Pydantic AI Gateway adds Jev, TypeSafe's classification model
Pydantic has made Jev, TypeSafe's classification model, available through Pydantic AI Gateway. Applications can call it with the same gateway key they use for language models, and Jev requests appear in the same spend view, limits and guardrails.
- Jev, TypeSafe's classification model, is now available through Pydantic AI Gateway.
- Jev returns an answer and a probability for each typed question about a piece of text.
- Requests appear in the same spend view, limits and guardrails as language models, under one gateway key.
- TypeSafe is a bring-your-own-key provider, set up in Pydantic Logfire under Gateway, then Providers.
- Pydantic AI supports Jev through TypeSafeModel, where each output field becomes a question.
What Jev does and how it is set up
According to Pydantic, Jev is not a chat model. It takes a piece of text and a set of typed questions, and returns an answer and a probability for each. Pydantic notes that David Montague has already shown how to use it as a judge in Pydantic Evals, at $0.042 per million input tokens.
TypeSafe is a bring-your-own-key provider. In Pydantic Logfire, users go to Gateway, then Providers, and add a provider, choosing TypeSafe AI. After pasting a TypeSafe API key and clicking Test & continue, the Gateway asks TypeSafe for its models with that key. A bad key therefore fails at this step and not on the first request. Once the provider is named, setup is finished.
The Connect dialog defaults to jev-latest and provides a snippet with the user's gateway key already filled in.
Calling Jev through the Gateway
Jev uses TypeSafe's System One API rather than the OpenAI chat format, so the Gateway forwards requests natively. A request goes to the /proxy/typesafe/v1/systemone path on the Gateway, authorized with the gateway key. The word typesafe in the URL is the route name given to the provider.
The request body contains the model name, a state field and a questions object. The state is the material being judged, and each entry in questions is one question about it. In Pydantic's example, the text "Help! My payouts have been failing for 3 days." is paired with a question named is_urgent that asks whether the text conveys urgency.
The TypeSafe Python SDK also works. After installing it with uv add typesafe-sdk, users point base_url at the Gateway and call system_one with their questions. The answer for each question can then be read from the response.
What the Gateway adds
Pydantic says the Gateway does not change what Jev costs or how fast it answers. It adds the same controls available for other providers:
- Cost and usage. Every request is metered, so Jev appears in the same spend view and limits as language models.
- Guardrails. Input guardrails run on the state, the instructions and the criteria before the request goes upstream.
- One key. The application holds a gateway key, not a TypeSafe key.
Using Jev from Pydantic AI
Pydantic AI can run an agent on Jev through TypeSafeModel, where each field of the output type becomes a question. It is installed with the typesafe extra: uv add "pydantic-ai-slim[typesafe]". To route traffic through the Gateway, the provider is given the gateway key and URL.
In Pydantic's example, a Ticket model has an urgent boolean and an area field limited to billing, bug, account or other. Given the message "You have charged me twice and my account is now overdrawn.", the agent returns urgent=True and area='billing'.
What it means for users
For teams already routing model traffic through Pydantic AI Gateway, Jev can be added without handling a separate TypeSafe key in the application. Classification calls are metered and limited alongside language model calls, and input guardrails apply before requests reach TypeSafe. Pydantic invites users who hit problems to open an issue or contact it in Slack.
Pydantic: Jev is now available in Pydantic AI Gateway
