Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

openpave-jev

PAVE skill and CLI for typed Choice, Score and Noul decisions via local Laya, any /v1/systemone server, TypeSafe Jev or an EPM gateway.

Agent skillbeta
View on GitHub ↗

Overview

openpave-jev is a skill for PAVE (and Claude) that asks non-autoregressive decision models for typed answers with calibrated probabilities, for classifying, routing, scoring and gating without an LLM round trip. A jev decide command takes a state and candidates and returns the answer, with a --min-confidence flag that exits with code 3 when the model is unsure.

Providers include local Laya via @receptron/laya (ModernBERT-large, about 1.7 GB), any HTTP server exposing /v1/systemone, the paid TypeSafe Jev API, a pave-epm Decisions API gateway that works inside openpave without key handling, and a deterministic mock. Auto-selection prefers mock, then EPM, then a self-hosted URL, then a Jev key, then installed Laya, so self-hosted options win over the paid API. The README cites roughly 35-500 ms per decision.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich of these candidates fits the state?—
ScoreLevel on an ordered rubric—
Yes / noCalibrated yes/noyes · no

How it works

  1. Install the skill into PAVE.
  2. Choose or auto-select a provider.
  3. Run jev decide with state and candidates.
  4. Gate on confidence; exit code 3 when unsure.

Strengths

  • Prefers self-hosted providers over the paid API.
  • Exit-code gating for scripts.
  • Mock provider for testing.

Caveats

  • PAVE-specific install path.
  • Laya provider downloads about 1.7 GB of weights.

Getting started

git clone https://github.com/cnrai/openpave-jev ~/pave-apps/openpave-jev && pave install ~/pave-apps/openpave-jev, then jev providers.

At a glance

Type
Agent skill
Status
beta
Language
TypeScript
Runtime
PAVE skill, CLI
Has code
yes

Use cases and integrations

How curated lists describe it

  • PAVE skill for non-autoregressive decision models (Jev / Laya / TypeSafe)…

    wh000wh000/awesome-jev-live

  • PAVE skill for non-autoregressive decision models (Jev / Laya / TypeSafe): typed Choice, Score and Noul answers with calibrated probabilities.

    hellogumbo/awesome-jev

Frequently asked questions

Which providers are supported?

Local Laya, any /v1/systemone server, TypeSafe Jev, a pave-epm gateway and a mock.

How do I gate on confidence?

Use --min-confidence; the CLI exits with code 3 when unsure.

Does it need a TypeSafe key?

No, local Laya or a self-hosted server works without one.

Related projects

  • system-one

    Provider-neutral TypeScript runtime: write choice, noul and score once and run them on Jev, Reflex or any System One endpoint.

    ServicebetaTypeScript
  • jev-mcp (freepik-company)

    Go MCP server giving any client typed Jev decisions via OpenRouter or TypeSafe, with validated answers and no paid retries.

    MCP serverproductionGo
  • askjev

    Unofficial hosted or local Jev MCP server giving agents a quick second opinion: it picks yes/no, scale or choice and returns probabilities.

    MCP serverbetaTypeScript
  • decidealot

    Docker service exposing local Laya, Von, a CLM head or hosted Jev through one TypeSafe-compatible HTTP and MCP endpoint.

    MCP serverbetaPython
  • gojev

    Go harness for System One models: hosted Jev, open Kev in-process via llama.cpp, and Cloudflare Clef, with typed helpers and a CLI.

    ServicebetaGo
  • instinct-ai

    Open-source dual-brain runtime that answers typed decisions with a fast local engine or the Jev API before falling back to LLMs.

    ServicebetaPython, TypeScript