping, the handlers with their audit lines and clamps, the session-time-zone guard) and a browser smoke tests/browser/recovery_watch_smoke.mjs (6 checks: real login, every state rendered, Edit / Mute / Unmute / Forget history through the real forms, phone width). tests/recovery_drill.php 125 (the 6 h window became 24 h, failed runs carry a message), tests/wave1_integration.php 87 and tests/endpoint_agent_module.php 87 (the 2.6.92 step and its idempotence), tests/rmm_ui.php 1755, tests/endpoint_agent_smoke.php 47, tests/rmm_installer_ui.php 141, golden replay 10, RivetCore PHPUnit 229 tests, and the security / crypto / platform / compliance / mail-intake suites all pass. Known failures, identical on 26.10.19 (verified on a clean checkout): sla_pause 2, recovery_integration 3, the strict golden replay.

Release notes

Adopts RivetCore 1.0.0-rc.16. Two owner requests. (A) The Mesh options live only in a stamped installer, so an agent that enrolled without them, or one that updated itself, never got a Mesh agent: a technician can now push the install (and the removal) from the device page, and install it on many devices from the fleet page. (B) The server could hand out only the stamped exe: it can now also hand out an .msi, built from an uploaded MSI template.

Deploy notes: back up the database, pull this tag, composer install --no-dev --no-plugins --no-scripts, php scripts/update_cli.php --update_db (2.6.90 to 2.6.91: Core migration 0025, one new table endpoint_agent_msi_templates; safe before or right after the code, and until it runs everything works, only publishing an MSI template needs the table). No new cron line, no nginx change. Agents: a pushed Mesh install needs the rc.16 agent (an older agent shows a disabled button, "update the agent first", and the API answers 409 unsupported); the MSI needs the rc.16 setup --from-msi inside the template. Nothing changes for the exe installer or for older agents. Build the template with RivetCore's make msi (release tag agent-v* attaches it with SHA256SUMS.msi) and upload it under Administration > Endpoint agent > MSI templates (or php scripts/endpoint_agent_publish.php <file>.msi --msi --version X --arch amd64 --activate).

(A) Push the MeshCentral agent

  • Device page, Remote tab (Remote access card): Install MeshCentral agent while MeshCentral is on with an https address and a device group and the device does not report the agent (disabled, with the reason, for an agent that did not announce job:mesh_install); Remove MeshCentral agent (page confirmation dialog with a box, a destructive job offered once, never retried; also offered after MeshCentral was switched off) and Run the install again once it reports the agent. The page follows the job result. The caller sends nothing but force: address, device group and pin come from the saved settings when Core creates the job, and the group id is never in the job row, the page or any result. agent/post/rmm_remote_tools.php actions mesh_install / mesh_uninstall.
  • Agent fleet page, bulk install: a collapsed card (every device I may manage, a client, a tag or a group; Count the devices, tick the confirmation, Install) that shows what Core queued and what it skipped (installed, queued, agent too old, no permission, error). Capped by bulk_run_max; refused above approval_bulk_threshold (the approval queue only knows script runs). agent/post/rmm_mesh_bulk.php, js/rmm_mesh_bulk.js. Never reaches a client the user may not manage.
  • Rules unchanged from Core: rmm.remote.manage and the remote-tools switch, idempotent (already_installed, already_queued), audited (Remote Tool Queued, Mesh Agent Bulk Install, Mesh Agent Installed / Install Failed).

(B) MSI installer

  • Format selector exe | msi in the Add device dialog (Agent fleet page and client page) and in Administration > Deployment; the msi option is disabled, with the reason, when no template is current for the architecture (Core builds x64 only; an ARM64 PC uses the exe). The download is the template plus the payload, never a change to the stored template, named RivetIT-Agent-<client>-x64.msi; the Also install the MeshCentral agent box works the same.
  • Deployment commands gain an msi block (msiexec, property form, silent downloader, uninstall, detection) for format msi.
  • MSI templates card: upload, list (version, size, SHA-256, where the platform was read from), make current, deactivate, reactivate; CLI --msi. Unsigned and UNVERIFIED on Windows: the card, the dialog and the commands say so (SmartScreen; whether msiexec accepts a package with the payload appended, repair, upgrade and the custom actions have never run on Windows; RivetCore docs/rmm/MSI_VERIFY.md is the checklist).
  • CSP safe (no inline handlers, no native dialogs), CSRF on every handler, labelled controls and live regions, tables scroll inside their card on a phone, nothing is drawn and no RMM query runs while the module is off.

Pin and DB: rivet/rivet-core ^1.0.0-rc.16 (lock reference cb27c541b6459b3be032bd410812328d0bbdb8bc), vendored copy identical to git archive of the tag (and installed.json / installed.php). DB 2.6.91 runs Core migration 0025 through the Core runner; db.sql gains endpoint_agent_msi_templates and the ledger row (a fresh import and the 2.6.90-to-2.6.91 update produce the same schema, no uca1400).

Tests: tests/rmm_ui.php 1755 checks (88 new, tests/rmm_ui_p16_mesh_push.php), tests/rmm_installer_ui.php 144 (59 new: the Format selector, upload and its refusals, the download with the real wixl package and the full package from make msi: OLE header, decodable trailer, stored template untouched; the Administration cards; the CLI), tests/endpoint_agent_module.php 86 (the 2.6.91 step, twenty endpoint_agent_* tables), tests/wave1_integration.php 85, a tenth browser smoke tests/browser/rmm_p16_smoke.mjs (13 checks: Remote tab install and remove with the in-page dialog, bulk, Format selector, template upload through the real file input, dark mode, phone width), and the real agent proof (98 checks, 27 new: a real rc.16 Linux agent under systemd enrolled WITHOUT Mesh is pushed mesh_install through the technician API against the mock Mesh server: wrong pin, unknown group, oversize body, the good case, node mapping, idempotence, bulk, force, removal). The agent's own Go reader decodes a PHP-stamped copy of the real make msi package. Known failures, identical on 26.10.18 (rc.15): sla_pause 2, recovery_integration 3, the strict golden replay, and check 14 of the rmm_p68_admin_smoke browser smoke.

  • Not verified: a real MeshCentral (the pushed install ran against Core's mock behaviour only; the Linux agent ids, install arguments and node-id locations are as unverified as in rc.12), and anything on Windows: the pushed install on Windows, the MSI (unsigned; msiexec with an appended payload, repair, upgrade, custom actions), the Windows agent.
Release notes

Adopts RivetCore 1.0.0-rc.15. A real MeshCentral (an owner's server) showed that its device group id (meshid) is not hex: it is 64 characters of base64 in MeshCentral's URL spelling (@ for +, $ for /; 48 raw bytes), with no mesh// prefix, and the download address is https://<mesh>/meshagents?id=4&meshid=<64 characters>&installflags=0. Core now accepts that, and RivetMSP's Mesh card follows. DB 2.6.90: one idempotent step gated on 2.6.89 that runs Core migration 0024_rmm_mesh_group_url through the Core runner (one added column, endpoint_agent_mesh_settings.group_id_url varchar(64) NOT NULL DEFAULT '', guarded by information_schema; nothing else altered; db.sql carries it and a fresh import equals a migrated database). Nothing is switched on by this update. Rows saved with rc.12 to rc.14 keep working (the URL form is derived from their stored hex). Ported from RivetIT 26.10.41 (same Core pin and the same behaviours; RivetMSP's own card and page structure). The warning is shown inside the Mesh card, once, instead of in the toast.

Deploy notes: back up the database, pull this tag, composer install --no-dev --no-plugins --no-scripts, php scripts/update_cli.php --update_db (2.6.89 to 2.6.90; safe before or right after the code, and until it runs everything works: only the verbatim URL form is not remembered). No new cron line, no nginx change. Roll out an agent built from RivetCore rc.15 with this server if you use the optional Mesh install: the installer payload's mesh.meshid is now the URL form (same member, payload version 1), which agents from rc.12 to rc.14 cannot use (they record a Mesh error; the RivetMSP install itself is unaffected). After the update, re-save the group id once if you want the verbatim form MeshCentral printed kept; reissue installers from the updated server.

  • Device group id field (Administration > Endpoint agent > MeshCentral). Help text: paste the download address from MeshCentral (device group, Add Agent, Windows 64-bit installer) or just its meshid; it is 64 characters, not hex; treat it like an invitation key. The field is no longer limited to hex: it takes the whole pasted address, the bare 64-character meshid, mesh//<64> and the old hex form, trims only whitespace and sends the raw value to Core (mesh_group_id) with room for a pasted address (2000 characters). Core refuses whitespace inside, quotes, &, #, a wrong length and another alphabet with one sentence and writes nothing; what was typed is put back escaped. The field stays a masked password input with its show toggle and now reads back mesh_group_id_url (the canonical hex stays in mesh_group_id). Warnings from the save response (today: a pasted address for a host other than the saved MeshCentral address) are shown in the card after Save ("Saved, with a warning: ..."), once, role="alert".
  • Certificate pin. Help text: leave empty when the server certificate is issued by a public CA such as Let's Encrypt (a certificate that renews every 60 to 90 days would break a pin); use a pin only for a private CA or a self-signed certificate. Read it from the server shows Core's pin_probe.note and, when pin_probe.publicly_trusted is true (a second, verified handshake succeeded), shows a "publicly trusted: no pin needed" panel and offers no way to put the hash in the pin field; for a self-signed certificate the proposal, the note and the button are as before.
  • Core's note under Advanced now says what was checked against a real MeshCentral (the download address shape, agent id 4) and what was not (Linux agent ids, install arguments, node-id locations).
  • Installer payload and Linux command. mesh.meshid and the Linux --mesh-group-id carry the URL form (verbatim as pasted, else derived from the hex); install-linux.sh quoting ('...') is Core's, so the $ in an id is safe.
  • Pin: rivet/rivet-core ^1.0.0-rc.15 (lock reference b930c50063823d1ee324216b7612e518bbb87998), vendored copy identical to git archive of the tag; tests/wave1_integration.php follows the pin and the new step (82 checks, was 79).
  • Tests: tests/rmm_ui.php 1667 checks (was 1617; the Mesh fragment now covers the bare meshid, the whole pasted address for the saved host and for a different host (warning), hex, mesh//, whitespace trimming, 19 hostile or malformed values refused with nothing written, escaping of a hostile value put back, clearing, a row from before migration 0024, the new help texts, the self-signed probe note and the publicly-trusted panel, end to end through Core's verified handshake against a LOCAL CA (a second php -S started with curl.cainfo = that CA, plus the control that the same server is not trusted by the first web server; this proves the wiring, not Let's Encrypt itself; tests/support/tls_stub_server.py answers the HEAD that handshake sends)), tests/rmm_installer_ui.php 85 (payload decode shows the URL-form meshid, verbatim for a pasted address with @ and $), tests/endpoint_agent_module.php 84 (the 2.6.89 to 2.6.90 step, idempotent, equals a fresh db.sql import), the Mesh browser smoke 19 checks (was 15: the same forms through the real page, the warning, hostile values, and the payload decoded from a real installer download), the other eight browser smokes, golden replay, real agent (12) and real P2/P3 agent (30) suites with an agent built from the rc.15 tag, phpunit (229 tests) and the regression list. Known results, identical on 26.10.17 (rc.14, checked by running the same suites on that tree in the same environment): sla_pause 2, recovery_integration 3 (the 600000-iteration drill archive), browser rmm_p68_admin_smoke check 14 (queue-worker guidance text, environment dependent), and endpoint_agent_golden.php strict reporting only the four intentional 01-disabled differences (a disabled module answers 503 module_disabled; golden replay, which allows exactly those, is clean); rmm_smoke check 25 and a HQ substring in a CSRF token on the fields page of rmm_ui_p23_a_policies.php are intermittent. The real-agent proof (tests/real_agent_proof/run.sh, docker, a real systemd agent built from the rc.15 tag, a MOCK Mesh server) passes with 71 checks. The real-agent proof harness (tests/real_agent_proof) now uses a fresh random URL-form group id. Every group id in the tests is generated at run time or a fixed fake; no real MeshCentral address or id is in the repository.
  • Not verified: only the Mesh download address shape and agent id 4 were checked against a real MeshCentral (by Core, not here); the Linux agent ids, the install and uninstall arguments, where the node id can be read, and everything on a Windows host are not. Nothing was installed from a real Mesh server and nothing was deployed.
Release notes

Small patch release. Adopts RivetCore 1.0.0-rc.14, which fixes Cannot use object of type stdClass as array from AlertingActions::getThresholds: on an install that never edited its global checks (the built-in list, where pending_reboot has empty params) the thresholds part of the Alerting tab of an agent-linked device failed, was logged, and showed the "declares no thresholds" note for every check. Inside Core a check definition's params is now always an array; \stdClass exists only on the wire (ChecksValidator::toWire()), so the stored checks_json, the signed check messages and the policy version hashes are byte for byte unchanged (the golden transcripts still replay identically, 10 files). No migration: DB stays 2.6.89.

Deploy notes: back up the database, pull this tag, composer install --no-dev --no-plugins --no-scripts. No update_db step, no new cron line, no nginx change, no agent change (an agent built from rc.13 keeps working; the fix is server side).

  • Edition code audited, no change needed. Nothing in includes/rmm_*.php, agent/rmm_*.php, agent/post/rmm_*.php, js/ or src/Core/Adapter/Endpoint/ checks instanceof \stdClass, json-encodes a check list or reads params as an object; the thresholds editor and the policy viewer already treat params as an array, and the administration checks editor prints Core's wire-shaped settings summary ({}, asserted below).
  • Pin: rivet/rivet-core ^1.0.0-rc.14 (lock reference fae6d98e63669294ed1bbf6719d082b1bedeb600), vendored copy identical to git archive of the tag. tests/wave1_integration.php follows the pin (79 checks).
  • Regression test in tests/rmm_ui_p23_c_alerting.php: with no checks_json stored, a device that reports disk_c, pending_reboot, svc_eventlog and agent_update loads the asset page Alerting tab; the technician thresholds route answers for each defined check, every default check lists in the thresholds part, the server log holds no stdClass / Cannot use object / fatal line, and the administration checks editor still shows "params": {}. Run against the rc.13 vendored Core the same checks fail (pending_reboot params is an object, thresholds answer 500, the error is logged), so the test is not vacuous.
  • Tests: tests/rmm_ui.php 1617 checks (18 new), nine browser smokes, the RMM, golden (replay identical, strict identical), real agent and phpunit (229 tests) suites and the regression list all as on 26.10.16. Known failures, identical on 26.10.16 (rc.13 Core): sla_pause 2, recovery_integration 3 (the 600000-iteration drill archive), browser rmm_p68_admin_smoke check 14 (queue-worker guidance text, environment dependent); rmm_smoke check 25 (tag removal after reload) is intermittent. Nothing was deployed.
Release notes

Adopts RivetCore 1.0.0-rc.13 (rc.12: the optional MeshCentral agent install with automatic node-id mapping, migration 0023; rc.13: the RMM starter pack). Aligned with RivetIT 26.10.39 (same page name agent/rmm_starter_pack.php, same handler, same Mesh box rule: offered only once MeshCentral is on with an address). DB 2.6.89: one idempotent step gated on 2.6.88 that runs Core migration 0023_rmm_mesh_agent through the Core runner (three new tables, nothing existing altered; db.sql carries it and a fresh import equals a migrated database). The starter pack has no migration. Nothing is switched on by this update: no installer asks for the Mesh agent until an administrator saves the Mesh install settings and ticks the box, and nothing from the starter pack is installed until an administrator installs it. Older agents are answered exactly as before (golden transcripts replay identically, strict mode included).

Deploy notes: back up the database, pull this tag, composer install --no-dev --no-plugins --no-scripts, php scripts/update_cli.php --update_db (2.6.88 to 2.6.89; safe before or right after the code). No new cron line, no nginx change. To use the Mesh install, roll out an agent built from RivetCore rc.13 (an older agent never reports Mesh and ignores the installer members).

  • Optional MeshCentral agent (off by default). The Add device dialog and the Administration deployment form get Also install the MeshCentral agent (unticked every time, disabled with Core's own refusal sentence until Mesh is switched on with an https address and a device group id). Administration > MeshCentral gets the install settings: device group id (masked, it is a secret), certificate pin with Read it from the server (shows the SHA-384 and SHA-256 of the server's certificate and never saves; a button puts it in the field and you press Save), the overwrite switch, and JSON corrections of the agent-id and install-argument tables behind a warning that they are unverified. Core validates everything and writes nothing when any field is invalid. The device's Remote access card (and the Remote tab) shows the Mesh agent state: installed with its version or the install error, and what the server did with the node id (mapped; conflict, a visible alert that names the other device only when it belongs to the same client; kept_manual, a visible hint). The node id itself is shown to administrators only. Events rmm.mesh.agent_installed and rmm.mesh.node_conflict are in the shared picker.
  • RMM starter pack. Endpoints > Starter pack (administrators, scripts sub-switch): browse and filter 73 scripts (41 Windows, 32 Linux; 57 read only, 10 change the system, 6 disruptive), 3 check sets and 3 policy drafts (created switched off and unassigned), tick what you want, Preview, Install selected, Uninstall selected (preview first, then an in-page confirmation), a settings form per draft (time server, SSH unit, certificate paths, backup service...), Core's signing-key blocker shown first. An empty selection is never sent to Core (Core reads it as the whole pack). The script library marks starter scripts (Starter badge, risk chip, Update waiting for a parked newer version, with a link that compares it with yours); policies show a Starter badge and never print their marker, and saving a draft keeps it. Event rmm.starter.installed.
  • Tests: tests/rmm_ui.php 1599 checks (new fragments rmm_ui_p1213_mesh.php, rmm_ui_p1213_starter.php), tests/rmm_installer_ui.php 81 (payload decode: the Mesh members only when ticked, token-gated download, Linux flags), the pin probe against a local TLS server, auto-mapping from real check-ins (conflict, kept manual, overwrite), two browser smokes (Mesh 15 checks, Starter 16; desktop and phone, light and dark), the real Linux agent proof (71 checks, was 39) extended with a MOCK Mesh server and the starter pack through the API.
  • Not verified: anything that talks to a real MeshCentral server (download address and flags, agent ids, install and uninstall arguments, where the node id can be read), the Windows code on a Windows host, and all 41 Windows starter scripts (parse-checked by Core, never run). Nothing was deployed.

All RivetMSP releases on GitHub

v0.9.8

v0.9.8
Release notes

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.7...v0.9.8

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.7...v0.9.8

v0.9.7

v0.9.7
Release notes

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.6...v0.9.7

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.6...v0.9.7

v0.9.6

v0.9.6
Release notes

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.5...v0.9.6

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.5...v0.9.6

v0.9.5

v0.9.5
Release notes

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.4...v0.9.5

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.4...v0.9.5

v0.9.4

v0.9.4
Release notes

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.3...v0.9.4

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.3...v0.9.4

v0.9.3

v0.9.3
Release notes

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.2...v0.9.3

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.2...v0.9.3

v0.9.2

v0.9.2
Release notes

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.1...v0.9.2

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.1...v0.9.2

v0.9.1

v0.9.1
Release notes

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.0...v0.9.1

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.9.0...v0.9.1

v0.9.0

v0.9.0
Release notes

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.8.10...v0.9.0

What's Changed

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.8.10...v0.9.0

RivetIT-Mobile 0.8.8 beta

v0.8.8 Pre-release
Release notes

Android beta 0.8.8 improves the ticket creation result.

  • Open the new ticket directly after the server confirms creation and returns its ID.
  • Clear cached ticket responses after a confirmed create so the ticket list refreshes when users return.
  • Keep the draft and report the error when creation is not confirmed.

Validation: Android debug unit tests, lint, and APK assembly passed. An Android 36 tablet emulator using fictional data verified an HTTP 503 left the draft on screen, a 201 retry opened ticket #91, and Back showed #91 in a list that had previously cached an empty response. Live backend permissions and notification effects remain unverified.

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.8.7...v0.8.8

RivetIT-Mobile 0.8.7 beta

v0.8.7 Pre-release
Release notes

Android beta 0.8.7 improves ticket creation pickers for installations with many departments.

  • Search departments by name on the server, including those beyond the first API page.
  • Show loading, failure, and Retry states for department and category pickers.
  • Make both pickers easier to browse in expanded, scrollable sheets.
  • Add tablet mock-API QA evidence to the issue #9 screen audit.

Validation: Android debug unit tests, lint, and APK assembly passed. A 1600×2560 Android 36 emulator verified a department beyond the initial 30 and category failure/retry with fictional data. Live ticket creation and restricted-role behavior still need verification.

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.8.6...v0.8.7

RivetIT-Mobile 0.8.6 beta

v0.8.6 Pre-release
Release notes

Worksheet and outtake creation now waits for server confirmation before closing the sheet or navigating. While a request is pending, repeat taps are blocked. A failed request stays visible with an inline error, and an unconfirmed result directs the user to check the ticket before retrying. The worksheet picker now distinguishes loading, empty, and failed template requests and offers Retry.

Verified on a 1600×2560 Android 36 emulator against a temporary HTTPS mock with fictional data: each pair of quick taps sent one POST; 503 responses kept each sheet open; 201 retries opened the new forms. Unit tests, lint, and normal debug assembly passed. Screenshot QA is disabled in the published APK.

Issue #9 remains open for live-backend, restricted-role, accessibility, and broader device checks.

Full Changelog: https://github.com/TheTractorHacker/RivetIT-Mobile/compare/v0.8.5...v0.8.6

All RivetIT Mobile releases on GitHub

RivetMSP Android v1.34.0

v1.34.0
Release notes

What's Changed

New Contributors

Full Changelog: https://github.com/TheTractorHacker/rivetmsp-mobile/compare/v1.33.0...v1.34.0

RivetMSP iOS 0.1.0

ios-v0.1.0 Pre-release
Release notes

First RivetMSP iOS build. Unsigned and not yet run on a device or simulator: the .ipa must be signed with your own Apple team to install, and the simulator .app.zip is for a Mac. APNs push and an iPad layout are not included.

RivetMSP Android v1.31.0

v1.31.0
Release notes

RivetMSP Android replaces the legacy application branding with the approved RivetMSP identity.

Uploaded MSP company logos replace product artwork in sign-in, app chrome and the biometric lock screen. Branding is cached per server, refreshed on resume and reset when switching servers. A configured logo that cannot load shows the company name rather than a competing product logo. Launcher and OS splash resources use bundled RivetMSP artwork before a server is selected.

Company-logo synchronization requires RivetMSP Web v2.46.0 or newer. Existing authentication, APIs, package IDs and stored preferences are preserved.

The production APK uses the existing production signing identity and application ID com.foleyit.itflow, version 1.31.0 (code 40). It requires Android 14 or newer. The beta APK uses com.foleyit.itflow.beta and installs separately.

Validation: all 26 local JVM tests passed; local Android lint completed with zero errors. GitHub Actions passed lint, unit tests, the beta build and the signed production build. The downloaded production APK signature matches v1.30.1. No physical-device/emulator visual test was performed.

This release covers branding and company-logo precedence. The wider ticket-workflow UX redesign is not included.

Downloads

Release notes

Step 4 of the redesign — motion and the new brand identity.

  • Fade + subtle-rise screen transitions using the design reference's springy easing, wired once at the NavHost level so every screen in the app gets it.
  • A new reusable press-scale interaction (works with any Button/Card/FAB) applied to the floating bottom nav, all 4 FABs, and Dashboard's quick-action buttons.
  • New F-monogram logo mark replacing the old sync icon everywhere: top bar, drawer, both auth screens, and the launcher icon. Canvas-drawn rather than a static asset so its accent dot tracks whichever color seed is active.
  • Fixed an icon-contrast issue across all three logo-tile locations: the mark now uses the theme's onPrimary role instead of a hardcoded white, since this app's dark-mode seed colors are deliberately light/bright and white would wash out for several of them.

Scope note: press-scale went on the highest-visibility touch points, not retrofitted onto every button/card across all ~35 screens. Sheet and drawer motion already come from Compose Material3's own defaults.

Full Changelog: https://github.com/TheTractorHacker/itflow-msp-app/compare/v1.26.0...v1.27.0

Release notes
  • Gradient logo tile on both auth screens (matching Profile's gradient avatar), with a color-role fix caught in review so the icon stays legible in dark mode across all 5 seeds.
  • Login: monospace/letter-spaced 2FA code field; a distinct "Waiting for passkey…" state on the passkey button.

Server Setup's real certificate-pinning security (probes the actual server cert on an untrusted-cert error, requires typing the fingerprint's last 6 characters to trust-on-first-use) and Login's real passkey/WebAuthn sign-in via CredentialManager are both verified byte-for-byte unchanged.

This completes the screen-by-screen restyle pass across the whole app (v1.16.0 → v1.26.0): global chrome, Dashboard, Tickets, Clients, Assets, Appointments, Notifications/Alerts, Profile, Reports, Knowledge Base, Credentials/Quotes/Invoices/Expenses/Worksheets, and Auth. Motion/polish and a final pass verifying against the design reference are what's left.

Full Changelog: https://github.com/TheTractorHacker/itflow-msp-app/compare/v1.25.0...v1.26.0

Release notes
  • Credentials: shape-token consistency on the list; detail screen's password generator and notes split into their own clearly-labeled cards.
  • Quotes/Invoices: status-colored trailing pill wired to the shared financial-status color mapping wherever it was missing; shape-token consistency.
  • Expenses: shape-token consistency on the list; Add Expense's fields grouped into cards matching Create Ticket's pattern.
  • Fill Worksheet / Outtake Sign: field/content grouped into cards, sticky bottom action bars.

Verified in review by diffing the specific functions directly (not just trusting the summary): Credential Detail's signed biometric-challenge flow and Outtake Sign's custom finger-drawn signature canvas are both byte-for-byte unchanged — only surrounding layout moved.

Full Changelog: https://github.com/TheTractorHacker/itflow-msp-app/compare/v1.24.0...v1.25.0

v1.24.0
Release notes
  • List: card shape consistency.
  • Detail: added a category tag chip above the article content (the data was already fetched, just never displayed).

The WebView-based rich-content renderer and its http(s)-only external-link allowlist are unchanged — real functionality beyond the design mockup's plain "body text" description, and deliberate security code.

Full Changelog: https://github.com/TheTractorHacker/itflow-msp-app/compare/v1.23.0...v1.24.0

All RivetMSP Mobile releases on GitHub