Blog
Running an Open Model on a Rented GPU, Without the Trial and Error
Picking a GPU and vLLM settings for an open-weight model usually takes guesswork. ThetaServe derives them from the model's metadata and live GPU prices, then launches a secured endpoint that stops itself.
When Building Is Cheap, What Is a Product For?
AI makes custom software cheap to build and change. What that does to the build-versus-buy decision, and to products built by generalizing their customers' requirements.
Will It Run? How to Tell If an LLM Fits on Your Computer
Four simple formulas to estimate whether a local LLM fits in your memory, how fast it will run, and whether you need a GPU.
How Big Is a Token? Measuring What Actually Travels on the Wire
How many characters, UTF-8 bytes, JSON envelopes, and network packets does it take to deliver a single LLM token? We measured 10 tokenizer families to find out.
Why a Small Local Model Can Write the Right Shell Command
Small models know what find and sed do, but not which options your system accepts. ThetaTerm gives the model your own man pages, checks its answer, and asks before anything runs.