Email Agent Workflow

An automated email processing system using OpenAI Agents SDK with Ollama (qwen3:8b).

Architecture

New email arrives to <INSERT EMAIL>
         ↓
Agent 1: "Triage Agent"
- Analyzes email importance/urgency
- Decides: respond, ignore, or flag for human
         ↓
Agent 2: "Draft Agent" (if respond)
- Generates reply draft
- Passes to Agent 3
         ↓
Agent 3: "Quality Checker"
- Reviews draft for tone, accuracy
- Flags if needs human review (sends email with title "FLAGGED EMAIL")

Prerequisites

1. Python 3.9+
2. Ollama installed and running
3. Gmail API credentials

Setup

1. Install Ollama and pull the model

# Install Ollama from https://ollama.com
# Then pull qwen3:8b
ollama pull qwen3:8b

2. Install Python dependencies

cd email-agent-workflow
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

3. Set up Gmail API

1. Go to Google Cloud Console
2. Create a new project (or select existing)
3. Enable the Gmail API:
   • Go to "APIs & Services" → "Enable APIs and Services"
   • Search for "Gmail API" and enable it
4. Create OAuth 2.0 credentials:
   • Go to "APIs & Services" → "Credentials"
   • Click "Create Credentials" → "OAuth client ID"
   • Choose "Desktop app"
   • Download the credentials
5. Rename the downloaded file to credentials.json and place it in this directory

4. Configure environment

cp .env.example .env
# Edit .env if needed (defaults should work)

5. First run (Gmail authorization)

python main.py

On first run, a browser window will open for Gmail authorization. Grant the requested permissions.

Usage

# Ensure Ollama is running
ollama serve

# Run the workflow
python main.py

How It Works

Agent 1: Triage Agent

• Analyzes each unread email
• Classifies by category (business, personal, spam, etc.)
• Rates urgency (high, medium, low)
• Decides action:
  • respond: Pass to Draft Agent
  • ignore: Mark as read, skip
  • flag_for_human: Send notification, skip auto-reply

Agent 2: Draft Agent

• Creates a contextually appropriate reply
• Matches tone to original email
• Addresses all questions/points raised

Agent 3: Quality Checker

• Reviews draft for:
  • Appropriate tone
  • Accuracy
  • Completeness
• If approved: sends the reply
• If issues found: saves draft for review

Files

• main.py - Workflow orchestration
• agents_workflow.py - Agent definitions and tools
• gmail_tools.py - Gmail API functions
• config.py - Configuration (Ollama, email settings)

Customization

Change the model

Edit .env:

OLLAMA_MODEL=llama3.2  # or any other Ollama model

Adjust agent behavior

Modify the instructions in agents_workflow.py for each agent.

Troubleshooting

"Ollama connection refused"

• Ensure Ollama is running: ollama serve

"Missing credentials.json"

• Complete Gmail API setup (Step 3 above)

"Model not found"

• Pull the model: ollama pull qwen3:8b