"Do you wanna split this ice cream?" "No thanks, I'm watching my tokens."

Point your camera at any price tag. Find out what it costs in AI tokens. That $6 latte? 1.2 million Claude Sonnet tokens.

→ Try it live — no signup, no backend, keys stay in your browser.

1. Snap or upload a photo of anything with a price (tag, menu, receipt, screen)
2. Vision AI reads the price — or guesses it if there's no visible tag
3. Instant token breakdown across 10 AI models with a culturally-aware one-liner

• Camera-first mobile UI with live preview, flip camera, retake, and upload fallback.
• Two analysis modes:
  • Read tag for direct price extraction from images.
  • Guess price for estimated USD pricing when no tag is visible.
• Multi-provider own-key support in browser:
  • Gemini
  • OpenRouter
  • NVIDIA NIM
  • Cloudflare Workers AI (ACCOUNT_ID:API_TOKEN format)
• Shared-key mode via /proxy with password protection and provider fallback chain.
• Per-provider model selectors (Gemini, OpenRouter, NVIDIA, Cloudflare AI).
• Country-aware quip generation with two quip modes:
  • Normal mode: uses selected provider.
  • Simple mode: free backend quips via Workers AI.
• Token economics breakdown:
  • Hero number for primary model token equivalent.
  • Full comparison table across supported models.
  • Input/output/reasoning token split handling.
• Extra UX features:
  • Animated count-up results
  • Contextual fun facts
  • Share card image export
  • Setup/privacy drawers with key hints and safe defaults
• Zero-build static deployment on Cloudflare Pages, plus local preview with python3 -m http.server.

Bring your own key — all have free tiers:

Token counts are estimates based on an assumed split of 30% input / 20% reasoning / 50% output (40/60 for models without reasoning tokens). Real-world usage varies: chat workloads tend to be input-heavy due to long system prompts and conversation history; agentic and code workflows skew toward output and reasoning. The split is a general-purpose middle ground, not a measurement.

pip install -e .

tokeneyes photo.jpg              # read price from image
tokeneyes shoe.jpg --guess       # AI guesses the price (no visible tag)
tokeneyes --price 5.99           # skip vision, just convert
tokeneyes --list-models          # show all supported models
tokeneyes-web                    # start local web UI (port 8000)

Requires GEMINI_API_KEY or OPENROUTER_API_KEY in your environment or a .env file.

The web app is a single static HTML file — no build step, no server.

git clone https://github.com/disc0nnctd/TokenEyes
cd TokenEyes/cloudflare
python3 -m http.server 3000

Deploy to Cloudflare Pages by dragging the cloudflare/ folder to pages.cloudflare.com. See cloudflare/DEPLOY.md for full instructions including optional free quip generation via Workers AI.

• Receipt mode — scan a full receipt, see every line item as tokens
• Browser extension — hover any price on any webpage for an instant token tooltip
• Reverse mode — "I have $20 in API credits, what does that buy me in the real world?"
• Per-user agent consumption tracking — log your daily Claude Code / API usage and see it expressed as real-world costs ("this week's agent sessions = 3 lattes")
• AR glasses — real-time price-tag overlay when open camera SDKs become available (Meta Ray-Ban, etc.)

• Keys are memory-only in your browser — gone when you close the tab
• All API calls go directly from your browser to the provider — no TokenEyes server in the path
• Nothing is logged or stored by us

MIT