Environment Setup Walkthrough

This document explains how to bootstrap the BODMAS calculation engine from scratch.

1. Environment Variables (.env)

Copy the provided .env.example to .env:

cp .env.example .env

The system relies on LLMs. The fallback chain is:

1. OLLAMA_BASE_URL: The system attempts to ping this first. Default is http://localhost:11434.
2. OPENAI_API_KEY: If Ollama is down, it uses GPT-4o.
3. ANTHROPIC_API_KEY: If OpenAI is missing, it uses Claude 3.5 Sonnet.

If you wish to use Langfuse tracking, ensure these are populated:

• LANGFUSE_PUBLIC_KEY
• LANGFUSE_SECRET_KEY
• LANGFUSE_HOST=http://localhost:3000

2. Docker & Database Migrations

The system requires PostgreSQL to run Langfuse. Ollama is explicitly excluded from Docker and must be run on the host machine.

docker-compose up -d

Docker Compose will automatically spin up PostgreSQL, wait for it to be healthy, and start Langfuse. Langfuse will automatically run its schema migrations on boot.

3. Telemetry Endpoint Binding

Once Langfuse boots:

1. Navigate to http://localhost:3000.
2. Create an account and a new project.
3. Generate new API keys in the project settings.
4. Paste the LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY into your .env file.

The src/config.py uses get_langfuse_handler() to automatically bind these keys into a CallbackHandler which is injected into the LangGraph compile and invoke sequences.