Release v0.104.0
Aug 07, 2026
Quick Support brings agentless remote sessions with a one-time code, high-stakes AI actions now require a second approver, third-party patches get auto-approval on update rings, and locked-file backup genuinely works — plus a deep round of reliability fixes. Self-hosters: standard upgrade, no new required settings.
Remote Management
- Quick Support: help a machine that has no agent installed. Generate a one-time code, the user runs a small support client, and you get a single remote session on a temporary device record — the client removes itself when the session ends.
- Starting a remote desktop session no longer occasionally trips over itself when several session requests land at once.
- Session expiry is handled gracefully across the app: a clear notice on the login page, an idle warning before you're signed out, and you land back on the page you were on after signing in again.
AI & Automation
- High-stakes AI actions — payments and invoicing, tenant changes, Microsoft 365 and Google identity changes — now require approval from a second administrator (four-eyes). Routine device work stays one-click for the requesting technician.
- The AI assistant keeps answering questions while an action is waiting on approval, instead of going quiet.
- Billing and catalog questions work in chat again — the assistant's invoice, quote, and catalog tools were unreachable from the in-product chat.
- AI usage and cost tracking now count correctly for partner logins and include cached-context usage.
- Script output from automation runs is saved and shown in Run History.
Patch Management
- Third-party application updates can be auto-approved per update ring, with their own severity gating and an optional hold period after release.
- The full patch catalog is now pageable — fleets with more than 200 pending patches can view and bulk-approve everything.
Backup & Recovery
- Backing up locked files on Windows genuinely works now — the snapshot mechanism had never functioned. Verified end-to-end on real hardware this release.
- A new snapshot-health panel on each device's Backup tab shows per-writer status and any unprotected volumes.
- Backup jobs with many unreadable files are marked Partial instead of reporting success.
- If the backup helper dies mid-run, the job fails within seconds with a clear reason instead of hanging for ten minutes.
Visibility & Monitoring
- WAN IP and LAN IP columns on the devices list — opt-in, sortable, and persistent.
- Devices that re-enroll with an existing hostname are flagged as possible replacements instead of being rejected, so reimaged machines enroll cleanly.
- Unreachable SNMP devices are polled less and less often until they respond, instead of being hammered every minute forever.
- The server now tells apart a stalled application thread from a database problem, and its health metrics say which one is happening.
Platform & Infrastructure
- Large software packages upload in chunks with a real progress bar and mid-flight cancel.
- Accepted quotes get order tracking: see which lines still need a vendor order, mark them ordered and received, and export the list.
- Long-running scripts are no longer cut off at a flat ten minutes — the timeout you set on the script is the timeout that applies.
- The readiness endpoint reflects recovery immediately after a database or Redis blip, without a restart.
Security & Compliance
- Partner-wide scripts can now only be created, changed, or deleted by users with access to all of your organizations — technicians scoped to specific customers can use them but not manage them.
- Dependency security updates across the stack, including fixes for two recently published CVEs.
- Error reporting can no longer include a live API key in performance traces.
- Agent binaries no longer contain text patterns that tripped some antivirus heuristics.
Version 0.104.0 is about two things: helping machines you don’t manage yet, and trusting the platform more on the ones you do.
Quick Support is the headline. Someone calls with a broken laptop that has no agent on it — generate a one-time code from the Remote page, they run a small support client, and you’re in a real remote session. The device shows up as a temporary record that cleans itself up, the code works exactly once, and the client removes itself when the session ends. No install, no enrollment, no leftovers.
On the AI side, high-stakes actions now get a second pair of eyes. Payments, invoicing, tenant changes, and Microsoft 365 or Google identity changes require approval from a second administrator before they run; routine device work — run a script, restart a service, install a patch — stays a one-click approval for the technician who asked. If you’re a one-person shop, you can approve your own high-stakes actions with a hardware-backed passkey; register one under Settings → Profile before you need it.
Backup deserves an honest note. The mechanism that captures locked files on Windows had never actually worked — this release fixes it at the root, and we verified it end-to-end on real hardware, along with a new snapshot-health panel and a Partial status for jobs that skip too many files. The same hands-on testing also found three remaining issues with very large or permission-restricted file sets, which are documented with workarounds in the GitHub release notes. We’d rather tell you where the edges are than let you find them.
Self-hosting: this is a standard upgrade — pull the new images and restart; the 19 database migrations apply automatically and there are no new required settings. Two things to know: settings with typo’d values for IP allowlisting or agent auto-promote now stop the server with a clear error instead of silently falling back to a default, and alert rules built on the long-retired “custom” condition type (which could never fire) are cleaned up during the upgrade. Full details in the GitHub release notes.