Bespoke Automata

Bespoke Automata, An Introduction

About Bespoke Automata

Create and deploy sophisticated Agent AI's to a single API with Bespoke Automata. With Bespoke Automata, you can combine large language models running locally or remotely with instruments for database IO, dictionaries, arrays, logic, APIs and more into powerful Brains capable of pursuing goals set by their designers.

With Bespoke Automata, you can design and test brains via a Directed Graph GUI (powered by litegraph), and deploy them behind a single user friendly API, each brain a different endpoint.

Demo Video

⚠️ READ CAREFULLY, INSTALLATION IS NOT STREAMLINED ⚠️

This is a development release and while the software is maturing, I would recommend you approach the installation process as you would any software under development. If you encounter any problems or would like to propose an improvement, please raise an issue. Join us on Discord, we would love to hear about what you're building with Bespoke Automata.

How to install/run BA and it's stack:

Requirements

NPM
- Electron-forge
- Yarn
Python
- flask
- sentence_transformers

Optional GPU support

Cuda/Blas/etc setup
Cuda Toolkit
NVCC
For metal support in MAC OSX, llama-cpp-python should work out of the box

GUI

The bespoke automata GUI is a node graph UI using a modified litegraph, so it should be familiar to ComfyUI users

Installation

Clone the repository and open the directory

$ git clone https://github.com/C0deMunk33/bespoke_automata
$ cd bespoke_automata

Use yarn to install and run

$ yarn install
$ yarn run start

work through installing the modules until it works

BA API:

The BA API uses llama-cpp-python for text inference and vision

place text models in the folder ../models/text
place vision models in the folder ../models/vision
NOTE THIS IS AT THE SAME LEVEL AS THIS REPO, GGUF work best IMO, get then from Hugging Face.* NOTE: if you are running non-cuda (Apple silicon, AMD, Intel,CPU etc) you will need to follow the instructions on https://github.com/abetlen/llama-cpp-python to compile for your hardware NOTE: llama-cpp-python binaries on Apple M* hardware have been tested to be grand.
Metal OSX: CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
CUDA LINUX: CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python
CUDA WINDOWS: $env:CMAKE_ARGS = "-DLLAMA_CUBLAS=on" then pip install llama-cpp-python
CPU: pip install llama-cpp-python
cd bespoke_automata/APIs/
python omni_api.py
work through pip installs until it works
the server will be your_ip:5000
endpoint for text acts like GPT (and defaults to GPT, but that may be broken)

NOTE: On OSX, port 5000 collides with Airplay Receiver. You can either turn it off in Settings > General > Airdrop & Handoff or switch the port in the config.

Back end:

Once completed, a brain can be deployed as API endpoints.

save brain to bespoke_automata/bespoke_manager/graphs
cd bespoke_automata/bespoke_manager/
node server.js
work through any NPM install issues
Brains will be your_ip:9999
your_ip:9999/brains will list brains
your_ip:9999/brains/[brain filename sans extension] is brain endpoint
your_ip:9999/brains/[brain filename sans extension]/schema shows IO params for that brain