Key Takeaway

  • 📷 A Frigate NVR on Hostinger KVM works — inside honest limits: no GPU passthrough means CPU-only detection, which Frigate’s own docs discourage; the fix is architecture, not hardware.
  • 🧠 The 2026 detector map changed: Coral is no longer recommended for new installs; Intel iGPU/NPU via OpenVINO (~15ms) and Hailo-8/8L lead — none of which a KVM has, so the KVM plays the OFFSITE brain, not the detector.
  • 📐 The 1-3 camera rule: documented VPS deployments cap at one to three cameras with the detect stream at 720p or lower — recording can stay full resolution.
  • 🏗️ The winning build: at-home N100-class box (or old mini PC) runs detection; the Hostinger KVM carries recording, remote access, and offsite backup — the two boxes bridge via Tailscale and MQTT.
  • 🔐 Never expose Frigate raw to the internet — the VPS deploy guides’ hard rule; access rides the tunnel, and that rule is also your family’s privacy firewall.

Table of Contents: Why a Frigate NVR on Hostinger KVM tempts everyone · What the docs actually say · The two-box architecture · Step-by-step build on KVM 2 · The config that works · The weekly workflow · Cost ledger · FAQ

Setting up a Frigate NVR on Hostinger KVM is the goal of every Filipino homeowner who checks their cameras from Riyadh — and the search results never tell you the whole truth about what a KVM can and cannot carry. This piece does: Frigate’s own hardware documentation, the 2026 detector shift, the documented VPS deployment limits, and the two-box architecture that gives you LOCAL-grade detection with OFFSITE-grade resilience — the combination that actually matters when you’re three timezones from the gate it watches. One article, full build, every caveat priced.

Frigate NVR on Hostinger KVM two-box camera brain architecture concept
The KVM carries the offsite brain; the home box carries the detector.

Why a Frigate NVR on Hostinger KVM Tempts Everyone (and What It Gets Wrong)

The temptation is real: Frigate is the open-source NVR that reads your cameras with AI — person, car, cat, package — and a Hostinger KVM plan looks like the cheapest always-on box that could run it for a home in Las Piñas supervised from Al Khobar. But the pipeline of problems between “KVM plan” and “working Frigate” is specific. A KVM virtual machine has no GPU passthrough: you cannot hand the VM an Intel iGPU, a Coral Edge TPU, or a Hailo module, because the host’s shared hardware doesn’t work that way. Frigate is designed around detectors — dedicated inference devices that drop object-detection latency to milliseconds; its docs are blunt that offloading TensorFlow to a detector is “an order of magnitude faster” and dramatically reduces CPU load. On bare metal, your N100 mini PC’s iGPU runs OpenVINO detection at ~15-30ms per frame. On a KVM, CPU-only software inference runs 200-500ms+ per frame at useful resolutions — missing detections on fast motion, exactly the moments you bought cameras for.

What Frigate’s Own Docs Say About a Frigate NVR on Hostinger KVM (the 2026 Verdict)

The Frigate recommended-hardware page settles the detector question with a table worth memorizing: Intel N100 iGPU ~15ms (one detector instance), N150 ~15ms, Iris XE ~10ms, dedicated NPUs ~6ms Hailo-class — against which CPU-only detection is “absolutely discouraged” in the project’s own community support. The 2026 shift is as important: the docs now state the Coral Edge TPU is no longer recommended for new Frigate installations (except very low-power or otherwise-incapable hardware), pointing instead to Hailo-8/8L M.2 modules and Intel iGPU/NPU via OpenVINO. Translation for the Filipino builder: the budget detector everyone recommended for five years is being retired, the replacements are iGPU-class hardware a VM cannot reach, and the VPS path exists but only as what it honestly is — a limited bridge, per the RamNode VPS deployment guide: plan for one to three cameras detecting at 1280×720 or lower, recording may stay full resolution, and never expose the Frigate UI raw to the internet.

The Two-Box Architecture: Where a Frigate NVR on Hostinger KVM Actually Wins

Here’s the build that makes a Frigate NVR on Hostinger KVM worth the money anyway — the KVM takes the role VPS hardware is genuinely good at. Box one, at home: an N100-class mini PC (₱8-12K on Shopee, often secondhand office SFFs for less) running Frigate with its iGPU detector — this box sees the cameras on your LAN and does the millisecond inference. Box two, in the cloud: your Hostinger KVM 2 runs the offsite stack — continuous recording vault (Frigate’s media storage mirrored out), the remote-access gateway (Tailscale subnet router so your phone in Riyadh reaches the home Frigate WITHOUT any exposed port), the MQTT broker linking your HA Container instance (#21 of this series), and the alert relay that pushes detections to Telegram — the voice-assistant pairing with our HA Voice server build (#20) makes the same events speak out loud at home. The home box keeps inference local (fast, private, bandwidth-light); the KVM keeps everything else survivable — if a thief takes the home box, the recordings already left the building. That offsite-vault property is the actual product; detection was never the KVM’s job.

Step-by-Step: Building the Frigate NVR on Hostinger KVM Companion Half

Step 1 — plan sizing. KVM 2 (2 vCPU, 8GB RAM class, ~$7-9/mo intro) carries the vault + tunnel + broker + relay comfortably for 2-4 cameras’ recordings. Storage: recordings are the disk-eater — a 4TB block (Hostinger’s KVM plans take block storage) holds roughly 3-4 weeks of 4-camera continuous 1080p; enable Frigate’s retention (30 days detect-relevant, 7 days all-video) to make the math honest. Step 2 — OS and tunnel. Ubuntu 24.04, then install Tailscale and declare the KVM a subnet router: curl -fsSL https://tailscale.com/install.sh | sh && sudo tailscale up --advertise-routes=100.64.0.0/10 — approve the route in the admin console once; every device you own now reaches home through the KVM without a single forwarded port. Step 3 — broker. Docker Compose brings up Mosquitto in four lines; this is the same pattern our HA Container on KVM piece (#21) documented — the camera box, the HA instance, and the alert relay all publish/subscribe here. Step 4 — the media mirror. The honest-simple version:rsync over the Tailscale link from the home Frigate’s /media/frigate to the KVM on a nightly cron; the power version runs a second Frigate instance on the KVM pointed at the home box’s restream for select cameras. Step 5 — alerts. The home Frigate publishes MQTT events; a tiny script — the pattern our n8n on VPS setup automates properly — subscribes and forwards to your Telegram via bot API — detection events now hit your phone in Riyadh within seconds, with the recording safe on the KVM if the home box dies mid-event.

The Frigate NVR on Hostinger KVM Config That Works (Home Box → KVM Bridge)

On the home box’s Frigate config.yml, every camera carries a detect stream at the honest resolution and a record stream at full quality:

go2rtc:
  streams:
    gate_cam:
      - rtsp://frigate:S3cret@192.168.1.108:554/stream1

cameras:
  gate_cam:
    enabled: true
    ffmpeg:
      output_args:
        record: preset-record-generic-audio-copy
      inputs:
        - path: rtsp://127.0.0.1:8554/gate_cam
          input_args: preset-rtsp-restream
    detect:
      enabled: true
      width: 1280
      height: 720
      fps: 5
    objects:
      track:
        - person
        - car
        - motorcycle

The 5fps detect stream at 720p is the documented VPS-and-home sweet spot: enough temporal resolution for person/car tracking, low enough to keep the CPU cool, recorded stream stays separate at full res. The KVM never touches this config — it mirrors, brokers, and relays. Separation of duties is the whole architecture.

The Weekly Workflow: Running Your Frigate NVR on Hostinger KVM From Riyadh

The two-box build settles into a five-minute weekly rhythm: Monday morning, check the KVM’s vault disk gauge (block storage fills are the one operational failure this build has); scan Frigate’s events console for the week’s clips worth keeping (the garage delivery that isn’t yours); and once a month, test the failure path — pull the home box’s power, confirm Telegram still delivers the “connection lost” alert from the KVM’s own watcher, confirm the nightly mirror catches up when power returns. That failure-path test is what separates a security system from a security hobby; the KVM’s offsite vantage makes it a two-minute exercise instead of an afternoon of panic. For the family back home, one more habit from the surveillance piece: cameras watch the gate and the garage — interior privacy zones stay interior, and the family knows exactly which zones record. A security system the household trusts is the only one that survives the household.

The Cost Ledger for a Frigate NVR on Hostinger KVM (Complete, No Surprises)

Home box (N100 mini PC or secondhand SFF): ₱8,000-12,000 one-time · Hailo-8L M.2 module if you step up from iGPU detection: ~₱9-10K (optional — iGPU first) · Hostinger KVM 2: ~$7-9/mo intro (~₱420-540) · 1TB block storage: ~$5/mo (~₱300) — recordings vault · Tailscale: free tier is permanent-fit for this design · Total monthly carry: under ₱1,000 — cheaper than a single month of a commercial cloud-camera subscription with better privacy and footage you own outright. vs the “pure VPS Frigate” fantasy: same KVM cost, but honest about detection limits — no amount of spend inside a VM adds an iGPU. The money goes further split across two boxes than concentrated in one.

Frequently Asked Questions

Can a Frigate NVR on Hostinger KVM run detection without a GPU?

Technically it runs; practically it’s discouraged — Frigate’s docs and community treat CPU-only detection as the last resort (hundreds of ms per inference vs ~15ms on an N100’s iGPU). The honest KVM role is the offsite brain: recordings, tunnel, broker, alerts. Detection belongs on a home box that has an iGPU.

The Frigate NVR on Hostinger KVM ceiling is real.

How many cameras can a VPS Frigate handle?

The documented guidance for VPS deployments: one to three cameras with detect streams at 1280×720 or lower — recording can stay full resolution. Our two-box build sidesteps the ceiling entirely: the home box detects all your cameras locally, and the KVM carries the vault and alerts.

Is a Coral TPU still worth buying for a 2026 Frigate build?

For new builds, no — Frigate’s documentation no longer recommends Coral for new installations, pointing to Hailo-8/8L M.2 modules and Intel iGPU/NPU detection via OpenVINO. Existing USB Coral units keep working fine; the money for a new purchase goes to the iGPU-class box.

How do I access my home Frigate securely from abroad?

Through the tunnel, never a forwarded port: the KVM runs Tailscale as a subnet router, your phone joins the tailnet, and the home Frigate’s UI opens over the private network. Zero exposed ports is the deployment guide’s hard rule — and it’s what keeps random scanners out of your cameras.

What happens if the thief takes the home box?

The nightly mirror already moved the recordings to the KVM — footage survives, the tunnel dies (which your phone notices immediately via the KVM’s connection-lost alert), and the MQTT relay keeps the rest of your HA devices talking. Recovery is a new box plus a config restore.

Which KVM plan is right for the offsite stack?

KVM 2 (2 vCPU / 8GB class) carries the vault, tunnel, broker, and alert relay for a typical 2-4 camera home. The variable is storage: size the block-storage add-on to your retention math — 1TB holds about a month of 4-camera 1080p continuous with standard retention rules.

If this intelligence helps you, you can add WorldNgayon as a preferred source on Google — free, one click, and it helps other Filipinos find the answers faster.

Financial Disclaimer: This article is for general information and education only and does not constitute financial advice. Pricing figures are plan snapshots at writing; verify current rates with providers before purchase decisions.

Editorial Transparency Note:WorldNgayon uses AI-assisted tools in parts of its editorial workflow. For our editorial standards, sourcing practices and use of AI, see worldngayon.com/about/. Article bylines and source credits identify the stated authorship; this general note does not certify how an individual archive article was originally produced. Report factual errors through worldngayon.com/contact-us/.

Leave a Reply