OpenJev

Open source System One–style decision playground: state + typed questions → structured answers.

Three backends:

Mode	What it is
parallel	LLM micro-scorers (one tiny {p} call per option, then softmax)
oneshot	Single structured JSON call to any OpenAI-compatible model
decider	Mapika/decider — real System One weights (calibration-aware RL on v10)

Tutorial

https://youtu.be/xtXq279B4Go

Quick start (LLM backends)

cd OpenJev
npm install
cp .env.example .env   # set OPENAI_API_KEY
npm run dev

Open http://localhost:3001 — mode parallel or oneshot.

Integrate Mapika/decider (RLCD / System One)

decider is an open System One model family (Qwen3.5 fine-tunes).
decider-2b v10 includes calibration-aware RL. It speaks TypeSafe’s wire format: POST /v1/systemone.

1. Serve the model (needs a CUDA GPU, ~4 GB for 2B)

pip install "git+https://github.com/Mapika/decider#egg=decider[serve]"
# or: git clone https://github.com/Mapika/decider && cd decider && pip install -e ".[serve]"

scripts/serve.sh Mapika/decider-2b 8000

Smoke test:

curl -s localhost:8000/v1/systemone -H 'content-type: application/json' -d '{
  "state": "My card was charged twice.",
  "questions": {
    "team": {
      "type": "choice",
      "instructions": "Which team?",
      "criteria": { "billing": "charges, refunds", "technical": "bugs, outages" }
    },
    "refund": { "type": "noul", "instructions": "Is a refund needed?" }
  }
}'

2. Point OpenJev at it

In the UI: set mode = decider, decider url = http://localhost:8000

Or via API:

curl -X POST http://localhost:3001/api/evaluate \
  -H "Content-Type: application/json" \
  -d '{
    "mode": "decider",
    "decider_url": "http://localhost:8000",
    "state": "Charged twice again!!",
    "questions": {
      "department": {
        "type": "choice",
        "instructions": "Which team?",
        "criteria": {
          "billing": "Charges, refunds",
          "technical": "Bugs",
          "other": "Else"
        }
      },
      "angry": { "type": "noul", "instructions": "Strong frustration?" }
    }
  }'

Or in .env:

DECIDER_BASE_URL=http://localhost:8000

3. Python (no OpenJev) — direct

from decider.infer import Decider
d = Decider("Mapika/decider-2b")  # ~4 GB VRAM
print(d.system_one(
    "I was charged twice for order A-104.",
    {
        "department": {
            "type": "choice",
            "instructions": "Which team?",
            "criteria": {
                "billing": "Charges, refunds",
                "technical": "Bugs",
            },
        },
        "refund": {"type": "noul", "instructions": "Is a refund needed?"},
    },
))

How OpenJev wires it

UI / curl  →  POST /api/evaluate { mode: "decider", ... }
                →  OpenJev server
                →  POST {decider_url}/v1/systemone   (TypeSafe shape)
                →  Mapika/decider GPU server
                →  typed answers + probabilities

Same question types as TypeSafe Jev: choice, score, noul.

Environment

# LLM backends
OPENAI_API_KEY=
# OPENAI_BASE_URL=
# OPENJEV_MODEL=gpt-4o-mini

# decider backend
DECIDER_BASE_URL=http://localhost:8000
# DECIDER_API_KEY=local

Layout

OpenJev/
├── server/index.ts        # Express + Vite, POST /api/evaluate
├── server/loadEnv.ts
├── src/lib/evaluate.ts    # parallel | oneshot | decider
├── src/lib/types.ts
├── src/App.tsx
└── package.json