Jev vs centralized decisions
Centralized decisions means one service owns questions, thresholds, and traces. Jev is often the model behind that service — not a replacement for having a service.
Unofficial. Platform-team habits live on Jev for platform teams. docs.typesafe.ai.
Comparison scope
If every app embeds raw SDK calls, you will fork criteria and floors. Centralize the contract; still call TypeSafe (or a gateway) from that one place.
Criteria that decide the architecture
| Axis | Jev (System One) | Centralized decision services |
|---|---|---|
| Question git history | Your mono-repo module | Scattered prompts |
| Model pin | One config | Drift |
| Door adapters | One matrix | Copy-paste boolean vs noul bugs |
| Jev itself | The model | The model |
Decision quality and control
Centralization without eval is a single point of confident failure. Pair with offline evaluation.
Integration trade-offs
Apps call route(ticket) -> enum. The service maps enum ← Jev answers. Product teams never see raw Nouls unless they opt in.
TypeSafe’s public models page lists jev-1.13 at $0.042 per million input tokens with output tokens free — a vendor claim, not a jev.pro measurement. Confirm on the models page before you bid.
When each approach fits
Prefer Jev when
- You want one decision layer (Jev-backed)
Prefer Centralized decision services when
- A single app with three questions — a library may suffice
Central service + Jev model + edge caches if needed (see edge page).
One adapter, many callers
The pairing is not “Jev versus a decision service.” It is “raw SDK in every app” versus one service that owns questions, floors, and door adapters. Sketch: product apps call route(ticket) -> enum. The service maps that enum from answers on a pinned jev-1.13.0 pack. Boolean vs noul lives in one matrix. response.model and question SHA are logged once. Product teams never paste Playground JSON into a lambda.
TypeSafe’s $0.042 / M input list price is a vendor claim. Centralizing the adapter does not change the bill; it stops you from sending the same CRM row twelve times from twelve repos. Official how-to: code owns control flow. The agent skill (npx skills add typesafe-ai/skills) is for coding agents that call this service, not a replacement for it.
How a “central brain” still fails
- Centralization without eval. One bad criteria deploy hits every tenant. Pair with a gold set and offline evaluation before you lower floors.
- Forked questions anyway. A “temporary” copy in a growth squad becomes the second source of truth. Git history on the pack is the product.
- Treating Jev as a BRMS. There is no FICO-style rule UI here. Hosted model + your service. Multi-tenant criteria stay yours; Jev is not fine-tuned per tenant.
- Schema-safe enum as policy. A valid Choice is not “this is the right credit decision.” Your thresholds and humans still own irreversible acts.
- Generation of the decision memo. Jev does not write the explanation essay. Persist the question pack; that is the audit trail.
When not to use Jev here
If a single app has three questions, a library may suffice — do not stand up a mesh for that. If you need a business-rules workbench, buy a BRMS, not a model pin. If you must run offline without TypeSafe, this pairing does not help; see vs edge decisions.
What this page does not claim
-
Not an enterprise architecture sales pitch.
-
Schema-safe output is not the same as factual correctness.
-
No independent bake-off numbers live here.
FAQ
Is TypeSafe a decision manager like FICO Blaze? Different era, different contract. Jev is a hosted model, not a BRMS UI.
Multi-tenant criteria? Your service. Jev is not fine-tuned per tenant.
Does TypeSafe host the decision service?
No. Evaluation is POST /v1/systemone. The facade, floors, and traces are yours.
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.
Hub: Comparisons. Siblings: vs edge decisions, for platform teams, reference architecture. Canonical: https://docs.typesafe.ai.