Leo · 2026-11
On handover day the argument is rarely about whether the video plays. Two devices can both list ONVIF and still fail to talk, because the term covers discovery, streaming, recording control and metadata as four separate capabilities. ONVIF camera compatibility is therefore a test sequence rather than a datasheet line, and the sequence below is ordered the way failures actually appear.

This checklist is written for the person who signs the acceptance document, not for the person who writes the brochure.
Four things, and they fail independently. Discovery lets the recorder find a camera on the network. Streaming carries the video. Recording control lets the recorder ask an edge device to store or replay. Metadata carries what the analytics decided. A camera can stream perfectly and still expose no usable event data.
The specification itself is published and public at onvif.org/profiles. Read it before a tender, because a clause that names no profile is a clause you cannot enforce at handover.
One clarification is worth making early. On a recorder, ONVIF support is a capability that the manufacturer implements and documents. It is not a certificate the buyer can hang on the wall, and a supplier who presents it as one is overselling. The published woCyber product pages describe PoE IP camera support over ONVIF on the S1 and A2 Pro and stop there, which is the correct level of claim.
Name the profile, not the protocol. Profile letters describe which capability set a device implements, and the differences decide whether your analytics and your archive work.
| Profile | What it enables | What to test in the field |
|---|---|---|
| Profile S | Base video streaming between a camera and a client | Stream opens on the recorder, correct resolution and frame rate, no re-encode surprise |
| Profile T | Advanced streaming, including H.265 and bidirectional audio paths | Codec negotiation, audio in both directions, stream restart after a network drop |
| Profile G | Recording control and edge storage on the camera side | Playback request reaches the camera, timestamps match the recorder’s clock |
| Profile M | Metadata and event delivery from analytics | Event arrives with a usable label, and the recorder stores the label with the clip |
Most disputes I have seen came from a tender that asked for ONVIF without a letter, then expected analytics metadata to arrive from cameras that only implemented Profile S.
A profile letter is also an enforcement tool. Once the clause names Profile T for the new cameras and Profile S for the retained stock, the acceptance test writes itself and the argument about “it should work” disappears. Write the exception in the same clause. Legacy cameras with streaming only are acceptable if the analytics runs on the recorder instead of the camera.
That exception is where ONVIF camera compatibility stops being a protocol question and becomes an architecture choice. Metadata can come from the camera or from the recorder, and only one of those paths survives a mixed estate.
Nearly every first-day failure is a network condition, not a protocol bug. Cameras and recorders discover each other over multicast on the local segment, and multicast does not cross a router by default.
Four conditions break discovery in practice. The camera sits on a different VLAN from the recorder. The switch drops multicast as unknown traffic. The camera clock is hours away from the recorder, so timestamps in the discovery response look stale. The camera has a default credential that the recorder will not use, and it answers discovery but refuses the session.
Whether a single appliance should carry both jobs at all is a separate question, taken up in our nvr-vs-nas-for-surveillance-which-appliance-fits-your-cameras comparison. For the failure modes: fix those four before you blame the firmware. Put both devices on one segment for commissioning, set NTP on both, and confirm the credential the recorder stores is the one the camera accepts. Do that during commissioning, not after the complaint arrives.

On a European site the answer is usually “inside the building”, and the reason is legal rather than sentimental. Video of identifiable people is personal data. If a provider holds a copy, that provider can be asked to produce it. If the recording never leaves the customer’s own network, there is nothing to produce.
That has procurement consequences. The customer’s data protection officer will ask for a processing agreement under Article 28 of the GDPR, a retention period per camera, an access record, and a deletion routine. Answer those four in writing and the compliance conversation ends early. Answer them with adjectives and it restarts in week six.
Retention is the lever most projects get wrong. Longer is not safer. A retention period should match the purpose the cameras were installed for, and it should be enforceable per camera group rather than globally. On this line, the published S1 specification lists retention options of 7, 14 and 30 days with event-triggered and continuous recording modes.
If you need the retention arithmetic rather than the policy language, our surveillance-storage-retention-policy-nas-for-eu-projects guide walks the capacity side of the same decision.

Run the test on real camera models, at the firmware versions you will actually ship. A datasheet test is not a test.
The export test catches more problems than the rest combined. A clip that only plays inside one vendor’s client is not an archive, it is a hostage.
Two calculations decide whether the site works, and both are short enough to do during the survey.
Aggregate ingest is camera count multiplied by bitrate, and the figures below are worked examples rather than measurements. Thirty-four cameras at an average 4 Mbps come to about 136 Mbps on paper, which is comfortable on a 1GbE recorder port but has to be planned on the switch uplink once you add workstation playback. Fifteen 4K cameras at 12 Mbps reach 180 Mbps and start to stretch that uplink.
Storage follows the usual relation: terabytes = cameras × bitrate in Mbps × 86,400 × retention days ÷ 8, then ÷ 1,024 three times to move from megabytes to tebibytes. Worked through: 8 × 4 × 86,400 × 30 ÷ 8 = 10,368,000 MB, which is about 10.4 TB decimal, or roughly 9.9 TiB the way an operating system reports capacity.
Run both numbers before you choose the appliance. A 34-camera site at 30-day retention is an archive problem, and no amount of protocol support fixes a capacity shortfall.
The metadata path becomes the product. A recorder that only stores video forces a human to watch it. A recorder that classifies events on the device turns hours of footage into a short list.
On the woCyber S1, person, vehicle and pet classification runs on the appliance, with false alarms published at under 3 percent, and the account spaces are physically isolated so one user cannot browse another’s recordings. Isolation inside the organisation is the feature a data protection officer actually cares about, and our nvr-with-person-vehicle-and-pet-detection-2026-buyer-checks checklist raises the same question for shared recorders.
There is a trade-off worth stating. On-device classification costs compute on the recorder, and it constrains which event types you get. If your project needs licence plate recognition or face matching, verify that the firmware exposes it before you promise it in a tender.
Verification is the other half. A classifier that reports an event still needs a human to confirm it before the clip is used as evidence, and that review step has to be scheduled into the site’s process. Projects that skip the scheduling end up with accurate labels and nobody acting on them.

Get seven items on paper. Camera models and firmware versions tested. The profile letter implemented. Maximum concurrent streams at the specified resolution. Retention options and how they are enforced per group. The export format and whether it needs vendor software. The access control model and log destination. And the certification set the supplier can document for import: on this line that is CE, UKCA, FCC, RoHS, REACH and WEEE.
Consider a hypothetical integrator delivering 34 cameras across two sites for a mid-size logistics operator, with two camera brands in the bill of materials because the client’s existing stock had to be reused. The acceptance clause says ONVIF. The old cameras implement Profile S only. The new analytics requirement needs Profile M metadata. That gap is found in commissioning unless it is found in the survey, and finding it in the survey is the whole job. The fix costs one line: the profile letter per camera model goes into the bill of materials, and the analytics requirement moves to the recorder, where it can be tested and signed.
If the project needs the recorder to classify events on the device rather than in a cloud service, the products/s1-security-kit is the appliance in this line built for that role: PoE IP cameras over ONVIF, on-device classification, local retention, and isolated account spaces.
The expensive failure is rarely a camera that will not stream. It is a site that streams perfectly and cannot answer a question three months later, because nobody kept the profile letter, the firmware version or the retention decision on file.
Keep the commissioning record with the handover pack, and put the retention table into the contract rather than an email attachment. Both survive the person who wrote them leaving the company, which is the test most site documentation fails.
A specification that names the profile beats a longer feature list, not because features are unimportant, but because a named profile can be tested, and a test can be failed honestly.
Whether you need surveillance storage, private cloud for business data or an OEM/ODM partner, our export team will map the right platform to your application — reply within one working day.
Browse all platforms →An off-grid security camera with local storage records onto hardware at the site, powered by the site's own supply, with no cloud…
November 20, 2026A NAS backup strategy for business rests on three pillars in 2026: a copy that is separate from the primary array, a…
November 20, 2026A security camera system without monthly fees records to storage you own, on hardware that processes video locally, instead of uploading clips…
Ask our export team for the OEM/ODM cooperation guide and a written quotation.