Pydantic AI RAG Agent in Docker Sandbox

An enterprise-grade Pydantic AI Retrieval-Augmented Generation (RAG) agent running inside a Docker sandbox environment, integrated with:

• 🎯 Qdrant Vector Database: Vector storage and similarity search at http://localhost:6333
• 📦 RustFS / MinIO S3 Object Server: High-performance S3-compatible document storage holding gxp-docs at http://localhost:9100
• 🔍 Langfuse: LLM observability and tracing platform at http://localhost:3000
• 🤖 Ollama: Local embedding model server running qwen3-embedding:8b at http://localhost:11434 integrated via Pydantic AI's native Embedder (pydantic_ai.embeddings.Embedder)
• ⚡ Pydantic AI Agent: DeepSeek model (deepseek-chat / deepseek-v4-flash) powered by Pydantic AI v2.26.0+.

---

🚀 Execution Modes

Mode 1: Docker Sandboxes CLI (sbx create & sbx run)

For interactive management inside the Docker Sandboxes (sbx) Dashboard & TUI:

1. Create the sandbox:

   sbx create --name rag-agent-sbx -p 8000:8000 shell .

2. Run or attach to the interactive agent sandbox:

   sbx run --name rag-agent-sbx

3. Launch the RAG agent API server inside the sandbox:

   sbx exec rag-agent-sbx python3 -m uvicorn app.main:app --host 0.0.0.0 --port 8000

4. View active sandboxes in the sbx dashboard:

   sbx ls
   # or interactive dashboard:
   sbx tui

---

Mode 2: Docker Compose (docker compose up -d)

If Qdrant (:6333), RustFS (:9100), Langfuse (:3000), and Ollama (:11434) are running on your host machine:

1. Configure .env:

   S3_ENDPOINT_URL=http://localhost:9100
   S3_ACCESS_KEY_ID=RFS8K9X2M4P7Q1W3V6N5
   S3_SECRET_ACCESS_KEY=a7F9mK2vX8pQ5wL1zN4bR0cT6yH3jS8d
   S3_BUCKET_NAME=gxp-docs
   
   QDRANT_HOST=localhost
   QDRANT_PORT=6333
   QDRANT_COLLECTION_NAME=gxp_documents
   
   OLLAMA_BASE_URL=http://localhost:11434
   EMBEDDING_MODEL=qwen3-embedding:8b
   
   OPENAI_BASE_URL=https://api.deepseek.com/v1
   OPENAI_API_KEY=your-deepseek-api-key
   LLM_MODEL=deepseek-chat
   
   LANGFUSE_PUBLIC_KEY=pk-lf-8d6eb8e5-c24b-4914-a8c7-b9e21ff9290c
   LANGFUSE_SECRET_KEY=sk-lf-b29640ce-6ec0-4298-8a10-5262a727f2bd
   LANGFUSE_HOST=http://localhost:3000

2. Start the containerized agent sandbox:

   docker compose up -d --build rag-agent

---

📄 Document Seeding & Vector Ingestion

To upload sample GxP Standard Operating Procedures (SOPs) into RustFS S3 (gxp-docs bucket) and index vector embeddings into Qdrant using Ollama qwen3-embedding:8b:

python scripts/seed_gxp_docs.py

---

📡 API Endpoints

The containerized RAG Agent API exposes endpoints on http://localhost:8000:

Method	Endpoint	Description
GET	/	System status & service configuration
GET	/health	Connectivity check for S3 (RustFS) & Qdrant DB
GET	/documents	List objects in gxp-docs bucket
POST	/ingest	Ingest a single document by S3 key
POST	/ingest/all	Sync all documents in gxp-docs into Qdrant
POST	/query	Execute Pydantic AI agent RAG query

Example Query Request

curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "What are the gowning requirements and wet contact time for Class B cleanroom sanitization?"
  }'

Example JSON Response

{
  "answer": "Based on SOP-QA-001_Cleanroom_Sanitization.txt, personnel entering Class B areas must wear sterile non-linting coveralls, triple gloves, sterile face mask, safety goggles, and sterile dedicated footwear. Sanitization requires a minimum of 10 minutes wet contact time for all surfaces.",
  "sources": [
    "SOP-QA-001_Cleanroom_Sanitization.txt"
  ],
  "confidence_level": "High"
}

---

🔍 Observability with Langfuse

• Navigate to http://localhost:3000.
• Traces show full execution telemetry for:
  • Document ingestion & text chunking
  • Local embedding calls via Pydantic AI's native Embedder (qwen3-embedding:8b)
  • Qdrant similarity searches (gxp_documents collection)
  • S3 raw document fetches from RustFS
  • DeepSeek LLM reasoning & Pydantic AI tool invocations.

---

🏗 System Architecture

                 +--------------------------+
                 |  RustFS (S3 Object Store)|
                 |    http://localhost:9100 |
                 |     Bucket: gxp-docs     |
                 +------------+-------------+
                              |
                     1. Upload / Fetch
                              v
+-----------------------------+-----------------------------+
|        Pydantic AI Agent (Docker Sandbox / sbx)           |
|                   (app.main / app.agent)                  |
+--------------+------------------------------+-------------+
               |                              |
 2. Generate   |               3. Vector      | 4. Traces
 Embeddings    v               Search         v & Metrics
+--------------+----------+   +---------------+----------+   +------------------------+
|      Ollama Service     |   |    Qdrant Vector DB    |   |    Langfuse Platform   |
|   http://localhost:11434|   | http://localhost:6333|   |  http://localhost:3000 |
| model: qwen3-embedding:8b|  | Collection: gxp_docs  |   |    OpenTelemetry/SDK   |
+-------------------------+   +--------------------------+   +------------------------+