Know where people are.
Without watching them.
Spatial awareness for the rooms a camera can't go — and shouldn't. Server halls, vaults, clean rooms, executive floors. Vantage reports anonymous positions and zone events, not faces.
A volumetric awareness layer built on the Beamforming Feedback Information your radios already emit. No new sensor, no firmware fork. The decoder runs on the node you ship.
A standardized sensing substrate, a decoder trained on emitter geometry, and a privacy boundary that is architectural — not a setting. The combination is the defensible part.
replay data · no live capture yetPresence is worth knowing. Identity is a liability.
Cameras see everything — which is the problem. They carry privacy exposure, they have blind spots, and they cost money to place, light, and store. In a secure facility you often want one answer: is someone where they shouldn't be? A lens is a heavy way to ask a light question.
Every 802.11ac / ax / be device returns Beamforming Feedback Information to the access point as normal MU-MIMO operation. Today that feedback is computed, used once, and discarded. It carries the spatial signature of the room. No one ships volumetric awareness from it on-device.
The substrate is free, standardized, and already deployed across the installed base. The hard parts — turning that feedback into volumetric tracks, and proving the emitter-geometry step is load-bearing — are where the IP sits. That is the wedge.
Feedback in. Anonymous tracks out. One node.
BFI
The compressed feedback matrix devices already send back to the access point.
Decoder
A neural decoder trained on the fixed emitters' geometry fingerprint for the venue.
Tracks
Anonymous x / y / z positions with a confidence value. A point, never a likeness.
Zone logic
Armed schedules, restricted areas, dwell. A decision gate filters low-confidence noise.
Audit
A tamper-evident record of every event. Nothing raw leaves the node by default.
Self-calibration infers the relative positions of the venue's fixed emitters during a short startup window — no site survey, no floor plan supplied. The decoder weights stay fixed across venues; only the input normalization adapts.
It runs itself. Then it's yours to drive.
Real scenario data through the real event engine. Watch a track cross into an armed zone and write an event — then take the controls: drag the subject, scrub the timeline, switch scenarios, arm and disarm zones.
Readout
Scenario
Armed zones
Event log · tamper-evident
Anonymous by default. Identity only by consent.
No raw egress
Raw feedback, decoder activations, and outputs stay on the node unless an authorized channel is opened. A wiring fact, not a checkbox.
Anonymous tracks
The system produces unlabeled positions. There is no passive identification step to turn off — it isn't there.
Consent re-ID
Recognizing a specific person requires prior enrollment, an enrollment signature, and a consent token held by the consenting party.
For a buyer this is the difference between a system legal and security will approve, and one they'll fight. You get the event, not the person.
For an integration partner the boundary is enforced in the runtime, so the privacy story travels with the silicon instead of depending on each deployment.
The architectural restriction — a system that could re-identify but is built to refuse without consent — is the cleanest patentable surface in the set.
No hand-waving. Here's the line.
Proven — in simulation
- ✓Substrate carries the signal.First viability gate passed on held-out data — the feedback is separable, not noise.
- ✓The geometry step is load-bearing.Ablation: emitter-geometry conditioning collapses planar error ~0.88 m → ~0.06 m in clean simulation, 5/5 seeds. A relative ~15× gap, not a real-world accuracy figure.
- ✓Full product logic + UI, built and tested.Seven-stage pipeline, decision gate, detectors, tamper-evident audit, nine operator pages. You're driving it above.
Pending — the real test
- →First real over-the-air capture.A ~$300 bench captures real feedback and answers the one falsifying question: does a body move the signal at all?
- →Real-world accuracy numbers.No public benchmark exists for this regime. Expect room-zone resolution, not pose. Synthetic figures are not quoted.
- →Provisional filed · one paid pilot.Provisional first (no hardware dependency), then a zone-grade pilot at a commercial site with UWB ground truth.
The moat is the combination, not any one piece.
Pieces of this exist in prior art — the substrate itself is conceded. The defensible position is the system claim that requires all five together, anchored on the emitter-geometry step. No competitor recites the set.
BFI substrate
Standardized feedback. Conceded — never claimed alone.
Geometry self-cal
Emitter geometry inferred from the feedback. The load-bearing core.
Volumetric
x / y output as a required element. Planar today.
On-node
No raw egress as an architectural boundary. The cleanest surface.
Consent re-ID
Identity gated behind enrollment + consent.
Prior-art mapping is complete; the primary BFI exposure is tracked and a freedom-to-operate review is scoped. Active work is the provisional and the geometry-ablation re-confirmation. Full memo in the data room.
For a silicon partner the relevant element is (i) — it lands on the feedback your radios already produce, with the decoder running on your compute.
For a buyer this means the capability isn't a commodity you can source three ways — it's one defensible system.
Run a pilot in a space that matters.
Put it on your silicon.
See the data room.
One node, one restricted area, ninety days. We bring ground truth and a scored report at zone grade. You decide if anonymous beats a camera.
A reference decoder against your feedback output, on your NPU. We start with the substrate gate on a bench you can replicate.
Prior-art memo, ablation results, pilot economics, and the capture plan. Enough to underwrite the next milestone, not a vision deck.