Laplace-1-Vector

Laplace-1-Vector

Scores the options an agent could take, or predicts what happens next, before the action runs. A demo and MCP server for the 4B Laplace-1-Vector decision model, for agent builders.

Made by JuneoAI on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: Laplace-1-Vector is a 4B decision model from Juneo Lab that judges what an agent is about to do before it does it. This Space is a demo of it, and it also runs as an MCP server. The model scores options and predicts outcomes, and it never carries out an action.

Who it suits: Developers who build AI agents and want a check before an action runs, and teams that mix people and agents and want to compare options first.

Industry | Who uses it | How they use it | Example
AI agents | Agent developers | Score the actions an agent may take before one runs | Rate a sub-agent request that reaches past its authority
Operations | Teams of people and agents | Compare ways to split work between people and agents | Score candidate task splits for one project
Purchasing | Teams that deal with suppliers | Judge how to answer a supplier agent that asks for data | Score the replies your policy allows
Research | Machine learning researchers | Study a causal-interventional decision model | Run it on a state and read the spread over candidates

Anyone who wants a model that performs the action itself can skip it.

Where to use it: It works in these places.

• A web browser, on the Space's page at Hugging Face
• An MCP client, because the Space also runs as an MCP server
• Windows, with Python 3.12
• macOS, with Python 3.12
• Linux, with Python 3.12

Use case | Who | What they get back
Score candidates | Agent developers | A probability for each candidate your application allows
Predict the next observation | Agent designers | The generated outcome, shown before the action happens
Call it as tools from an MCP client | Developers | Two tools, score_candidates and predict_observation

How to set it up: Using the Space online needs only a browser. A copy on your own computer needs the following.

• Python 3.12 and pip
• The huggingface_hub package, which gives you the hf command
• Disk space for the repository, which holds 4 files

1. Install the huggingface_hub package.
2. Download the Space's files into a local folder.
3. Move into that folder.
4. Install Gradio at the version the Space uses, then install the packages in requirements.txt.
5. Start the app.

pip install -U huggingface_hub
hf download JuneoAI/Laplace-1-Vector --repo-type space --local-dir Laplace-1-Vector
cd Laplace-1-Vector
pip install gradio==6.17.3
pip install -r requirements.txt
python app.py

How to use it: Open the Space's page, or run the app on your computer and open the page Gradio shows. Pick one of two modes, scoring candidates or predicting the next observation. The source names the MCP tools but gives no connection settings.

python app.py

Tool | What it does
score_candidates | Takes a state, a question and your allowed candidates, and returns a distribution over exactly those candidates
predict_observation | Takes environment rules, context and an action, and generates the next observation

Example: You describe a state where a sub-agent asks for more access than it was given, add a question about what to do, and list the responses your application allows. It returns a spread of probabilities over exactly those responses, from one forward pass.

For newcomers and experts: Before an AI agent acts, this model can tell you how likely each of your allowed options is. The first small thing to try is one of the examples set inside the demo.

The model is described as causal-interventional. Scoring uses one forward pass over your candidates. The probabilities are relative to the list you supply and are not a calibrated chance of being correct. The next-observation mode generates text from the rules, context and action you give it.

You give a state, a question and candidates
The model runs one forward pass
It returns a probability for each candidate
Your application decides what to do

Good to know: The model scores and predicts, so your own code stays in charge of what runs.

• License: apache-2
• Cost to run: none named
• Data it reaches: the state, rules and context you type or send
• Not for: carrying out actions, and its numbers are not a calibrated chance of being correct
• Hardware: the Space is set to zero-a10g at Hugging Face, and what your own computer needs is not stated

On Hugging Face, 13 November 2026

• Likes: 1
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/JuneoAI/Laplace-1-Vector
• Files: https://huggingface.co/spaces/JuneoAI/Laplace-1-Vector/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.