jev.proField guide · curious → Pro

Open · Last updated

Jev for platform teams

Platform teams usually want one decision microservice (or library) that product agents call. Jev is a good fit as that library’s model, not as the agent host. This unofficial open-cluster page is for internal platform owners. Replicas here are not Jev and not endorsed. Confirm APIs on docs.typesafe.ai. We do not issue keys.

This page is for platform owners crossing from curious to production-minded Jev use. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. This is not a TypeSafe playbook.

Open-cluster pages are independent field-guide notes. Replicas and third-party interfaces mentioned anywhere on jev.pro are not Jev and not endorsed.

Install the skill where agents write code

Official install (Claude Code / other agents) lives on the agent skill page. Platform rule: one install method per repo; pin skill updates; reject PRs that invent request fields (stale skill symptom).

TypeSafe’s own review advice: put questions and thresholds in one file so humans can diff them. Agents are weak at writing criteria — expect to edit. Criteria are descriptions a person could apply. Structured instructions may be objects or arrays; backtick field paths.

Teams moving off a shared mega-prompt should dual-run through the migration guide before you flip the platform route() to Jev-owned auto-act.

Shared contract

  1. Pin jev-1.13.0 in production config; aliases in dev only.
  2. Log model, usage, probabilities, policy decision.
  3. Door-specific model ids stay in an adapter (channel matrix).
  4. Product teams consume your route() enum, not raw Nouls.

LangChain depth: optional langchain-typesafe middlewares for model routing and tool gates — harness. Treat as an integration, not the platform. That blog is not Jev and is not endorsed as official TypeSafe architecture.

First-party evaluation is POST /v1/systemone plus GET /v1/models. Do not let product teams invent /classify or /score HTTP routes on your gateway unless they are adapters that still call System One. Compose atomic questions in code; one POST, many questions.

Your route() enum is the action space. Persist a decision trace on the platform side so product apps cannot “forget” to log. Wire-up details: implementation checklist. Ship gates: launch checklist.

Jev is the router, not the actor

Coding agents, browser-use, and computer-use still execute tools. Jev answers “may I?”, “which model?”, “which queue?”. See vs agent loops.

Choice picks an unordered option (max 255; add other / none). Score places a state on 2–10 levels — keep arithmetic in the platform; do not treat score as dollars. Noul is P(yes); 0.5 is uncertain, not medium. Confidence is peakedness of Choice/Score; Noul has none. Product teams must not copy a Noul τ onto a Choice winner.

Production gates (SEO-GEO)

Scope: when to use Jev. If the platform becomes the agent host, you are in an anti-pattern. Hub: Open. Sibling: implementation checklist.

What this page does not claim

FAQ

Should product teams call System One directly?

Prefer one platform route(). Direct calls fragment pins, floors, and traces.

Is the LangChain harness the platform?

No. It is an integration. Your enum and traces stay yours.

When do we migrate existing agent prompts?

Shadow-mode dual-run first. Follow the migration guide.

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.

Sources