openmacro 0.2.6

Multimodal Assistant. Human Interface for computers.

Project is in its early stage of development. Optimisations & safety not guarenteed until first version release.

openmacro is a multimodal personal agent that allows LLMs to run code locally. openmacro aims to act as a personal agent capable of completing and automating simple to complex tasks autonomously via self prompting.

This provides a cli natural-language interface for you to:

• Complete and automate simple to complex tasks.
• Analyse and plot data.
• Browse the web for the latest information.
• Manipulate files including photos, videos, PDFs, etc.

At the moment, openmacro only supports API keys for models powered by SambaNova. Why? Because it’s free, fast, and reliable, which makes it ideal for testing as the project grows! Support for other hosts such as OpenAI and Anthropic is planned to be added in future versions.

This project is heavily inspired by Open Interpreter ❤️

Quick Start

To get started with openmacro, get a free API key by creating an account at https://cloud.sambanova.ai/.

Next, install and start openmacro by running:

pip install openmacro
macro --api_key "YOUR_API_KEY"

Not working? Raise an issue here or try this out instead:

py -m pip install openmacro
py -m openmacro --api_key "YOUR_API_KEY"

Note: You only need to pass --api_key once! Next time simply call macro or py -m openmacro.

Profiles

openmacro supports cli args and customised settings! You can view arg options by running:

macro --help

To add your own personalised settings and save it for the future, run:

macro --profile "path\to\profile.py" --save

Profiles are python files to allow dynamic integration and type safety (with TypedDict)! Better docs for Profiles will be added in upcoming versions.

What your profile.py might look like:

# imports
from openmacro.profile import Profile
from openmacro.extensions import BrowserKwargs, EmailKwargs

# profile setup
profile: Profile = Profile(
    user = { 
        "name": "Amor", 
        "version": "1.0.0"
    },
    assistant = {
        "name": "Basil",
        "personality": "You have a kind, deterministic and professional attitude towards your work and respond in a formal, yet casual manner.",
        "messages": [],
        "breakers": ["the task is done.", 
                     "the conversation is done."]
    },
    safeguards = { 
        "timeout": 16, 
        "auto_run": True, 
        "auto_install": True 
    },
    paths = { 
        "prompts": "core/prompts",
    },
    extensions = {
    # type safe kwargs
        "Browser": BrowserKwargs(headless=False, engine="google"),
        "Email": EmailKwargs(email="[e-mail]", password="password")
    },
    config = {
        "verbose": True,
        "conversational": True,
        "dev": False
    },
    languages = {
    # specify custom paths to languages or add custom languages for openmacro
      "python": ["C:\Windows\py.EXE", "-c"],
      "rust": ["cargo", "script", "-e"] # not supported by default, but can be added!
    },
    tts = {
    # powered by KoljaB/RealtimeSTT
      "enabled": True,
      "engine": "SystemEngine"
    }
)

If you want to build your own app with openmacro, you can simply extend your profile to use it:

...

async def main():
    from openmacro.core import Openmacro

    macro = Openmacro(profile)
    macro.llm.messages = []

    async for chunk in macro.chat("Plot an exponential graph for me!", stream=True):
        print(chunk, end="")

import asyncio
asyncio.run(main)

You can also switch between profiles by running:

macro --switch "amor"

Profiles also support versions for modularity (uses the latest version by default).

macro --switch "amor:1.0.0"

Note: All profiles are isolated. LTM from different profiles and versions are not shared.

To view all available profiles run:

macro --profiles

Extensions

openmacro supports custom RAG extensions for modularity and theoretically infinite capabilities! By default, the browser and email extensions are installed.

Writing Extensions

Write extensions using the template:

from typing import TypedDict
class ExtensionKwargs(TypedDict):
    ...

class Extensionname:
    def __init__(self):
        ...
      
    @staticmethod
    def load_instructions() -> str:
        return "<instructions>"
        

You can find examples here.

Note, classname should not be camelcase, but titlecase instead.

Note, creating a type-safe kwargs typeddict is optional but recommended. If extesions does not contain a kwarg class, use:

from openmacro.utils import Kwargs

Upload your code to pypi for public redistribution using twine and poetry. To add it to openmacro.extensions for profiles and openmacro.apps for the AI to use, run:

omi install <module_name>

or

pip install <module_name>
omi add <module_name> 

You can test your extensions by installing it locally:

omi install .

Todo's

• AI Interpreter
• Web Search Capability
• Async Chunk Streaming
• API Keys Support
• Profiles Support
• Extensions API
• WIP TTS & STT
• WIP Cost Efficient Long Term Memory & Context Manager
• Semantic File Search
• Optional Telemetry
• Desktop, Android & IOS App Interface

Currently Working On

• Cost efficient long term memory and conversational context managers through vector databases. Most likely powered by ChromaDB.

• Hooks API and Live Code Output Streaming

Contact

You can contact me at [e-mail]

Contributions

This is my first major open-source project, so things might go wrong, and there is always room for improvement. You can contribute by raising issues, helping with documentation, adding comments, suggesting features or ideas, etc. Your help is greatly appreciated!

Support

You can support this project by writing custom extensions for openmacro! openmacro aims to be community-powered, as its limitations are based on its capabilities. More extensions mean better chances of completing complex tasks. I will create an official verified list of openmacro extensions sometime in the future!