ForageQuest: A Single-File AI That Tells You Not to Eat the Mushroom

Description and use

ForageQuest is an offline, privacy‑preserving web app that identifies mushrooms using a local CLIP model, applies safety gates to avoid false positives, and always advises consulting a mycologist.

Problem

Foragers need a reliable, offline, privacy‑preserving tool to identify edible mushrooms without risking poisoning.

Outcome

ForageQuest delivers safe, offline mushroom identification in a single HTML file, eliminating server costs, protecting privacy, and ensuring the system fails safely by refusing to name species when uncertain.

Workflow

• Capture mushroom photo in browser
• Model (CLIP via Transformers.js) scores image against 16 mushroom genera and 14 non‑mushroom labels
• Apply confidence gate (threshold 0.70)
• Apply look‑alike blocklist (e.g., chanterelle → jack o'lantern)
• Apply never‑eat rule (always recommend local mycologist)
• Display candidate match or safety warning
• Optional: fetch weather, nearby observations, and rarity via Open‑Meteo, iNaturalist, GBIF APIs

Evidence

Every foraging app wanted a server, an API key, and a monthly fee. That's backwards. On a trail, a cloud app is useless exactly when you need it most.

Source

One HTML file. No backend. No API keys. No data leaves your device.

Source

The app never says "safe to eat." It says "candidate match." Three tiers: Confidence gate, Look‑alike blocklist, Never‑eat rule.

Source

Right: Identified a puffball at 82%. Flagged the young‑amanita look‑alike warning. The Touch Grass timer worked - I put my phone down and found a slime mold I'd have walked past.

Source