Ozark-Connect/NetworkOptimizer v2.7.0 on GitHub

On-Site Agent update available. If your agents monitor gateways, updating picks up automatic fan speed polling and charting on Device Stats. Optional: existing functionality is unaffected without the update.

v2.7.0 brings topology-aware firmware upgrades to Network Optimizer. Plan a rollout for your entire site: waves ordered by depth, per-SKU canaries, a post-soak report, and Autopilot for the sites you trust. Also in this release: ISP Health incidents with PDF export, Live View playback showing clients as they actually were, and client-outcome evidence in channel recommendations. If you were running the preview, thanks for testing - and see "Coming back to stable" below for switching back to the stable tag.

What you missed on v2.6.x

If you're upgrading from v2.6.0, the four patches since then carried some significant work:

ISP Health - loaded loss is now attributed to the upload or download phase that caused it, instead of being reported as ISP loss.
Cable Modem Stats - Technicolor CGA Series and Vodafone Station (ARRIS TG3442DE) support.
Cellular tower distance and DNS Lookup in Network Tools, plus a Reset Radio button for cellular modems.
Wi-Fi Optimizer - the per-floor load imbalance score measures about each AP's own client mean rather than a flat share, so a high-density AP is not penalized for carrying more clients by design.
Monitoring charts sync their hover crosshairs across every chart on the tab, and links between tabs carry where you came from so the back button works.

See the v2.6.4, v2.6.3, v2.6.2, and v2.6.1 release notes for the full detail, and v2.6.0 for the full v2.6 feature set.

Firmware Rollout

UniFi Network updates every device at once: no ordering, no canaries, no record of what happened. For a home network that's fine. For a 30-device site with a maintenance window, or an MSP managing remote sites where a bad firmware means a truck roll, Firmware Rollout plans the upgrade the way you would - except it knows the topology.

Topology-aware waves - your site upgrades outer-to-inner by depth, so the switch a device depends on never reboots while that device is mid-upgrade. APs that share coverage are spread across waves so a floor's Wi-Fi is never entirely down.
Per-SKU canaries - one device of each model goes first and soaks. If it comes back wrong the rest of that model are held: fourteen U7-Pro-XGs and the first one comes back bad, the other thirteen wait.
Autopilot - runs unattended on the sites you trust, with a heads-up alert before it starts. For the finicky client site, per-wave approval keeps a human in the loop.
Drift detection - if new firmware lands after you've scheduled a rollout, the page names the delta and offers a Re-plan that reopens the wizard with fresh data. You can also kick off a scheduled rollout early with Deploy Now.
Post-soak report with PDF export - per-device before/after CPU and memory, gateway downtime, console version from-to. Something to attach to a change record or hand a client.
SSH fallback for console updates - if the console doesn't see an update through its API, the rollout falls back over SSH with a best-effort backup first. Cool-down is 5 minutes for APs and switches, 10 for gateways.
Shared firmware catalog - firmware builds seen by any site on the same channel are available to every other site. Ubiquiti progressively ungates EA and RC firmware, so one console might see a build days before another on the same channel. The rollout planner now draws from the combined catalog across your install, so you get what your channel entitles you to without waiting for Ubiquiti's per-console rollout.
Alert suppression - device offline/recovered and Console connection alerts stay quiet during the rollout: per-device while it upgrades, site-wide while the Network application or UniFi OS is updating, and for five minutes after so everything can reconnect.

UniFi OS on a self-hosted UniFi OS Server is deliberately left alone: the Network application and your devices are what it manages.

Monitoring

ISP Health

Incidents - each loss spike, latency shift, or path change is tracked as a one-line incident with the paths it touched. Export the view as a PDF with transit weights: when the ISP says they don't see anything on their end, this is what you send them.
Loaded loss and latency recommendations now account for how your SQM rates compare to what the line actually delivers - when nominal rates are already close to measured throughput, the advice says so rather than telling you to keep lowering them.
Brief bufferbloat under load no longer counts against your ISP - a download that saturates your line for a few minutes was being classified as ISP congestion and penalizing the score. ISP Health now recognizes the uniform latency floor as your own access link under load and suppresses it from scoring.
LAN/Gateway outages classify consistently across all time ranges - a brief outage where the gateway went dark could read as a whole-WAN outage on the 30-day view while the 7-day view correctly showed LAN/Gateway. Now consistent regardless of the window.

Live View

Playback shows the clients that were actually connected at the rates and link speeds they were running. Scrub back to the time someone reported a problem and see the network as it was, not as it is now.
Each WAN's globe shows its own latency during playback. On a multi-WAN site you can tell whether a complaint lines up with one circuit degrading or with something inside the LAN.
Devices you place stay where you put them - through roaming, through being offline, and through a scrub.

Device Stats

Fan Speed - gateway fan speed (RPM) is now polled automatically and charted alongside temperature, CPU, and memory. If you had a custom OID configured for this with the field name fan_speed_rpm, it's absorbed into the standard chart on upgrade.

Cable Modem Stats

Technicolor CGA4233 support (#1152).

Network Tools

Gateway Diagnostics and Console Support File - run diagnostics against the gateway or generate a support bundle for Ubiquiti, from within Network Tools (#1129). Saves logging into the console separately, especially across managed sites.

Wi-Fi Optimizer - Channel Recommendation

Channel history follows the spectrum, not the control channel - at 5 and 6 GHz a bonded channel covers the same spectrum whichever number it goes by. Time on 149 and time on 157 count together now, so a wide channel gets recommended on everything actually measured there.
Client outcomes can block a move - where you have monitoring history, the PHY rates your clients were negotiating on the current channel (which fold in signal, noise and interference at once) weigh into the recommendation. A channel that measured clean on the radio side but ran your clients slower will not be recommended.
Old evidence fades instead of vanishing - a channel whose record had aged out was treated as never measured, which is more forgiving than known-mediocre. On a crowded band where things change fast, older readings count for less; on a quiet band they hold their weight.
Recommendations are computed once an hour rather than on every visit. The card shows when the plan was last computed, and Recompute rebuilds it on demand.

Security Audit

Isolated management networks missing firmware download access are flagged. The check already covered cloud management, and now covers fw-download.ubnt.com too.
LAG parent ports are no longer falsely flagged in Security Audit or Config Optimizer (#1142, thanks @RichardD012).

UniFi Console Connection

Self-recovery and fast failure - a lost console connection recovers on its own rather than staying down until the next poll, and pages that don't need the console render immediately instead of waiting on a connection that might not be there (#1136).

Performance Tweaks

UniFi OS 5.1.30 - now supported. A security and maintenance patch over 5.1.29: kernel unchanged, SGMII+ module byte-identical to 5.1.26 through 5.1.29.

Fixes

Site Admins can act on their own site - placing a device on the map, editing floor plans, writing UPnP notes and adding custom OIDs all demanded install-wide Admin. Those actions now work for Site Admins and Operators.
Coming back to the app no longer refreshes the page out from under you - mostly on mobile with Network Optimizer installed to your home screen: it would open, work for a couple of seconds, then reload and throw away where you were.
Production logs are much quieter - per-poll diagnostics that ran at Information moved to Debug. About 4,000 lines per hour before anyone did anything.

Coming back to stable

If you were running the :preview Docker tag during the preview line, it only ever points at preview builds. Pulling images on that tag after this release gets you the next future preview, not v2.7.0. Changing back to the stable channel is explained below:

Docker: retag your images to :latest, then docker compose pull && docker compose up -d
Proxmox:

pct exec <CT_ID> -- bash -c 'cd /opt/network-optimizer && sed -i -e "s#network-optimizer:preview#network-optimizer:latest#" -e "s#speedtest:preview#speedtest:latest#" docker-compose.yml && docker compose pull && docker compose up -d && docker image prune -a -f'

Windows: install the v2.7.0 MSI over the top
macOS: git checkout main && ./scripts/install-macos-native.sh

Installation

Windows: Download the MSI installer below

Docker (Upgrade):

docker compose pull && docker compose up -d

macOS (native, recommended for accurate speed tests vs Docker Desktop):

git clone https://github.com/Ozark-Connect/NetworkOptimizer.git && cd NetworkOptimizer && ./scripts/install-macos-native.sh
# or if you already have it cloned
cd NetworkOptimizer && git pull && ./scripts/install-macos-native.sh

Proxmox:

bash -c "$(curl -fsSL https://raw.githubusercontent.com/Ozark-Connect/NetworkOptimizer/main/scripts/proxmox/install.sh)"
# or if you just need to update
pct exec <CT_ID> -- bash -c 'cd /opt/network-optimizer && docker compose pull && docker compose up -d && docker image prune -f'

For other platforms (Synology, QNAP, Unraid, native Linux) or new installations, see the Deployment Guide.