Key concept learned:
Jev is a hosted API model, not something installed locally. What you install (typesafe-sdk, pydantic-ai-slim[typesafe]) is just a client library — code that formats requests and talks to Typesafe’s servers. The actual model runs entirely on their infrastructure. Test: if cutting your internet breaks it, you’re calling something remote.
What it does differently from a normal LLM:
Instead of generating text, Jev takes a state (context) and questions (with types Choice, Score, or Noul) and returns structured, probabilistic answers — e.g. a chosen category plus confidence and a full probability breakdown across options.
Practical steps you walked through:
- Installed
pydantic-ai-slim[typesafe]— pulled intypesafe_sdkas a dependency - Got an API key from console.typesafe.ai
- Learned the correct SDK usage by inspecting the actual package (
TypeSafeClient,system_one()method,Choiceclass) rather than guessing names - Ran a real request classifying a support message (“payouts failing”) — got back
billingas the top choice with 88% probability vs. 12% technical, 0% sales - Practiced key hygiene: rotated an API key after accidentally pasting it in full, and reconfirmed the new key worked
Security habit reinforced:
Never paste real credentials into a shared chat — type them directly into your terminal instead, and rotate immediately if one leaks.
Want this saved as a note in your cyber range or blog docs, or kept just as a chat recap?