dbx-unifiedchat

Development Guide

This guide covers the three supported development workflows for the Multi-Agent Genie System. Choose the workflow that best fits your current task.

Workflow 1: Local Development (Fastest Iteration)

Best for: Writing unit tests, modifying single agent logic, fast iteration without deploying.

Setup

# Setup virtual environment
python3 -m venv .venv
source .venv/bin/activate
pip install -e .

# Setup pre-commit hooks for code quality
pip install pre-commit
pre-commit install

Configuration

Copy the example environment file and fill in your values:

cp .env.example .env
# Edit .env with your credentials and resource IDs

Testing & Running

# Test individual modules
pytest tests/unit/test_planning_agent.py -v

# Test full system locally
python src/multi_agent/main.py

Workflow 2: Databricks Notebook Dev (Real Services)

Best for: Integration testing, debugging with real Databricks services (Genie, Vector Search, Lakebase) before deployment.

1. Sync code to Databricks

We recommend using Databricks Repos to keep your Databricks workspace synced with your git repository:

databricks repos update <repo-id>

2. Configuration

The notebook tests use dev_config.yaml. Make sure your variables in this file point to your development/sandbox resources.

3. Run Notebooks

Open the Databricks UI, navigate to the synced repo, and run:

1. Notebooks/test_agent_databricks.py
2. Iterate on src/multi_agent/ code (changes will auto-reload if using %autoreload 2)
3. Debug directly against real services without packaging the model.

Workflow 3: Production Deployment (CI/CD)

Best for: Final deployment to Model Serving endpoints.

Method A: Via GitHub Actions (Recommended)

Our CI/CD pipeline automatically runs tests and deploys based on branch pushes.

1. Deploy to Dev:

   git checkout develop
   git commit -m "feat: add new agent"
   git push  # Auto-runs tests, validates bundle, and deploys to dev workspace
   

2. Deploy to Prod:

   # Create a Pull Request to main branch
   # Once merged:
   # Auto-runs tests, validates bundle, and deploys to prod workspace
   

Method B: Via CLI (Manual)

If you need to deploy manually using the Databricks Asset Bundle (DAB):

# Validate the bundle
databricks bundle validate

# Test agent in Databricks (Dev)
databricks bundle run agent_integration_test -t dev

# Deploy to Dev
databricks bundle run agent_deploy -t dev

# Deploy to Prod
databricks bundle run agent_deploy -t prod

This site is open source. Improve this page.