One JuL to
rule them all.

ANY MODEL · ONE COMMAND · MEASURED

Any model, tested as a decider.

C:\> jul models add qwen3-emb \
       --repo ollama:qwen3-embedding:0.6b --backend api
qwen3-emb on api: center options, tau 0.05019
  dev accuracy 0.635 ± 0.034 (n=200); ECE 0.172; 2237 ms per decision
C:\> jul ask choice "Which team?" -o billing -o tech \
       --state "I was charged twice" --model qwen3-emb --backend api

A model on the Hugging Face Hub, an embeddings API (Ollama, OpenAI, Mistral, Voyage, any OpenAI-compatible), your own fine-tune: jul models add fits it on development sets and tells you how good a decider it is. Then the same code uses it. Some models are weak zero-shot: measure, keep the one that wins on your data.

Numbers above are from a real run (Ollama on a CPU, hence 2.2 s), output trimmed. Every model, backend and provider →

LOCAL FIRST · A BIGGER DECIDER WHEN UNSURE

Escalate only the hard ones.

question ──► local 4B (97 ms on an A10G, no API bill)
               ├─ confidence ≥ 0.8 ──► answer
               └─ unsure ──► Jev · Clef · Nimble · your server

On the decision bench (2,108 Choice, Noul and Score questions), with jul-decision-wemm-4b local and Jev as the next tier at a 0.8 bar: 80% of the questions stay local, and accuracy goes from 0.849 (local alone) to 0.866. Choice and Noul mostly stay local (95% and 89%); Score questions mostly go up (93%).

Computed from the recorded predictions of both systems on the bench, not a live run; the share depends on your data and your bar: measure yours with jul bench test.jsonl --models accurate --escalate-to typesafe. How to set it up →

ONE API · EVERY DECIDER YOU ALREADY HEARD OF

Run them, or route to them.

DEVICE              WHAT JUL DOES                          HOW
Kev models          runs them locally (decision.json)      jul models add NAME --repo HF_REPO
Kev server          routes unsure questions to it          jul serve --escalate-to https://...
Laya                runs it locally (its own runtime)      --model laya | laya:multilingual
Jev (TypeSafe)      routes unsure questions to it          jul serve --escalate-to typesafe
Any of them         benches them on your rows, with JuL    jul bench test.jsonl --models accurate,typesafe
Clef · Nimble       routes unsure questions to them        --escalate-to cloudflare | ollama
Embedding LLMs      runs them locally (MLX, PyTorch)       jul models add NAME --repo HF_REPO
Encoders            runs them on CPU, in a Lambda (ONNX)   --backend onnx
Embeddings APIs     reads them over HTTP                   --repo ollama:M | openai:M | mistral:M
Your code           unchanged                              from jul import TypeSafeClient

Same Choice / Noul / Score calls and the same response whichever runs. The hub, in detail →

0 tokens generated · 0 bytes sent · no key · $0 per call.

AG NEWS · BANKING77 · EMOTION · 300 EXAMPLES

Accuracy, zero-shot.

HOW IT ANSWERS

Read the mind, skip the mouth.

STATE "I was charged twice"
               │
┌──────────────┴──────────────┐
│ WeMM-4B 4-bit (your laptop) │
│ layers 1 → 31, then stop.   │
└──────────────┬──────────────┘
               │ hidden state
               ▼
      cosine(state, option) ◄── options
               │         (encoded once,
               ▼              cached)
       softmax(cos / τ)
               │
               ▼
billing    ████████████████  0.88
technical  ██                0.11
sales      ▏                 0.01

WHY LOCAL

Small model. Short answer. Your box.

• Private by construction. Tickets, emails, medical notes: the text never leaves the process.
• Latency you can loop over. 55 ms is cheap enough for every row, every request, every keystroke.
• No meter running. No API key, no rate limit, no invoice. Ten thousand calls cost electricity.
• Runs anywhere. MLX on Apple Silicon, PyTorch on CUDA, CPU or MPS.
• Your model. Swap the backbone, register your own with one command, fine-tune the head on your data.

ALREADY ON JEV?

Change one import.

- from typesafe_sdk import TypeSafeClient, Choice, Noul, Score
+ from jul import TypeSafeClient, Choice, Noul, Score

client = TypeSafeClient(model="wemm-4b-4bit")

r = client.system_one(
    state={"ticket": "I was charged twice this month."},
    questions={
        "team": Choice(
            instructions="Which team should handle this?",
            criteria={"billing":   "payments, refunds",
                      "technical": "bugs, errors",
                      "sales":     "pricing, plans"}),
        "is_bug": Noul(instructions="Is it a bug report?"),
    },
)
r.choices["team"].choice         # "billing"
r.choices["team"].probabilities  # {"billing": 0.88, ...}

Same imports, same calls, same response shapes. Arguments that only mean something remotely are accepted and ignored.

ZERO-SHOT · AG NEWS, BANKING77, EMOTION · 300 EXAMPLES

Up to 0.877 zero-shot, 0.897 with labels.

Click a column to sort. Highlighted: the best value in each column.

TWO LINES · ON PYPI

Up in one coffee.

C:\> pip install jul
C:\> jul setup
  backend ....... mlx (Apple Silicon)
  model ......... wemm-4b-4bit, 2.6 GB, downloaded once
  check ......... 1 decision OK

jul setup picks the backend (MLX on Apple Silicon, PyTorch elsewhere), installs it if missing, downloads the model once and runs one real decision.

No terminal handy? Let an agent do it →

OR FROM A PROMPT

Let a big LLM install the small one.

Paste this into Claude Code, Codex, or any agent that has a shell.

Install jul (https://pypi.org/project/jul/), a local library that answers typed questions
with a 4B model, and check that it works on this machine.

1. Create a virtualenv with Python >= 3.10.
2. In that venv: pip install jul   then   jul setup
   jul setup picks the backend (MLX on Apple Silicon, else PyTorch), installs it, downloads
   WeMM-Embedding-4B (2.6 GB, 4-bit) and runs one test decision.
3. Then run:
   jul ask choice "Which team should handle this ticket?" -o billing:"payments, invoices" -o technical:"bugs, errors" --state "I was charged twice"
   and show me the JSON.

Do not use sudo and do not install anything outside the venv. If a step fails, show me the error and
what you suggest before trying something else. At the end, tell me in two lines: the backend, the
latency per decision, and the answer to the ticket.

BY JÉRÔME GUYON · JUL-SHOWCASES

Seen in the wild.

Seven small demos, each one an idea from jevable.com, run locally through from jul import TypeSafeClient.

JuL, Just use Less. A headless decision runtime for local models. Apache-2.0.

CREDITS
  bdauzats ........ author
  jlestel ......... contributor
  JeromeGuyon ..... contributor

github.com/usejul/jul · pip install jul · weights on Hugging Face · Not affiliated with TypeSafe AI or Jev.

C:\JUL> exit_

Game of Life. Also runs locally.