magnetic
by Toolkit
magnetic
is a way to run shareable, and reusable tool-powered LLM / agent workflows. We think of it as an agent/workflow runner & package manager for AI tools and workflows, that can easily be run, expanded and combined to powerful effect.
Here's a few examples to give you the vibe:
Installation
-
Clone the repository:
git clone https://github.com/toolkit-ai/magnet-cli-v2.git
-
Navigate to the project directory:
cd magnet-cli-v2
-
Install dependencies:
bun install
Usage
magnet-context
offers several commands:
run
: Executes a workflowcreate-recipe
: Creates a new recipe (composition of workflows)list-recipes
: Lists available recipes (compositions of workflows)list-models
: Lists available AI modelslist-workflows
: Lists available workflowscreate-workflow
: Creates a new workflow filepipe
: Runs a series of workflows, piping the output of each to the next similar to the UNIX pipe commandagent
: (Coming soon, an agent that composes and runs workflows for you)
Example usage:
bun run start create-recipe
bun run start run --recipe recipes/project-overview.yaml
bun run start run --workflow generate-knowledge --context.directory . --context.instructions "Please write a project overview" --metadata.title "Project Overview" --metadata.description "Writes an overview of the current project including general pattern, stack, and structure"
bun run start list-workflows
bun run start create-workflow
bun run start pipe read-file count-words write-file
Development
To run tests:
bun run test
Configuration
The tool supports various configuration options, including:
- Debug mode
- Custom directories for output and logs
- AI model selection
These can be set via command-line options or environment variables.
Contributing
Contributions are welcome! Please feel free to submit a Pull Request.