lime-ai 0.1.9

Lime AI — An agentic markdown runner.

Lime

A code agent with programmable markdown files

Lime wants you have fun programming Agents/LLMs and bring back more control into your hands.

Features

• Get all the benefits of margarita files now with agentic capabilities.
• Bring your favourite coding agent (Copilot for now) (roadmap for Claude code, codex and more soon)
• React like composability through includes [[ my-component is_admin=true ]]
• Store variables in state, and have the LLMs update/fetch state as needed.
• Don't waste tokens on having LLMs run functions. Run it locally and then pass the results to the LLM.
• Forget context explosion issues. Surgically control what context is sent to the LLM.
• Add only the tools you need for a query. Nuke them, then add others and repeat. This keeps context size small and relevant.

Example

// file:example.mgx
---
description: Add metadata
team: Can put anything in metadata
---

// Import python functions for use with @effect func
from math import add, subtract, multiply, load_files

// Supports all markdown. Places this into the agent's context.
<<
You are an expert mathematician.
Your task is to solve addition problems accurately and efficiently.

When given a problem, you should:
1. Read the problem carefully.
2. Identify the two numbers to be added.
3. Calculate the sum of the two numbers.
4. Provide the final answer clearly.
>>

// Include other margarita files into the context
[[ create-a-react-component.mg ]]

// Execute a loop 
for i in items:
    // run Python functions and store results in state.
    @effect func add(12, test.data) => result

// The agent can access/ set state variables too!
<<
Add 12 + test.data and store the result in the variable 'result'.
>>

// Add tools, note: AddToolParam extends BaseModel from pydantic
@effect tools add(params: AddToolParams) => int

// Run the agent using tools and the context you built up.
@effect run

// clear the context and tools after running to avoid context explosion in future runs.
@effect context clear
@effect tools clear

// use the state result variables with a new context.
<<
Validate the following:
- The addition tool correctly adds two numbers.
- The subtraction tool correctly subtracts two numbers.
- The multiplication tool correctly multiplies two numbers.
- The load_files function correctly loads and reads files from the specified directory.
>>

// conditonal logic
if result.failed:
    <<
    The test failed. Please review the implementation of the math tools and the
    load_files function for any errors.
    >>
    @effect run
else:
    // We're done!

Hopefully this gives you a taste of the possibilities with Lime!

Examples

The examples/ directory contains ready-to-run .mgx workflows you can use as a starting point:

Example	Path	Description
Hello World	examples/helloworld/helloworld.mgx	Minimal workflow to get started
Plan & Implement	examples/plan-implement/pi.mgx	Plan then implement a feature in a loop
Ralph	examples/ralph/custom-ralph.mgx	Multi-file workflow with custom tools and task breakdown

Installation

Run the following command for your platform to install Lime:

Linux:

curl -fsSL https://raw.githubusercontent.com/Banyango/lime/main/install-linux.sh | bash -s -- --option

MacOS:

curl -fsSL https://raw.githubusercontent.com/Banyango/lime/main/install-macos.sh | bash -s -- --option

Windows (PowerShell):

iwr -useb https://raw.githubusercontent.com/Banyango/lime/main/install-windows.ps1 | iex

Logo Designed by Freepik

Note: The CLI execute command supports a --headless flag which disables the textual UI and suppresses stdout/stderr output; in headless mode the process will exit with non-zero codes for missing files or when interactive input/permissions are required.

Class documentation standard

See the project class documentation standard for required docstring sections and concrete examples (data class, service class, abstract/base class): docs/class-documentation.md