Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/beefy/workoutassistant

Email based AI assistant
https://github.com/beefy/workoutassistant

ai-agent ai-assistant algotrading cryptocurrency email llm moltbook raspberry-pi solana webcrawling

Last synced: 3 months ago
JSON representation

Email based AI assistant

Awesome Lists containing this project

README 

          
# workoutassistant



An LLM that will answer emails. Capable of running on a raspberry pi with 6gb of RAM.



 - Email uses gmail SMTP/IMAP

 - LLM uses phi-3-mini and runs locally

 - Image captioning uses blip and runs locally

 - Image generation runs in the cloud using a hugging face interface (free)

 - Image to image (image editing) runs in the cloud using a hugging face interface ($0.02 to $0.03 per call) 

 - Approve List for approved email senders is stored in SQLite

 - Web search uses duck duck go



Tasks that run on a regular schedule:



 - Heartbeat

 - Check and process emails

 - Create and send a daily newsletter based on the front page news of apnews.com

 - Browse Moltbook, making posts, making comments, or upvoting/downvoting posts

 - Trade cryptocurrency on the Solana blockchain based on 5 indicators from the last 72 hours of price data



### Environment Variables



See `.variables.example` for expected environment variables.



### Docker Commands



Install

```

# Install Docker

curl -fsSL https://get.docker.com -o get-docker.sh

sudo sh get-docker.sh

sudo usermod -aG docker $USER



# Install Docker Compose (if not included)

sudo apt install docker-compose-plugin

```



Ensure it starts on boot

```

sudo systemctl enable docker

```



Commands

```bash

# View logs

docker compose logs -f workoutassistant



# Check status

docker compose ps



# Update to latest version

docker compose pull && docker compose up -d



# Stop containers

docker compose down



# Restart

docker compose restart



# SSH into the docker container

docker exec -it workoutassistant /bin/bash

cat /app/logs/output.log



# Fix permissions issue

docker exec -it --user root workoutassistant chmod -R 777 /app/src/generated_images



# Build locally

nohup docker build -t workoutassistant-local . > docker-build.log 2>&1 &



# Run with local image override

docker-compose up -d workoutassistant

```



To fix permission denied error

```

# First, check if docker group exists

grep docker /etc/group



# If docker group doesn't exist, create it:

sudo groupadd docker



# Add your user to docker group

sudo usermod -aG docker $USER



# Verify the user was added (should show your username)

getent group docker



# MUST logout/login or restart SSH session for permanent fix

reboot



# Temporary fix for current session only:

newgrp docker



# After logout/login, verify it worked:

groups | grep docker

docker ps

```



More setup

```

mkdir -p ./logs ./data ./generated_images && chmod 755 ./logs ./data ./generated_images

```

or

```

sudo chown -R $USER:$USER ./logs ./data ./generated_images 2>/dev/null || echo "Some directories don't exist yet - that's ok"

```



To make docker start on startup

```

sudo systemctl enable docker

sudo systemctl enable docker.socket

```



To make docker not start on startup

```

sudo systemctl disable docker

sudo systemctl disable docker.socket

```



To start docker

```

sudo systemctl start docker

sudo systemctl start docker.socket

```



To check if docker is running

```

sudo systemctl status docker

```



To check docker events

```

docker events

```



### System Monitoring

```bash

# Monitor temperature

vcgencmd measure_temp



# Check throttling status

vcgencmd get_throttled



# Monitor power supply

vcgencmd measure_volts

```



### 🔄 Automatic Updates



The deployment includes a script which:

- Polls DockerHub every 5 minutes for new images

- Automatically pulls and deploys updates



To run it

```

# Edit crontab

crontab -e



# Add this line (checks for updates every 5 minutes):

*/5 * * * * /home/bob/Code/workoutassistant/src/scripts/update_cron.sh >> /home/bob/Code/workoutassistant/update.log 2>&1

```



### Setting Up CI/CD



To set up automatic Docker image building and publishing:



**Set up DockerHub account** and create a repository named `workoutassistant`



**Configure GitHub Secrets** in your repository settings:

   - Go to Settings → Secrets and variables → Actions

   - Add these repository secrets:

     - `DOCKERHUB_USERNAME`: Your DockerHub username

     - `DOCKERHUB_TOKEN`: Your DockerHub access token ([create one here](https://hub.docker.com/settings/security))



### 🏥 Health Monitoring



The container includes health checks that monitor if the main process is running. Container will restart automatically if health checks fail.



### 🔧 Troubleshooting



**Container won't start:**

- Check if `~/.variables` file exists and has the required variables

- Verify Docker is running: `sudo systemctl status docker`

- Check logs: `docker compose logs workoutassistant`



### Prerequisites



Install system dependencies:

```bash

# Headless web search

sudo apt install -y chromium chromium-driver



# Python and pip (usually pre-installed)

sudo apt install -y python3 python3-pip



# SQLite (usually pre-installed)

sudo apt install -y sqlite3

```



### Install Python Dependencies



```bash

# Clone the repository

git clone https://github.com/YOUR_USERNAME/workoutassistant.git

cd workoutassistant



# Install Python packages

pip install -r requirements.txt

```



### Setup Local LLM



```

# Create models directory

mkdir -p ~/models

cd ~/models



# Download Phi-3-mini 4k version (~2.4GB)

wget https://huggingface.co/microsoft/Phi-3-mini-4k-instruct-gguf/resolve/main/Phi-3-mini-4k-instruct-q4.gguf



# Download Phi-3-mini 128k version for longer context windows

wget https://huggingface.co/mradermacher/Phi-3-mini-128k-instruct-ft-i1-GGUF/resolve/main/Phi-3-mini-128k-instruct-ft.i1-Q4_K_M.gguf?download=true

```



You may need to create a hugging face user and browse the web UI for models: https://huggingface.co/docs/huggingface_hub/v1.4.0/quick-start#authentication



### Local Image Captioning Setup



The image captioning module automatically downloads the BLIP-base model (~1.2GB) on first use. For manual installation:



```bash

# Install required packages

pip install torch transformers pillow



# Test the image captioning (will auto-download model)

python src/image_captioning.py

```



**Model Details:**

- **Model**: Salesforce BLIP-base (blip-image-captioning-base)

- **Size**: ~1.2GB download

- **RAM Usage**: ~2-3GB during inference

- **Performance**: 10-30 seconds per image on Raspberry Pi, 2-5 seconds on desktop CPU

- **Capabilities**: Image captioning and visual question answering



**Raspberry Pi Optimization:**

- Uses CPU-only inference (no GPU required)

- Optimized for low memory usage with `low_cpu_mem_usage=True`

- Automatically detected Pi hardware for performance messaging



**Usage Example:**

```python

from src.image_captioning import caption_image_local, ask_about_image_local



# Basic image captioning

caption = caption_image_local('path/to/image.jpg')

print(f"Caption: {caption}")



# Visual question answering

answer = ask_about_image_local('path/to/image.jpg', 'What colors are in this image?')

print(f"Answer: {answer}")

```



### SQLite Database Setup



SQLite comes pre-installed on most systems. To verify:



```bash

# Check if SQLite is installed

sqlite3 --version



# If not installed:

# Ubuntu/Debian:

sudo apt-get install sqlite3



# macOS (if not present):

brew install sqlite



# CentOS/RHEL:

sudo yum install sqlite

```



Create the initial tables

```

PYTHONPATH=src python src/scripts/create_tables.py

```



### Email setup



1. Create a gmail account for this project

2. Enable 2FA

3. Generate an App Password: https://myaccount.google.com/apppasswords



Store the necessary environment variables

```

export GMAIL_ADDRESS=

export GMAIL_APP_PASSWORD=

export TEST_EMAIL=

export ADMIN_EMAIL=

export APPROVED_PHRASE=

```



### Moltbook setup

```

curl -X POST https://www.moltbook.com/api/v1/agents/register \

  -H "Content-Type: application/json" \

  -d '{"name": "YourAgentName", "description": "What you do"}'

```



```

export MOLTBOOK_API_KEY=

```



### API integration for tracking status

```

export TRACKING_API_USERNAME=

export TRACKING_API_PASSWORD=

```



Create account

```

python src/scripts/register_tracking_api.py

```



### Crypto Setup



Create crypto wallets

```

python src/scripts/create_crypto_wallets.py

```



Acquire and set these environment variables

```

export SOLANA_ADDRESS=

export SOLANA_PRIVATE_KEY=

export JUPITER_API_KEY=

export BIRDEYE_API_KEY=

```



Check crypto balance

```

python src/scripts/get_crypto_balance.py

```



### Lichess Setup

Create a new account on Lichess. Convert it to a bot account with this curl command. Note: the account cannot have any games played.

```

curl -d '' https://lichess.org/api/bot/account/upgrade -H "Authorization: Bearer "

{"ok":true}

```



### Youtube / Discord Setup

```

brew install ffmpeg

```

See DISCORD_BOT_README.md



### Run the script



Start it in the background

```

# Use -u flag for unbuffered output and full path for log file

nohup python -u src/main.py > $(pwd)/output.log 2>&1 &



# Alternative: Set PYTHONUNBUFFERED environment variable

nohup env PYTHONUNBUFFERED=1 python src/main.py > $(pwd)/output.log 2>&1 &

```



Find the process to kill it when done

```

# Find by script name

ps aux | grep src/main.py

```



Monitor the output in real-time

```

# Follow the log file continuously (Ctrl+C to stop)

tail -f output.log



# Alternative: Show last 50 lines then follow

tail -n 50 -f output.log



# Less command with follow mode (F key to follow, q to quit)

less +F output.log

```



Kill the process

```

# Kill by PID

kill 12345



# Force kill

kill -9 12345

```



### Run the gh actions runner



```

nohup ./run.sh > output.log 2>&1 &

```