When to use Jev
Use Jev when a knowledgeable person could answer in a few seconds from text you already have. Skip it when the work is exact compute, long reasoning, generation, or a regulated attestation.
Independent “when” page — not TypeSafe’s use-case map and not a chatbot pitch. Official: System One vs tasks that need extended reasoning. docs.typesafe.ai. Replicas are not Jev and not endorsed. We do not issue keys.
Independent angle (cover, do not clone)
Win delta (jev-vs-llms): hybrid decide-then-draft; cost/latency honesty; when not to use System One — Kahneman framing lives on the explainer, production “when” lives here.
Use Jev
- Intent / queue / tool / model routing
- Policy and injection screens (not a WAF)
- Completeness / consistency flags on text packets
- Composite Scores you weight in code
- Confidence-gated auto vs review
Do not use Jev
- Math, counts, date order (jaggedness)
- Image/audio/video as raw input
- Being the agent, the ESP, the bidder, or the KYC vendor
- On-device offline
- “Write the email”
Compose with LLMs
Decide (Jev) → draft (LLM) → optional second Jev screen on the draft. Label every latency and dollar figure as vendor unless you measured it.
A one-line test
“Could a sharp teammate answer this in a few seconds from this text, and can our code act on a label?” If yes, Jev is in-scope. If you need a proof, a picture, a generated paragraph, or a legal attestation, it is not.
Snap language judgment, then draft
The one-line test above is the scope gate. In-scope usually means intent / queue / tool routing, policy and injection screens (not a WAF), completeness flags, composite Scores you weight in code, confidence-gated auto vs review.
Compose with LLMs as decide then draft: Jev first, LLM second, optional second Jev screen on the draft. Label every latency and dollar figure as vendor unless you measured it. List price $0.042 / M input (confirm models page) is not a reason to generate on the decide hop.
Kahneman framing lives on System One vs Two. Motivation lives on why use Jev. Prompt-only systems comparison: vs prompt-only.
Boundaries the short list undersells
The “Do not use Jev” bullets above cover math, generation, media, and “be the whole stack.” Three more that launch reviews miss:
- Persistence — Jev is not a database, feature store of record, or audit log. You store traces.
- Regulated attestation — clinical protocols, hiring, credit, identity proofing, automated legal advice: start from no unless counsel designs a narrow admin hop.
- On-device offline — hosted Jev is a door you call. A laptop clone is not Jev.
Noul 0.5 is equally likely yes/no — not a reason to auto-act. Schema-safe ≠ the teammate would have been correct. Pin jev-1.13.0 if you ship floors.
When not to use this page’s advice
Do not treat the lists as TypeSafe’s complete use-case map (see their use-case page and our /usecases). Do not stretch “snap judgment” to cover proofs, pictures, or licenses. Do not evaluate SemIf as a “when to use Jev” substitute.
What this page does not claim
- Not a complete use-case map — see TypeSafe’s use-case page and our /usecases.
FAQ
System Two work? Keep it in tools and checkers. See System One vs Two.
Healthcare / hiring / credit? Assume no unless counsel designs a narrow admin hop. See those comparison pages.
Can Jev store the decision for next time? No. Your trace store persists. Jev evaluates the state you send now.
Decide-then-draft if the draft is the product? Still decide first (route, safety, intent). The LLM writes. Jev does not.
Disclaimer
This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.
Open-cluster pages are independent field-guide notes. Replicas and third-party interfaces mentioned anywhere on jev.pro are not Jev and not endorsed. Hub: Open. Siblings: why use jev, system one vs system two, vs prompt only systems. Canonical: https://docs.typesafe.ai.