Hi Product Hunt, I built Prisnoir because I was frustrated by how quickly existing $20 AI subscriptions become restrictive during real agentic work. Kimi K3 is capable enough to compete in serious coding and research workflows, but running it locally is impractical for almost everyone. API pricing also makes long sessions difficult to use without constantly thinking about token costs. Prisnoir turns that into a straightforward subscription. We host Kimi K3 using its native MXFP4 weight and MXFP8 activation configuration. Instead of maximizing speed for a small number of users, we are optimizing the system around high cache reuse, sustained concurrency, and generous monthly usage. Our initial target is approximately 30+ tokens per second. That is intentionally a little slower than some premium services, but the tradeoff is that users should be able to complete far more actual work before reaching their limits. I would especially appreciate feedback from people who regularly use Claude Code, Codex, Kimi Code, OpenRouter, or other coding and research agents. Website: https://prisnoir.com

Prisnoir

Generous Kimi K3 usage and powerful self-improving harness

Generous Kimi K3 usage and powerful self-improving harness