This project provides a no-code environment for building and executing LLM-powered scripts within Obsidian, targeting users who want to automate tasks and create custom chatbots without writing traditional code. It enables users to visually define LLM logic, data flow, and actions using Obsidian Canvas.

How It Works

Cannoli scripts are constructed as Directed Acyclic Graphs (DAGs) within Obsidian Canvas, using nodes (cards) and edges (arrows) to represent variables, logic, and LLM interactions. This visual approach allows for intuitive script creation, abstracting away complex programming. The system interprets the Canvas structure according to a defined schema to execute LLM calls, process responses, and perform actions like HTTP requests.

Quick Start & Requirements

• Install via Obsidian's community plugins.
• Requires Obsidian.
• Supports OpenAI, Groq, Anthropic, Gemini, and local LLMs via Ollama.
• Ollama requires setting the OLLAMA_ORIGINS environment variable to *.
• See walkthrough folder in plugin settings for samples.

Highlighted Details

• Visual, no-code interface for LLM script creation using Obsidian Canvas.
• Supports custom chatbots with streaming and customizable formatting.
• Can trigger scripts via audio recordings (transcribed by Whisper).
• Allows defining custom HTTP requests within scripts.

Maintenance & Community

• Development instructions are available in DEVELOPMENT.md.

Licensing & Compatibility

• License not specified in the README.

Limitations & Caveats

• Full documentation website is forthcoming.
• Changing models within a single script requires reloading, potentially increasing execution time.