{"title":"FAQ","html":"\u003cp\u003eHyper, from \u003ca href=\"https://hyper.charm.land\"\u003eCharm\u003c/a\u003e is high-quality, open-weight AI inference built for\nagentic coding. We focus on performance, reliability, and developer experience,\nwhile working to make great inference more affordable for everyone.\u003c/p\u003e\n\u003ch2 id=\"why-open-weight-models\"\u003eWhy open-weight models?\u003c/h2\u003e\n\u003cp\u003eBecause, they’re very good, they’re cost-efficient, and we can run, optimize,\nand train them ourselves. Open-weight models give us the freedom to build the AI\nplatform we want without depending on frontier labs.\u003c/p\u003e\n\u003ch2 id=\"why-hyper\"\u003eWhy Hyper?\u003c/h2\u003e\n\u003cp\u003eAgentic coding is about more than inference, and at the same time, inference\nquality matters immensely. We care about fast, reliable, high-quality inference\nand a great developer experience. We're building Hyper as a platform to support\nthe entire agentic coding workflow.\u003c/p\u003e\n\u003ch2 id=\"how-does-hyper-deliver-inference\"\u003eHow does Hyper deliver inference?\u003c/h2\u003e\n\u003cp\u003eHyper runs a mix of dedicated inference and inference through close partnerships\nwith cloud, silicon, and inference providers. We work across the stack to get\nthe most out of open-weight models. Models are typically served using FP8 or\nNVFP4 precision.\u003c/p\u003e\n\u003ch2 id=\"whats-prism\"\u003eWhat’s Prism?\u003c/h2\u003e\n\u003cp\u003ePrism is a first purpose-built model for agentic coding. It's a decision model\nwith an 8B parameter backbone, which reads the prompt like an LLM, and routes\nthem to the most appropriate model for the task. Simpler tasks and routine work\nwork goes to smaller, faster models, while more demanding tasks get the bigger\nmodels they need. The model is being constantly retrained for new models as they\nbecome available.\u003c/p\u003e\n\u003ch2 id=\"why-does-charm-train-its-own-models\"\u003eWhy does Charm train its own models?\u003c/h2\u003e\n\u003cp\u003eBecause AI inference is still way too expensive for individuals, and we want to\nchange this. We believe there's a new class of models built specifically for\nagentic coding that can make this possible, and we're working to make it happen.\u003c/p\u003e\n\u003cp\u003ePrism, the routing model, is actually our first step in that direction. By\ntraining models on real sessions, we believe we can make high-quality,\nopen-weight inference way more affordable.\u003c/p\u003e\n\u003ch2 id=\"how-do-subscriptions-work\"\u003eHow do subscriptions work?\u003c/h2\u003e\n\u003cp\u003eThere are two kinds of subscriptions: Shred for individuals and Business for\nteams or anyone who wants ZDR. Every plan includes a weekly Hypercredit\nallowance, a rolling five-hour usage limit, and a concurrency limit. Bigger\nplans give you more of all three. Your five-hour capacity replenishes\ncontinuously, while your weekly Hypercredits reset every Sunday.\u003c/p\u003e\n\u003ch2 id=\"how-do-five-hour-rolling-windows-work\"\u003eHow do five-hour rolling windows work?\u003c/h2\u003e\n\u003cp\u003eYour five-hour capacity replenishes continuously throughout the day as long as\nyou have Hypercredits remaining in your weekly allowance. As the window moves,\nearlier usage falls outside it, freeing up capacity in real time.\u003c/p\u003e\n\u003cp\u003eThere are no resets every five hours. The rolling window simply looks at your\nusage over the previous five hours to calculate availability. Usage is counted\nwhen each request completes. For example, usage recorded at 10:00 stops counting\ntoward your five-hour limit at 15:00. Usage recorded at 10:15 stops counting\nat 15:15, and so on.\u003c/p\u003e\n\u003ch2 id=\"how-do-concurrency-limits-work\"\u003eHow do concurrency limits work?\u003c/h2\u003e\n\u003cp\u003eConcurrency determines how many requests you can run at the same time. As\nrequests finish, you can start new ones. Larger plans support more concurrent\nrequests.\u003c/p\u003e\n\u003ch2 id=\"when-do-my-weekly-hypercredits-reset\"\u003eWhen do my weekly Hypercredits reset?\u003c/h2\u003e\n\u003cp\u003eEvery Sunday at 00:00 UTC. Unused subscription Hypercredits don't roll over to\nthe following week. If you need more credits, you can purchase prepaid\nHypercredits to keep going without waiting for your subscription limits to\nrefresh, or you can upgrade your plan.\u003c/p\u003e\n\u003ch2 id=\"whats-a-hypercredit\"\u003eWhat’s a Hypercredit?\u003c/h2\u003e\n\u003cp\u003eHypercredits are how you pay for inference on Hyper. One Hypercredit is worth\n5¢. We could have called them nickels, eh?\u003c/p\u003e\n\u003ch2 id=\"can-i-use-hyper-without-a-subscription\"\u003eCan I use Hyper without a subscription?\u003c/h2\u003e\n\u003cp\u003eTotally. You can purchase prepaid Hypercredits starting at five bucks, use them\nwhenever you like, or enable automatic top-ups with an optional cap.\u003c/p\u003e\n\u003cp\u003ePrepaid Hypercredits are held separately from your subscription allowance. You\ncan see both balances in the app. They never expire, always use ZDR, and aren't\nsubject to concurrency limits. Go crazy.\u003c/p\u003e\n\u003ch2 id=\"wheres-the-free-plan\"\u003eWhere’s the free plan?\u003c/h2\u003e\n\u003cp\u003eWe've retired the monthly free Hypercredit allowance. You can still use Hyper\nwithout a subscription by purchasing prepaid credits whenever you need them for\nas little as five bucks.\u003c/p\u003e\n\u003ch2 id=\"does-hyper-train-on-my-data\"\u003eDoes Hyper train on my data?\u003c/h2\u003e\n\u003cp\u003eIt depends on your plan and credit source.\u003c/p\u003e\n\u003cp\u003ePrepaid Hypercredits (including automatic top-ups) always use Zero Data\nRetention. Business plans are also ZDR. In both of those cases, we'll never\nstore prompts or train on your data.\u003c/p\u003e\n\u003cp\u003eShred subscriptions don't include ZDR. Prompts and responses may be used to\ndevelop, train, and improve Charm's models. We use automated tools to remove\npersonal information before training and won't sell your data. Still, if you're\ndoing very sensitive work, it's worth considering using a ZDR-based credit\nsource.\u003c/p\u003e\n\u003cp\u003eAlso, nothing submitted before October 15, 2026 will be used for training.\u003c/p\u003e\n\u003ch2 id=\"im-on-the-original-shred-plan-what-happens-to-my-existing-subscription\"\u003eI’m on the original Shred plan. What happens to my existing subscription?\u003c/h2\u003e\n\u003cp\u003eYour plan is still active through the end of your billing cycle. At the end of\nthe billing cycle, you'll be switched to Pay As You Go, at which point you can\nchoose a new plan.\u003c/p\u003e\n\u003cp\u003eIn the meantime, your existing subscription remains covered by the same ZDR\nprotections for as long as your legacy plan is active. Nothing you send under\nthat plan will be used for training. That would be crazy!\u003c/p\u003e\n\u003ch2 id=\"any-other-questions\"\u003eAny other questions?\u003c/h2\u003e\n\u003cp\u003eWe're here to help. Email us at\n\u003ca href=\"mailto:hypersupport@charm.land\"\u003ehypersupport@charm.land\u003c/a\u003e or come chat in #hyper\non \u003ca href=\"https://charm.land/discord\"\u003eDiscord\u003c/a\u003e.\u003c/p\u003e\n\u003cp\u003eYou can also read our \u003ca href=\"https://hyper.charm.land/terms\"\u003eTerms of Service\u003c/a\u003e and\n\u003ca href=\"https://hyper.charm.land/privacy\"\u003ePrivacy Policy\u003c/a\u003e.\u003c/p\u003e\n"}
