Promptivore came out of a conversation with my wife about the Bolt hackathon. She’s a climate activist, I’m a tech builder and we were talking about AI.
I’ve been working with large language models, and I’m fascinated by what’s possible with them. But I also had this nagging thought: what is the environmental cost of all these prompts?
So I started digging.
There wasn’t a simple answer. A few academic PDFs here, some wildly different numbers in blogs, nothing usable for the average person. It wasn’t easy to calculate your AI footprint unless you knew how to parse energy usage in kWh, translate that into grams of CO₂, then tie that back to token usage across different models.
That’s when I realised this would make a great hackathon project. The world’s biggest AI hackathon was running on Bolt and I needed an idea. Within an hour, I had a name, a blank canvas and a mission: build a lightweight, friendly tool that helps people track their AI usage and get a sense of its environmental impact without feeling guilty about it.
Building Promptivore in 48 hours
The goal was to build something complete in two days, from scratch, that worked well enough to be launched. I wanted real data, clean UX, and something quirky enough to stand out but practical enough to be useful.
Stack overview:
- Frontend & UI: Built entirely in Bolt, which made it possible to move incredibly fast. I didn’t need to manage a separate frontend repo or fuss about routing or state, Bolt’s visual builder and file system let me keep momentum.
- Backend: Supabase handled all the backend tasks. That included:
- User authentication (login, registration, and RLS policies)
- Relational database tables for AI usage sessions and user stats
- Storing session history with timestamps, tokens, models, task types, and CO₂ data
- XP calculation logic
- GitHub: I used GitHub integration for version control and sharing the project with a few testers during dev.
- SEO: Keywords like “sustainable AI” and “carbon footprint calculator” were integrated into the landing page to improve discoverability, tracked with dev.post analytics during the event.
How it works
Users log their AI sessions, either manually or via a tracker form. For each session, they enter:
- The model they used (e.g. GPT-4, Claude 3 Sonnet, etc.)
- Number of tokens used
- The task type (creative writing, code help, research, etc.)
From there, Promptivore calculates the estimated CO₂ emissions using researched values per model. This was one of the harder parts to get right, I had to dig through various whitepapers, model specs, and benchmarks to get semi-reliable CO₂-per-token estimates for the most common models. The app includes context notes for each model to clarify how these values are determined.
After each session, users are shown their session impact in relatable terms, e.g. “equal to charging your phone 17 times”, and are given tips for optimising future sessions.
Features I managed to squeeze in
- Real-time CO₂ calculation per session
- XP and gamified eco-badges that reward efficient use (e.g. lower impact models, fewer tokens, etc.)
- A dashboard with interactive charts and breakdowns:
- CO₂ trends over time
- Model usage breakdown
- Task types
- Token consumption
- Session history with filters, sorting, and exports
- Carbon offset calculator and a carousel of tips for greener use
- Custom badge generator with a downloadable badge for users to embed on their own site
- Lottie animations and light gamification for delight
What I’m proud of
Two things: the clarity and the speed.
The UX is tight. It feels playful and welcoming, which was important — I didn’t want this to feel guilt-driven or preachy. It needed to educate without judging.
The other thing: building this whole thing in under 48 hours while staying close to the original idea. I didn’t pivot, I didn’t overbuild, I didn’t hit major blockers. That’s rare.
What’s next
There’s still plenty to polish — auto-tracking, team-level stats, streaks, achievements, and live model benchmarks — but the core tool is here. I’ve already got testers using it.
You can try it out at promptivore.site
Interested in more articles?
Read more here
