Launching today
Kontinuity
Never loose the thread - token and context health meter
1 follower
Never loose the thread - token and context health meter
1 follower
Every AI conversation has two sides. The one you see — and the one nobody shows you. The system prompt loading before you type. The context window quietly filling up. The session degrading while the interface looks identical. Kontinuity sits next to Claude and ChatGPT and shows real-time token costs, session costs, and context health — in dollars, on the page. The 98% of tokens you never typed but are paying for. Local-only. No accounts. 14-day free trial. $5/month after.
Hey Product Hunt! 👋
I'm Praveer, building solo from Kodaikanal, Tamil Nadu, India.
There's a specific weekend I can point to where Kontinuity started. Early March 2025. Vercel had just dropped v0. I was consulting for a company that needed a project management tool shaped around their actual workflow — not Asana, not Linear, something specific. I sat down on a Friday and started building.
The first few hours were electric. Features came together fast. The dashboard was clean. The AI was sharp. I kept going. By Sunday night I'd been at it for 56 hours straight, and somewhere in the middle of that, without me noticing, the build had quietly collapsed. The AI was confidently producing code that referenced functions that didn't exist. Bugs that had been fixed kept coming back. The whole thing was held together by hope. I abandoned it.
The frustrating part wasn't that it broke. The frustrating part was that I never saw it breaking. The interface looked the same at hour 4 and hour 50. The model was operating at full capacity at the start and barely holding the thread by the end, and nothing in the UI told me which state I was in. I'd been the last person to know.
I wrote about it here if anyone's interested: https://medium.com/@85.pac/the-most-expensive-thing-youll-build-is-the-project-you-never-ship-340d10161a46
That experience is what Kontinuity is, basically. A meter on the side of every AI conversation. Per-message and per-session token cost in dollars. System prompt overhead — the 22,000 tokens loading before you type "hello." A context health indicator that changes color before the quality silently degrades. Nothing leaves your browser; everything's local IndexedDB. I'm not interested in collecting your conversation data.
Built solo, with Claude as my primary collaborator across this and prior projects. 162 people installed it from the Chrome Web Store with zero paid marketing — that's the only reason I had the nerve to launch here.
Two asks, honestly:
1. Install it and tell me what surprises you. If the dashboard shows you something about your own usage that you didn't expect, I want to hear that specifically.
2. Tell me what's missing. The roadmap is shaped by what real users actually need, not what I find interesting.
Happy to AMA on tokens, context windows, building solo with AI agents, MV3 service worker pain, the economics of $20 AI subscriptions, or what happens when you skip sleep for 56 hours.
Thanks,
— Praveer Concessao