Pydantic

@pydantic

The Pydantic Stack: Logfire, Monty, Validation, AI, Evals, and AI Gateway

London

Pinned
Pydantic
@pydantic
Sep 23
One Jev request, three scores: a policy label, a rubric level, and the probability the reply asks for a password, all recorded in Logfire. Modeled cost of 1M evaluated replies: $62, vs $4,542 with Braintrust before data charges. Blogpost
Cheap AI scoring with Jev and Pydantic Evals
From pydantic.dev
6
Pydantic
@pydantic
Oct 31
Changing a dynamic system prompt no longer has to rewrite the start of your conversation. Pydantic AI can append the change to history, keeping the earlier prompt prefix available for caching. pydantic.io/uaGgo
3
Pydantic
@pydantic
Oct 31
Your agent falls back to a working model. On the next turn, it tries the failing primary again. ModelFallback's sticky option keeps the conversation on the model that answered. pydantic.io/j6Qq5
4
Pydantic
@pydantic
Oct 30
We're starting work on Pydantic V3. No V1-to-V2 style migration this time: breaking changes are limited to correctness fixes and things you can opt back out of. Also: V1.10.27 is the final V1 release, and V1 support has ended. pydantic.io/m6MR7
7
Pydantic
@pydantic
Oct 29
The UK Ministry of Justice, @justiceaiunit builds AI tools for judges and probation staff. Every structured AI response is validated with Pydantic before calling code sees it, and the same models generate the TypeScript client contract. Read the case study:
1