Pydantic AI Extended

This sample shows a multi-service AI app built with Pydantic AI, FastAPI, PostgreSQL (with pgvector), and Redis, deployed with Defang from a single Docker Compose file.

A background worker generates sample support tickets and system alerts with an LLM, pushes them to a queue, then they get classified and stored with vector embeddings in PostgreSQL for semantic search. A Pydantic AI copilot agent uses tools to inspect the current state before answering questions.

Prerequisites

1. Download Defang CLI
2. (Optional) If you are using Defang BYOC authenticate with your cloud provider account
3. (Optional for local development) Docker CLI

Development

To run the application locally for development, use the development compose file:

docker compose -f compose.dev.yaml up --build

This will:

• Start the FastAPI app on http://localhost:8000
• Start PostgreSQL on port 5432
• Start Redis on port 6379
• Start a background worker for item classification
• Start Docker model-provider services for chat + embeddings

Local development uses:

• ai/qwen2.5:3B-Q4_K_M for chat/tool-calling
• mxbai-embed-large for embeddings

This relies on Docker Model Runner / model-provider support being available in your local Docker installation. The first run will download both models, so startup can take a few minutes. If docker compose -f compose.dev.yaml up fails with exec: "model": executable file not found in $PATH, your local Docker installation does not have Docker Model Runner enabled yet. To keep iteration practical on CPU-only setups, compose.dev.yaml enables LOCAL_FAST_DATA=true, which uses deterministic sample generation and classification locally while still exercising the real chat and embedding services.

In deployed environments, the app uses dedicated chat and embedding model services defined in compose.yaml. Defang injects OpenAI-compatible CHAT_URL / CHAT_MODEL and EMBEDDING_URL / EMBEDDING_MODEL environment variables automatically, so the application code stays platform-independent.

Configuration

For this sample, you will need to provide the following configuration. Note that if you are using the 1-click deploy option, you can set these values as secrets in your GitHub repository and the action will automatically deploy them for you.

POSTGRES_PASSWORD

The password for your PostgreSQL database. You need to set this before deploying for the first time.

You can easily set this to a random string using defang config set POSTGRES_PASSWORD --random

Usage

1. Open the app.
2. Click Generate sample items.
3. Watch the worker create 10 tickets and 10 alerts, then fan out per-item classify/embed jobs (progress updates in real time via polling).
4. Ask questions like:
   • What should I look at first?
   • Summarize the current tickets and alerts.
   • Which items seem related?
   • Find alerts similar to the payment outage.

Deployment

Note

Download Defang CLI

Defang Playground

Deploy your application to the Defang Playground by opening up your terminal and typing:

defang compose up

BYOC (Deploy to your own AWS or GCP cloud account)

If you want to deploy to your own cloud account, you can use Defang BYOC.

The default sample uses Defang's managed model provider services:

• chat uses chat-default
• embedding uses embedding-default

If you want to pin different models, edit the provider.options.model values in compose.yaml.

Warning

Extended deployment time: This sample creates a managed PostgreSQL database which may take upwards of 20 minutes to provision on first deployment. Subsequent deployments are much faster (2-5 minutes).

---

Title: Pydantic AI Extended

Short Description: A Defang sample where background jobs classify and embed support tickets and system alerts, and a Pydantic AI copilot answers questions with tools.

Tags: Pydantic AI, FastAPI, PostgreSQL, Redis, AI, Agents

Languages: Python, Docker