r
realdavidmckee
Request: Add TypeSafe AI's Jev to Hatz as a selectable model, in its own "System One" category separate from the chat/LLM models.
What Jev is:
Jev (released Sept 15, 2026, early access) is TypeSafe AI's first "System One" model. It doesn't generate text. You give it state plus typed questions, and it returns structured answers (choices, scores, yes/no) with calibrated confidence. All questions are answered in parallel in a single pass.
Why it matters:
  • Cost: $0.042 per million input tokens, output is free. That's a small fraction of what we burn today using full LLMs for simple decisions.
  • Speed: TypeSafe reports 40-200x faster than frontier LLMs on decision tasks, with responses in the 70-500 ms range.
  • Reliability: output is constrained to the answer set you define, so there's no free-text parsing and no "creative" answers breaking a workflow.
MSP use cases (we'd use these immediately at Innovative):
  • Ticket triage: classify queue, priority, issue type, and sub-issue type on inbound PSA tickets
  • Alert filtering: decide whether an RMM alert is actionable before it becomes a ticket
  • Routing: pick which Hatz model/agent should handle a request, cheap model for simple tasks, capable model for complex ones
  • Security: phishing/spam verdicts on user-reported emails
  • Guardrails: score or flag LLM output before it goes to a client
Why a separate category:
Jev is a decision engine, not a chat model. Putting it in the normal model picker would confuse users expecting text back. A "System One" category makes it clear it's meant for workflows, agents, and automations, where it pairs with a generative model instead of replacing one.
Access: available via TypeSafe's API, LangChain (TypeSafeClassifier), and Vercel AI Gateway.