multimind 0.1.0

Local-first reasoning pipeline wrapper for Ollama and LM Studio

🧠 MultiMind AI

A local-first web UI that adds a reasoning pipeline on top of small local models.

MultiMind AI acts as an intelligent reasoning pipeline for your local AI models. It effortlessly auto-discovers endpoints like Ollama and LM Studio (OpenAI-compatible) and lets you orchestrate dedicated models for different logical phases: Planning, Execution, and Critique.

---

✨ Features

• 🧠 Adaptive Reasoning Modes: Toggle between Off, Medium, and Hard modes to dictate the depth of the model's reflection.
• 🔌 Zero-Config Auto-Discovery:
  • Automatically hooks into local Ollama endpoints (http://127.0.0.1:11434).
  • Supports optional discovery for LM Studio (http://127.0.0.1:1234).
• 🎯 Precision Model Mapping: Assign distinct models to handle the different stages of thought (plan, execute, and critique).
• 💬 Immersive UI: Enjoy a streaming timeline interface with collapsible "thought blocks" to keep your UI clean while the AI thinks.
• 📝 Native Markdown & Math Support:
  • Final outputs are beautifully rendered as HTML in the chat view.
  • Inline and block math equations are flawlessly typeset using a bundled local KaTeX build.
• ⚡ Frictionless Setup: Purely in-memory settings. Zero .env setup required for your first run.

🚀 Quick Start

Get up and running in your local environment in seconds:

# 1. Create a virtual environment
python3 -m venv .venv

# 2. Activate the virtual environment
# On macOS / Linux:
source .venv/bin/activate

# On Windows:
.venv\Scripts\activate

# 3. Install the package
pip install -e .

# 4. Launch the application
multimind AI

Next: Open your browser and navigate to http://127.0.0.1:8000 🎉

🔌 Supported Backends

MultiMind AI works seamlessly with standard local APIs:

• Ollama: Connects via /api/chat and /api/tags
• OpenAI-Compatible Servers (e.g., LM Studio): Connects via /v1/chat/completions and /v1/models

If no provider is automatically detected, you can easily point the backend to your local OpenAI-compatible endpoint using the application's settings panel.

💡 How It Works

MultiMind AI splits inference into modular steps, elevating the capabilities of standard models:

1. Plan: Formulates a structured approach to the prompt.
2. Execute: Generates the primary response.
3. Critique (Hard Mode): Evaluates the execution pass as a rough draft and streams refined, critiqued output as the final answer.

📝 Note: Chat history is intentionally in-memory only for the current MVP.