FAQ
Frequently Asked Questions about Jev
Short answers for agent developers — including API keys, pricing, and whether you can run Jev locally. For deeper setup advice, read the getting started guide or the Access & Use Jev section.
What is Jev AI?
Jev is the System One decision model released by TypeSafe AI. It is designed to produce reliable structured decisions for AI agents, with the core idea "Decisions not strings".
Is Jev a replacement for ChatGPT style LLMs?
No. Jev is not a full replacement for general purpose LLMs. It is built for structured decision tasks. Most agent architectures use Jev together with general LLMs.
Who should use the Jev SDK?
AI agent developers, backend engineers building multi-agent systems, teams building agent routing, scoring and classification workflows.
What does "Decisions not strings" mean?
It means Jev prioritizes stable structured decision outputs instead of generating free-form text strings, reducing unpredictable outputs inside agent pipelines.
Is this website official TypeSafe AI website?
No. This is an independent community resource. We are not affiliated with TypeSafe AI.
What is the difference between System One and System Two in AI?
System One (Jev) is built for fast, consistent decision making. System Two refers to generative LLMs built for flexible reasoning and writing. Many agent stacks combine both.
Can I replace GPT / Claude with Jev?
No. Jev is a specialized decision model. It cannot replace general LLMs. They complement each other in agent pipelines.
Where can I get a Jev API key?
You can apply for early-access waitlist at official TypeSafe AI website. Jev is also available via third-party AI gateways including OpenRouter, Vercel AI Gateway and Cloudflare AI.
How much does Jev cost? Are output tokens free?
Input tokens cost $0.042 per million tokens. Output decision tokens are completely free, you will not be charged for model output. Gateway providers may apply additional markup.
Can I download Jev model weights and run Jev locally on my own hardware?
No. At present Jev is API-only service. There are no public downloadable model weights for local self-host deployment. All inference runs on cloud.
What are Choice, Score and Noul primitives inside Jev?
They are three core question types for Jev API request. Choice selects options from predefined list; Score returns graded rating value; Noul returns yes-or-no probability judgement.
Source reference: