Shadow Tribunal

A local-first, privacy-friendly LLM app that generates a multi-voice psychological analysis (“Analyst”, “Critic”, “Protector”) from user input, running on Ollama and presented via a Streamlit UI.

What it does

• Runs entirely locally (Ollama on your machine)
• Produces structured output:
  • Psychological profile (archetype, traits, motivations, blind spots)
  • “The Tribunal Speaks” (multi-perspective synthesis)
• Designed for fast iteration on prompts + logic with a clean UI

Tech stack

• Python
• Streamlit
• Ollama (local inference)
• httpx (HTTP client)

Quick start (local)

1) Install Ollama + pull a model

Install Ollama and pull a model you want to use (examples):

• llama3
• mistral
• qwen2.5

Example:

ollama pull llama3

Start Ollama (if not already running):

ollama serve

2) Install Python deps

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

3) Run the Streamlit app

From the repo root:

streamlit run streamlit_app/app.py

Configuration

This app expects an Ollama generate endpoint. Defaults to:

• http://127.0.0.1:11434/api/generate

You can override via environment variables:

1. Copy:

cp .env.example .env

2. Edit .env if needed.

Repo layout

• streamlit_app/ — Streamlit UI
• core/ — prompts + logic + LLM client
• run_local.sh — convenience script

Notes for deployment

This repo is meant to be run locally. If you deploy, keep .env out of git and use environment variables.

Link: https://tribunal.bitshadow.dev/