TypeSafe AI decision API for typed automation

What does TypeSafe AI do in plain English?

TypeSafe AI gives developers a model called Jev for making small, well-defined judgments inside software. Supply text or structured text as the state, ask a typed question, and receive a value that code can use: a selected option, a score, or a probability. For example, a support app can decide which team should handle a ticket and whether it needs urgent review. Jev is designed for those decisions, not for writing an email or holding a free-form chat. The model introduction explains this input-and-output shape.

Where typed decisions help

A Choice selects from options you define and returns probabilities for each option plus confidence. A Score rates the state against ordered criteria. A Noul answers a yes/no question with a value from 0 to 1. One request can contain several independent questions about the same state. This suits ticket routing, content checks, relevance judgments, and scoring candidates before a separate system acts. Keep deterministic rules and calculations in your own code. Set a confidence threshold so uncertain cases reach a person instead of triggering an automatic action.

Try Jev, then call the API

Start in the official quick start: log in to the Playground, paste a sample ticket as state, add a Noul question such as whether the message is urgent, then compare it with a Choice for department and a Score for frustration. Inspect each returned value before designing your route. For an application, obtain an API key in the dashboard and send a JSON POST to https://api.typesafe.ai/v1/systemone with state, model, and named questions. Read answers under the same names and implement the threshold or fallback in code. Python and JavaScript SDKs are also documented.

Pricing and access as of September 23, 2026

The public Jev 1.13 model page lists US$0.042 per million input tokens (US$42 per billion); output tokens are free for that model. The customer agreement uses purchased credits and permits promotional credits at TypeSafe’s discretion. A permanent free allowance is not promised. The homepage announced on September 20, 2026 that the waitlist had ended and sign-up was open. Check local currency charges, taxes, and account terms before planning production spend.

Limits, language quality, and data handling

Jev 1.13 accepts text only, including strings or JSON objects/arrays of text values; images, audio, and video need preprocessing elsewhere. English is currently its strongest language, so test non-English examples on your own data. The documented jagged edges include arithmetic, date comparison, long irrelevant state, adversarial content, and open-ended generation. A typed response can still be a wrong judgment. TypeSafe says it does not train models on customer Input, but the privacy policy describes retention as reasonably necessary and US hosting; zero retention is an enterprise option, not a stated default.

Frequently asked questions

Is TypeSafe AI a chatbot?

No. Jev returns predefined typed decisions and probabilities for code. Use a generative model when the task needs free-form text.

How do I make a first Jev request?

Use the Playground to test a state and typed question, or get an API key and POST state, model, and questions to /v1/systemone. The quick start includes both routes.

Is TypeSafe AI free?

The public Jev 1.13 rate is US$0.042 per million input tokens, with free output tokens. Purchased credits fund use; promotional credits may be offered, but no permanent free tier is guaranteed.

Official resources and developer community

• Documentation and quick start — API setup and current reference
• GitHub organization — official SDKs and examples
• Discord community — developer discussion
• X updates — product announcements