Skip to content
← Back to release notes
v0.121.0 stable breaking major release

Release v0.121.0

Oct 02, 2026

One home for every AI model decision: AI Providers & Models settings, a chat model picker, bring-your-own OpenAI-compatible models with tool use, automatic failover, and AI usage chargeback to your clients. Backup storage credentials are now encrypted at rest, Quick Support works for standard Windows users, and two security issues are fixed. Upgrade recommended.

AI Models & Providers

  • Partner Settings → AI Providers & Models is now the single place for your AI connections, the models your techs may use, and the default model for each AI feature. Org Settings → AI → Model defaults can narrow that per client, and a blank field shows the value it inherits.
  • Techs pick the chat model and its options, such as effort and thinking, from the models you allow. Switching model mid-chat continues the same conversation when it fits; otherwise Breeze offers a linked new chat that starts from a summary. AI agents pick their model the same way.
  • Bring your own model: connect vLLM, LiteLLM, Ollama, OpenRouter or any other OpenAI-compatible endpoint, discover or enter its models, verify them, and use a verified model on every AI feature, including tool use.
  • Automatic failover: when a model is unavailable, rate-limited, overloaded or out of credit, AI calls move down an ordered backup list of up to five models. Moving between Breeze credits and your own API key only happens if you allow it.
  • AI agent runs use a model per stage (triage, analysis, remediation), and a run with no usable model is marked blocked and reported once a day instead of failing.
  • A residency switch disables every AI feature whose model can't guarantee data residency, and lists them before you confirm.
  • Every AI feature (chat, topology, the helper, script builder, Office, ticket drafts, the script reviewer, AI agents) is now priced from the same model rates and recorded per call, and spend budget alerts cover all of them.
  • When a model refuses a request, chat shows why, suggests other models and links to the docs instead of returning an empty answer.

AI Usage & Chargeback

  • Bill clients for the AI usage Breeze meters. On a billing profile, choose whether AI usage is billable, included or not billed, then add a markup or a per-model price list. Usage closes monthly into invoice lines.
  • A new AI usage by client business report, with PDF export, shows what each client used.
  • Nobody is billed by default, and usage from before this release is never billable.
  • Settings → AI Usage breaks spend down by model, feature, technician and client, with refusal rate and failovers, and a new Quality view sits beside Spend by model, feature or model family.

Remote Support & Agents

  • Quick Support can be started by standard (non-admin) Windows users. Each session keeps its files in its own folder, and the user sees an on-screen indicator while a technician is viewing their screen.
  • Breeze Viewer's in-app update on Windows installs again. Viewers stuck on an older version update on their next launch, and a failed update now says which stage failed.
  • Breeze Assist trusts certificates in the operating system's store, so self-hosted servers on a private certificate authority connect.
  • Typing quickly or holding a key in a remote terminal no longer ends the session.
  • Antivirus detection now recognises Emsisoft, Webroot, ThreatDown and WithSecure (including F-Secure).
  • Linux devices with an encrypted root on LVM, the default Ubuntu and Debian encrypted install, report as encrypted.
  • The enrollment rate limit is configurable for mass deployments behind one shared address.

Backup & Recovery

  • Backup destination credentials are now encrypted at rest. Existing destinations are encrypted automatically on the first start after the upgrade.
  • Windows backups capture folder junctions, such as a redirected Music folder, and recreate them on restore instead of skipping them.
  • Crash-consistent Hyper-V backup handles VMs that are off, saved or paused, and leaves them in the state it found them.
  • The backup readiness score no longer drops while verifications are still running.
  • Cancelling a queued bare metal rebuild closes the recovery so the device can be rebuilt again, and deleting a destination with job history explains what blocks it.

Fixes Across the Console

  • Personal API keys no longer stop working when their creator signs out. A password change, MFA change or account status change still invalidates them. Keys that stopped working on v0.120 need to be replaced.
  • Built-in alert rules can no longer be created twice. Existing duplicates are merged once at upgrade.
  • Network topology's overview reads like a network map: one card per site network, its gateway above the LANs behind it, and a tile for every device.
  • Patches that installed but await a restart show Installed, reboot required, and a per-device install supersedes an older failed job.
  • Vulnerabilities marked Ready now always mean Remediate will install a patch on that device.
  • Add Device reuses one installer enrollment key per site instead of creating a new one every time.
  • Invoice PDFs and emails include a unit price column, invoices without online payment no longer show a Pay button, and a sent quote uses the client's current tax rate.
  • Webhook settings no longer go blank after deliveries, signed automation webhooks no longer return Automation not found, and Xero and QuickBooks show a rejected customer or item once with the provider's reason.
  • Deleting a site that still has devices explains why it can't be deleted, and custom fields are listed alphabetically everywhere.

This release gives you one place to decide how Breeze uses AI. AI Providers & Models brings your connections, the models your techs may use, and the default model for each AI feature onto a single settings page, with per-client defaults on top. Techs can pick the chat model themselves, you can connect your own OpenAI-compatible model server (Ollama, vLLM, LiteLLM, OpenRouter) and use it everywhere, tools included, and when a model goes down or runs out of credit, calls fail over to the next model on your list.

Because every AI feature is now metered the same way, you can also bill it. Set AI usage on a billing profile to billable, included or not billed, add a markup or a per-model price list, and usage closes into invoice lines each month. Nothing is billed until you turn it on. Beyond AI, backup storage credentials are now encrypted at rest, Quick Support works for standard Windows users with an on-screen viewing indicator, Breeze Viewer on Windows updates itself again, and there’s a long list of fixes across patches, billing, topology and alerts.

Self-hosters, read before you upgrade: this release fixes two security issues, GHSA-vjqq-fmr4-m576 (High) and GHSA-7xq8-mcq6-j438 (Low), so upgrade from 0.120.0 or earlier. It is roll-forward only: once backup credentials are encrypted on first start, don’t roll back. Installs using MCP_LLM_PROVIDER=openai-compatible must set IS_HOSTED=false explicitly and can no longer point at localhost. Don’t skip this release on the way to v0.122, which removes the legacy AI model settings. The GitHub release notes have the full upgrade detail.