FAQ

Which hardware is supported?

Currently only the x86_64 architecture is supported. Raspberry Pi / Orange Pi / Apple Silicon Mac / ARM-architecture NAS (UGREEN DXP, Zspace Z2/Z4, and most home models) are not supported yet — please don't try.

Recommended starter hardware: an N100 / N305 mini-PC (about ¥1000) comfortably runs full-feature AI (motion + face + object) on 1–3 streams; for more streams, or to add fall / license-plate recognition, step up to an i5 / i7 with an iGPU, and to an NVIDIA discrete GPU beyond that. x86-architecture NAS units (e.g. Synology XS series, TerraMaster D series, some UGREEN x86 models) can also install it. Before ordering, run uname -m on the target machine — it must output x86_64 to install.

Do I need a public IP?

No. Run it on your home LAN or a NAS; for remote access we recommend a VPN via Tailscale / Wireguard — zero ops, zero exposure. To let family watch from outside, use a reverse proxy + a strong password + HTTPS.

How many streams can one machine handle?

The bottleneck is AI detection, not video forwarding. The more detectors you turn on (face / fall / plate / gesture / cry), the more compute it takes. Conservative figures at full features, 1080p, 5fps — turning detection off on less-important cameras (record-only) frees up compute for the ones that matter:

• N100 / N150 iGPU: 1–3 streams, full features (fall detection from N150 up)
• i5 / i7 + Iris Xe iGPU: 2–4 streams (incl. license-plate recognition)
• RTX 2060 / 3060 or similar discrete GPU: around 8 streams to start, depending on VRAM and which detectors are on
• Record only, no detection: limited by disk IO and bandwidth — well over a dozen streams (depends on how many MB/s your HDD can write)

Can it run purely on a LAN?

Yes. License activation needs a one-time internet connection, after which the server can go fully offline — a failed heartbeat doesn't affect functionality (see Licensing). Event push, TTS announcements, the Home Assistant bridge, and recording playback don't need the internet.

Why not let the app connect to the camera directly?

A single RTSP stream can only be subscribed to by a few clients. Live + detection + direct app connection is already 3, which will overwhelm a cheap ONVIF camera. SkyView relays so one camera connection is distributed to N clients, while converting RTSP into a format browsers and apps can play directly and handling segmented recording — doing it yourself amounts to rewriting an NVR.

Do you support Hikvision / Dahua?

Full lineup supported. Professional IPCs like Hikvision / Dahua / Uniview / TP-Link Vigi / Huawei HoloSens are all standard ONVIF and show up on scan. See Adding Cameras.

Does it support 4K recording?

Yes. SkyView forwards H.264 / H.265 directly; 4K is bandwidth-heavy but its CPU cost is nearly 0. AI detection automatically uses the substream (416x416) so it's unaffected. Note that one 4K stream is about 42GB per day — plan your disk accordingly.

Can the face library / config sync across devices?

Not currently (single-instance deployment). A multi-machine deployment needs a shared database + shared NFS recordings directory — undocumented but feasible. A one-click "config export / import" migration is planned.

Will there be a commercial SaaS version?

Not in the short term. SkyView is positioned as "self-hosted" — data lives entirely on the user's own server, and we upload no footage to us. This is both a privacy commitment and the core difference from big-vendor IPC apps.

Will data be uploaded to the cloud?

No. SkyView is purely self-hosted: all video, face, and event data exists only on your own server, and our servers store no footage. Apart from the fingerprint-validation heartbeat during license activation (just a string of hardware hash), the client sends us no business data. You can verify by observing egress traffic at the router level.

Does it support two-way audio?

Yes. Based on the ONVIF backchannel, both Android and iOS use Hold-to-Talk, and any camera with a backchannel works. Automation can also use text-to-speech (TTS) to announce proactively, e.g. automatically playing a prompt when someone loiters.

How do I connect cameras like Tapo / Xiaomi / Wyze?

SkyView natively connects only RTSP and ONVIF. For proprietary-protocol devices like Tapo / Xiaomi / Wyze / EZVIZ, first bridge them to standard RTSP with go2rtc, then add them as a regular RTSP source.

Does natural-language recording search need internet or an API key?

No. Semantic search is done by a local VLM (vision-language model), bidirectional Chinese/English, fully offline, calling no external LLM and needing no API key like OpenAI / Gemini.

Can it be installed on a NAS (Synology / Unraid / fnOS / TrueNAS)?

Yes. SkyView is a single Docker image supporting Synology (Container Manager), Unraid, fnOS, UGOS, TrueNAS SCALE, Proxmox VE LXC, and the Home Assistant OS add-on. See the installation docs for per-platform steps.

Where are recordings stored? Can they be uploaded to cloud storage?

Recordings are stored locally on the server by default. Optional upload backends: 115 Cloud (QR-code login + instant upload) or a local archive directory, pushed asynchronously by an independent upload process that doesn't tie up the real-time path.

How does it integrate with Home Assistant?

Via MQTT discovery it automatically exposes cameras and detection events to Home Assistant for linkage in HA; conversely, SkyView's automation engine can trigger actions like TTS announcements. See the automation docs.

Does it support Apple Home (HomeKit)?

HomeKit isn't integrated for now. SkyView uses its official iOS / Android apps and the web console as its primary clients, with live view over WebRTC that auto-degrades to HLS on handshake timeout.