# Grizzly Systems

> Off-grid, multi-sensor field monitoring for conservation, security, and research.

Grizzly Systems builds **GrizCam** — a ruggedized field unit that captures 360° video, directional multi-channel audio, environmental telemetry (temperature, humidity, barometric pressure, air quality), and chemical / gas signatures (smoke, VOCs, decay) from places without power or cell coverage. On-device AI filters noise at the source; cloud AI turns what's left into actionable alerts.

Headquartered in Paradise Valley, Montana. Founded 2021.

## Product

- **GrizCam device** (pre-production; specifications subject to change; price on request): Multi-sensor field unit. 280° field of view, up to 4K, with 360° motion detection (ANA watches all the way around and wakes the cameras and microphones). A multi-microphone array estimates the bearing (direction, not distance) a sound came from; with 3 or more GPS-synced units, GrizCam Desktop can locate it. Records audio at up to 96 kHz (sounds up to about 48 kHz) — not suited to most bat work. Chemical/gas sensors (CO, CO₂, NO₂, CH₄, PM2.5), environmental telemetry. Months on battery; continuous on solar. IP67. Edge AI inference on-device.
- **Edge AI**: Filters wind, vegetation, and shadow at the source. In a peer-reviewed three-month field trial in the Greater Yellowstone Ecosystem (Kaltenbach et al., Wildlife Society Bulletin, 2025, https://doi.org/10.1002/wsb.1598), GrizCam's edge-AI prototype captured false-positive images in about 1% of events at average conditions, versus about 50% for a Browning and 6% for a Reconyx trail camera, and nearly zero in heat of 30 °C or more. We keep testing and publish new results as they come in. Deployments generate signal, not volume. Cloud AI (Claude / GPT / Gemini class) then turns what's left into actionable alerts.
- **Connectivity**: Four redundant paths — private mesh (0.5–3 km node-to-node, self-forming), cellular (LTE-M / NB-IoT, carrier-agnostic eSIM), satellite (coming soon), and offline (encrypted local store-and-forward, synced on retrieval). Always one path available.
- **Software Suite**:
  - **GrizCam Portal** (web, real-time operations): Live fleet map, device health, detection feed filtered by species/confidence/time/location, search, and email alerts. Enterprise adds image and acoustic AI alerts, unlimited users and GrizCam Desktop.
  - **GrizCam Desktop Application** (local, research analysis): Thousands to tens of thousands of recordings, even on a laptop. Interactive spectrogram annotation. GPS collar playback. Accelerometer + magnetometer overlay. Opens standard audio (WAV/FLAC) and image/video (JPG/MP4) files from most recorders and trail cameras. Free tier after sign-in.

## Mission profiles (use cases)

1. **Land security** — Vehicles, gunshots, people, fence breaches across thousands of acres. Working ranches and large landholdings.
2. **Critical infrastructure** — Pipelines, substations, fence lines, remote facilities. No trenching, no power runs, no cell dependency.
3. **Conservation & research** — Species presence, herd movement, gunshot activity (anti-poaching), ecosystem change. Built with biologists; deployed in Yellowstone National Park.

## Why off-grid sensing matters

94% of the world's land area has no power outlet. That's where wildfires start, where criminals act, where wildlife actually lives. Traditional surveillance fails there — cell-dependent designs go dark, motion-trigger cameras flood the channel with wind and shadow noise (~65% false-positive rate), missed events go un-recorded. GrizCam closes the gap.

## Research — Cry Wolf Classifier

The **Cry Wolf Classifier** is the first wolf-vocalization detector trained on **The Cry Wolf Project** dataset (Yellowstone's Wolf Project — one of the world's largest libraries of wild wolf vocalizations, from the Greater Yellowstone Ecosystem). It is built in **GrizCam Desktop** by transfer learning: a lightweight linear probe trained on frozen embeddings from bioacoustic foundation models (Earth Species Project AVES2, Google DeepMind Perch 2.0), using only a researcher's own annotations — no code, no encoder fine-tuning. Reported accuracy is leakage-free held-out out-of-fold (OOF) F1 (~0.90 howl / ~0.98 chorus for BEATs transformer encoders). The same labeled dataset also produces a compact ONNX detector that runs on-device on GrizCam edge hardware (ESP32-S3-class) for real-time field detection. Conservation value: turning passive recordings into reliable occupancy and abundance signals for estimating wolf populations — the basis for management decisions like hunting quotas. Page: https://www.grizcam.com/cry-wolf-classifier

## Company

- **Headquarters**: Paradise Valley, Montana (Greater Yellowstone Ecosystem)
- **Founded**: 2021
- **Co-founders**: Cody Goldhahn (CTO, 5th-generation Montanan, embedded systems), Jeff Reed (CEO, PhD computational linguistics, 30+ years electronic manufacturing at Avnet/Microsoft/T-Mobile/Arrow)
- **Funding**: Raising **$1,888,000** through the **Invest in the Wild** campaign (a partnership with the Wild Livelihoods Business Coalition) to build GrizCams for conservation partners in the field — https://www.grizcam.com/invest-in-the-wild
- **Conservation partners**: Bear Awareness Gardiner (bear-proofing coexistence), Centro de Innovación Científica Amazonía / CINCIA (illegal Amazon mining), Yellowstone National Park (anti-poaching grid), Western Landowners Alliance (predator conflict), Adventure Scientists (biodiversity), Holland Ranch (biodiversity + non-lethal predator deterrence), The Cry Wolf Project (wolf-vocalization blueprint), Yellowstone Safe Passages (wildlife crossings)

## Invest in the Wild (fundraising campaign)

**Invest in the Wild** is a partnership between Grizzly Systems and the Wild Livelihoods Business Coalition, raising **$1,888,000** to build GrizCams for conservation partners doing the hard work in the field. Campaign page: https://www.grizcam.com/invest-in-the-wild

## Privacy

Analytics (Google Analytics 4 + Microsoft Clarity) on the marketing site loads only after explicit visitor consent. No marketing trackers, no ad networks. Customer Data (GrizCam captures: video, audio, telemetry) belongs to the customer, never sold or shared. Every customer can opt out of AI training. Full policy: https://www.grizcam.com/privacy

## GrizEyes — what the camera sees

**GrizEyes** is the imaging side of GrizCam. A single device carries **two 4K image sensors** whose views are stitched into one continuous panorama, so one camera on a tree, fence post, gate frame or pipe stand covers every approach around it at any bearing — the only thing it cannot see through is whatever it is mounted to. See it for yourself at https://www.grizcam.com/grizeyes: real panoramas and 360 video captured by deployed GrizCams, not renderings, in a viewer you can drag to look around, view flat, or open on a phone or headset. The same page explains why the traditional camera-trap heat trigger (PIR) both fires on nothing and misses real animals, and how **ANA** decides what is worth recording instead — capturing on the signature of motion rather than waiting to recognize a subject, so an unfamiliar animal is still recorded. ANA is an always-on, low-power watch that learns its own scene from the moment the device is deployed, with no zones to draw; only when it decides a motion is worth a look do the 4K cameras and the on-device AI wake to capture and label it (animals, people, vehicles), so the battery is spent on real activity and far fewer empty frames reach the user. Each event goes out over cellular, satellite or a private-area mesh network to the GrizCam Portal, where a whole deployment can be reviewed, searched and alerted on from any browser. In GrizCam Desktop, GrizEyes lays out every capture beside the audio from the minutes before and after it, and also opens photos and video from other camera traps such as Browning and Reconyx.

## Downloads

The **GrizCam Desktop** application is free for the conservation community and available to download. Downloads require a free account (each build is licensed to a user; the EULA is accepted once at sign-up — no payment required for Desktop). To get it: visit https://www.grizcam.com/download, click **Log On** (top right) and sign in or create a free account, then open **Settings (gear icon) → Downloads → GrizCam Desktop** and choose the **Windows** or **Mac** installer. **Training videos** for GrizCam Desktop are available from the same page: click **Training Videos** in the GrizCam Desktop header to open the video library, no account required. **User documentation is versioned:** each build that ships a guide shows a **Documentation** button beside its Download button, opening the user guide written for that exact version (an HTML guide with search and screenshots, plus a print-ready PDF). Older builds are never pointed at newer docs. Because the build list is behind sign-in, the Documentation button appears once signed in. GrizCam **firmware** downloads are coming soon. Public download info page: https://www.grizcam.com/download

## Contact

- Email: info@grizzlysystems.io
- Phone: 425.246.0025
- Web: https://www.grizcam.com (primary) · https://www.grizzlysystems.io (alias, same site)
- TED Talk (Jeff Reed, 2025): "Can AI help us speak with wolves?" — https://www.ted.com/talks/jeffrey_t_reed_can_ai_help_us_speak_with_wolves

## Press

Featured by NVIDIA, The Washington Post, BBC, PBS, TED, Yellowstonian, Mountain Outlaw, NPR.

## GrizCam Desktop: share audio and annotations with colleagues (2026-09)

GrizEars can share a recording straight from one GrizCam user's desktop to another's, with the sender's annotations already on the spectrogram. The audio is encrypted on the sender's machine before it leaves; the GrizCam service only carries the parcel. The recipient annotates and presses Sync with contact; only the changes travel back, as proposals the owner accepts or denies, and every shared annotation has its own conversation thread. Pair once with any GrizCam account; Pro subscription on both ends. How-to: https://api.grizcam.com/operations/downloads/docs/desktop/grizcam-desktop-user-guide_v0.23.0.html#howto-share-audio

## Greater Yellowstone Bioacoustics — uncommon owls (2026-09)

Passive acoustic monitoring across the Greater Yellowstone Ecosystem: Yellowstone scientists and citizen scientists run grids of autonomous recording units that listen around the clock, and the archive is where the region's rarest voices turn up. https://www.grizcam.com/gye-bioacoustics is a field guide to eleven uncommon owls (Flammulated, Great Gray, Boreal, Northern Saw-whet, Northern Pygmy, Long-eared, Short-eared, Western Screech, Burrowing, Snowy, with Great Horned as the common comparison): each species' Yellowstone status from the NPS 2021 checklist and the park's annual bird reports (1999–2019), the territorial call's measured fundamental frequency and cadence (measured from open iNaturalist recordings; cadence checked against Cornell, USDA GTR-260 and McCallum 1994), what to listen for on a spectrogram, and the confusions to rule out before confirming. Every number states where it came from. Candidates are found with GrizCam Desktop; a person confirms the sound.

## Project Centinela — Grizzly Systems and Planet in the Greater Yellowstone Ecosystem (2026-09)

Grizzly Systems and Planet (https://www.planet.com/project-centinela/) are piloting a sensor-fusion effort in the Greater Yellowstone Ecosystem: daily high-resolution satellite imagery and analytics from Planet, fused with GrizCam camera, acoustic and scent sensors on the ground. Yellowstone National Park is protected, but the private lands between the park and the national forests carry the migration corridors, trout water and wildlife that the region's billion-dollar tourism economy depends on, so the pilot puts verification under conservation on working land. https://www.grizcam.com/centinela describes the goals (spatial planning in corridors and riparian systems, conservation on private lands complementing protected areas, reducing human-wildlife conflict while sustaining working landscapes, and mobilizing private capital and tourism-driven conservation finance) and seven project pillars: fire response tools, invasive plant monitoring, wildlife connectivity on private lands, native fisheries and riparian restoration, beaver-mediated drought resilience, payment-for-presence programs, and wildlife crime and compliance monitoring. Partners include Wild Livelihoods, Montana Freshwater Partners, Yellowstone Safe Passages, the Upper Yellowstone Watershed Group, Montana State University, Utah State University, the Center for Large Landscape Conservation, USGS, the USDA Forest Service and the National Park Service.

GrizEars is the acoustics workspace in GrizCam Desktop, Grizzly Systems' free software for turning field recordings into findings. It opens audio from GrizCam devices, animal-borne biologger collars and third-party recorders such as Wildlife Acoustics units, handles tens of thousands of files at once even on a laptop, and lets researchers annotate by hand, then train, validate and run their own detectors on bioacoustic foundation models (Perch, AVES2) with no code. Paired with GrizCam hardware, a multi-microphone array lets it estimate the direction every sound came from, an onboard compass turns that into a true bearing on the map, and GPS plus a battery-backed real-time clock line up calls captured by neighboring devices to the second. With biologger collars it pairs the audio with the collar's motion sensors (IMU) to reconstruct the animal's path by dead reckoning.

The GrizEars page (https://www.grizcam.com/grizears) turns any recording into a 3D spectrogram in the browser with the same engine GrizCam Desktop uses; nothing is uploaded. Its Bioacoustic journeys section (https://www.grizcam.com/grizears#bioacoustic-journeys) follows wolf 1513F of Yellowstone's Rescue Creek pack across the park's Northern Range over 17 days in the winter of 2024-25: a reconstruction, not camera footage, rebuilt from her collar's GPS and a biologger's motion sensor, compass and microphone (GPS-anchored dead reckoning), with her howls and choruses placed on the route. Her collar recorded about 26 minutes of howling and calling a day. The video is passphrase-protected; the full methods and statistics are published on the page.