Confidence Score
Outdoor Trail Cue 👀
🍃 Touch Grass · Open-Source AI Challenge
Get off the screen.
Explore the wild with Open-Weight AI.
Identify leaves, songbirds, wildflowers, mushrooms, and bugs on the trail with zero cellular signal. Trail Companion runs an open-weight foundation model right in your browser, keeping your photos private while rewarding the time you spend with your phone in your pocket.
76+
Native Species Cataloged
~140ms
On-Device Inference Speed
100%
Offline (Airplane Mode Ready)
0 kb
Private Location/Photo Leaks
Autonomous On-Device Vision
Powered by open-weight CLIP ViT-B/16 quantized to int8 via Transformers.js and ONNX Runtime Web. Your camera photos never leave your device. Download the weights once on Wi-Fi (~87 MB), and identify flora and fauna deep in the backcountry with zero cell towers.
Screen-Free Walk Meter
We don't want you glued to your glass screen. Our Walk Mode measures how much of your outdoor time was spent off-screen, encouraging you to pocket the phone and observe nature directly.
"That's Not Grass" Rejection
Point your camera at a laptop or indoor desk, and the model's zero-shot negative bank flags it immediately, reminding you to close the laptop and step into the sunlight.
Synthesized Birdsong Audio
Hear authentic whistling calls and forest wind rustles generated entirely offline with the Web Audio API. Perfect for learning to recognize birds by their songs on the trail.
Zero-Shot Extensible
Encounter a rare endemic plant not in the guide? Type its name in our Studio. The open text model computes fresh embeddings on-device in seconds without any retraining.
Ready to test it right now?
Try one of our 6 instant field test photographs or snap an item near your window.
Open Field Scanner Workstation →Field Scanner Workstation
Capture live camera photos or choose a picture to identify on-device with zero signal.
Tap shutter to select or snap a photo
100% processed locally on your phone
Instant Sample Photo Test Deck
Get Trail-Ready
Download the open-weight CLIP ViT-B/16 vision model (~87 MB) once to your browser storage. Works completely offline afterwards.
Downloading model…
0%
Cached permanently on device. You only download this once.
Native Field Guide
Searchable biodiversity taxonomy pre-loaded for offline identification.
Walk Mode & Screen-Off Tracker
Start a hike, then pocket your phone. We calculate your percentage of screen-free time.
—
Off-Screen
00:00
Time Outside
0
Species Logged
0:00
Screen Time
Daily Trail Bingo
0 / 9Today's observations automatically mark tiles off the board.
Trail Badges & Milestones
Celebrate real-world outdoor exploration.
Recent Outdoor Walks
- No walks recorded yet. Step outside!
Field Journal
Your local nature observations, stored 100% on your device with zero cloud tracking.
Your journal is currently empty. Point your scanner at a leaf, flower, or songbird to log your first find! 🌱
Model Studio & Runtime Settings
Inspect the open-weight model runtime or teach new species on-device without retraining.
Runtime Specifications
- Model
- CLIP ViT-B/16 (open weights, ONNX int8, 87 MB)
- Runtime
- Transformers.js v4 + ONNX Runtime Web
- Device
- —
- Taxonomy
- —
- Network
- —
Teach a Local Species
Zero-shot expansion: Type any plant or wildlife common name. The open-weight text encoder computes fresh vectors directly in this browser.
Privacy & Data Control
All observation photos, GPS coordinates, and walks are stored strictly within IndexedDB on this device.