Developer tools
Small tools, taken further
A growing set of utilities, each a simple idea built sharper. It starts with the Cookbook, which works out what a model needs and what that costs here, and grows from there. Every figure a live tool shows is computed from published specs and our own rate card, never a benchmark we imply we ran.
The Cookbook
Answer one question: can I run this model, and what will it cost? Pick your card and see what runs on it; pick a model and see which of the 7 cards holds it, how fast, and the price per million tokens. A one-line command reads the card you own. 57 models are in it, and it needs no account.
Open the cookbook- models sized
- 57
- cards priced
- 7
- benchmarks run
- 0
Model directory
The full catalogue of open models we can size, in one searchable place, filtered by kind. Each shows what it is, the memory it needs, and the cheapest card to rent it on. It links straight into the Cookbook to fit any one in detail.
OpenPrice comparison
Pick a card and see what we charge for it against the clouds and the open market, per hour and per region. Sorted by price, filterable, and drawn from the same rate card the rest of the site quotes, so nothing here can disagree with a figure you saw elsewhere.
OpenKracht Bazaar
The catalogue of tools an agent can call, grouped by capability with their endpoints and schemas: web, browser, data, a sandbox, and more. Browse it like a marketplace, import an OpenAPI spec, or add your own. The Utilities already run for real and free; the rest are on the way.
Coming soonKracht Skills
A skill is a SKILL.md plus the scripts and references it needs. Load one and an agent gains its instructions and the tools it names; save a workflow you liked and it becomes a skill the next agent can load. Loading and attaching need the agent runtime underneath.
Coming soonKracht Agents
Give a model a prompt, the tools it may call, the skills it has loaded and a memory of the session. The loop is the simple one: it decides, a tool runs, it reads the result and goes again. The model can run locally or on a rented card the Cookbook sizes.
Coming soonPlayground
Pick a model and a card, set a prompt, temperature and tools, and see the answer with the memory it takes and the speed it runs at. Memory and speed come from the Cookbook's arithmetic; running a live model needs the runtime underneath.
Coming soonCompare
Put a handful of models in the same row and read the trade-off: how fast each runs, the memory it takes, and what a million tokens costs, with the best value in each row marked. Speed, memory and cost come from the same arithmetic the Cookbook uses.
Coming soonPresets
A named bundle of settings: a model, precision, context, runtime, temperature and tools. Start from Fast, Low VRAM, Long context or Coding, or save your own. It is just JSON; applying one loads it into the Playground.
Coming soonModel caller
One key in front of many open models, each request routed to the cheapest card in the catalogue that can serve it, and metered by the second. It is where the cookbook's arithmetic naturally ends: it does not just tell you which card to rent, it rents it per call.
Coming soonThe live tools compute from published parameter counts and each card's published memory bandwidth, and the estimates err upward on purpose: a fit that turns out wrong costs a rental that runs out of memory a minute in and bills for it anyway. The coming-soon ones are named here because leaving them out would be its own kind of dishonesty, and held rather than linked because shipping them as if they worked would be worse.