This is Hyperinference

Build your next thing with purpose-built AI inference for coding, from Charm.

Start free Log In

Charm™ Hyper is an AI inference offering for AI coding. It works great with Crush, but you can use it with any tool. It’s fast and cost effective with a focus on privacy and open source.

What it is

Models as they were designed to be served When you call a model through Hyper, you get what the lab shipped: the same reasoning options, the same defaults, and the same behavior.
Cost-efficient Hyper’s typically more cost efficient than leading model providers. We optimize cost by working with cloud and silicon providers to optimize costs at the kernel and hardware levels.
Optimized for coding agents Hyper is built specifically for agentic coding. We want to make long, detailed coding sessions within reach and cost-effective.
Built for teams Master keys for the account and sub-key per developer, per project. See budgets, rotate keys, and see usage by user. Hyper is built for teams who want better governance around budgets and usage.

Privacy is a first class feature

No training on your data Your prompts and your outputs are never used to train any model.
No retention Prompts and outputs aren’t stored. In the event they’re held for debugging, abuse, or legal compliance they’re deleted within 30 days.
You own whatcha generate Charm assigns all right in generated output to you.
GDPR-ready Charm acts as a data processor for your content. DPA available on request.

Pricing

Hyper uses prepaid credits, called Hypercredits. 1 Hypercredit is 5¢ in token costs, though Hypercredits get cheaper if you buy ’em in bulk.

Free: Everyone gets 100 Hypercredits, every month
10-pack: $10 buys you 500 credits
20-pack: $20 buys you 2000 credits
Subscriptions: Coming soon! Predictable pricing for individuals and teams.

Bundles are prepaid to avoid surprise bills at the end of the month, and bundle credits never expire.

How we do it

We’re particular about how inference is run against AI models. We maintain close partnerships with AI cloud and silicon providers who give us control over the infrastructure on the hardware and kernel levels. We also maintain multi-region support to cover Charm’s diverse user base and proudly serve inference from North America, South America, Europe, Asian, the Middle East and North Africa.

Have questions? Want to say hello?