Geist

FOSS Project for natural language AI Assistants - using GPT, LlaMA, Mistral etc. architectures. Geist is a framework for multiple natural language AI Assistants to interact, train, and do work.

Roadmap

Create core OODA architecture for AI Assistants using various LLM transformer architectures.
Create core API surface area for inference for architectures.
Persist conversations, memory, and presets for various agents and feed these back into agent startup / initialization.
Create external world adapters for agents to interact with the real world, such as VoIP, Slack, Notion, etc.

Roadmap Agents

Software engineer spirit. (Reviews PRs, creates design docs, creates code from natural language description of features.)
Virtual assistant spirit. (Summarize and respond to emails, keep track of birthdays,
Business manager spirit. (Generate business ideas, affiliate marketing, etc. and provide detailed blueprints for execution.)
Research assistant spirit. (Summarize arxiv posts on machine learning / deep learning, provide briefs on latest changes in the field.)

Core Architecture

The Geist Architecture consists of three main components: a world model, a task creation model, and an execution-based model. The diagram below illustrates the architecture and relationships between these components.

flowchart LR
    A[World Model] -->|Updates| B[Task Creation Model]
    B -->|Generates Tasks| C[Execution-based LLM Model]
    C -->|Execution Results| A
    D[Plugin System] -->|Provides Capabilities| C
    E[Long-term Memory] -->|Base Knowledge| A
    F[Short-term Memory] -->|Context-specific Info| C

Versioning and Setup

Install Postgresql

Version 16.2 on Mac or Windows

Install Miniconda

Create a python 3.10 environment solve and install with windows_x64_environment.yml or mac_arm_environment.yml

Setting up your environment

Make sure that your .env file is initialized - the following values are included but you may not need to set all of these depending on agent utilization and DEV/PROD settings:
- OPENAI_TOKEN = TOKEN
- POSTGRES_PWD = PASSWORD
- POSTGRES_DB = geist
- TWILIO_SID = ACCOUNT_SIDE
- TWILIO_SOURCE = SOURCE_NUMBER
- TWILIO_TOKEN = API_TOKEN
- ENHANCED_LOGGING = FALSE
- DB_HOST = localhost
- HUGGING_FACE_HUB_TOKEN = TOKEN_VALUE
Copy any model weights into app/models/weights/MODEL_NAME.
- Currently supported Models: llama_3_1
- This can be run from python scripts/download_models.py
- If the download script doesn't work, direct download links can be found at : https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct/tree/main

client/geist/.env settings: - REACT_APP_API_BASE_URL = http://localhost:3000

Starting the solution

Start the postgresql server PATH/pg_ctl -D DATA_PATH -l LOG_PATH start
Run python bootstrap.py

Starting the solution with docker compose

Run docker compose up

Supported Environments

Mac OS - ARM (m series) with linux container.

DavidDworetzky/Geist