Services

hemlock-lab runs 6 Docker containers on a hemlock-net bridge network, with Ollama running on the host. This page documents each service's configuration and management.

---

ChromaDB

The vector database backing all 5 RAG pipelines.

Property	Value
Image	chromadb/chroma:0.6.3
Port	8000
Volume	chromadb-data
Network	hemlock-net
Healthcheck	curl -f http://localhost:8000/api/v2/heartbeat

Configuration

ChromaDB runs as an official Docker image with persistent storage via a named volume:

# docker/docker-compose.yml (excerpt)
chromadb:
  image: chromadb/chroma:0.6.3
  ports:
    - "8000:8000"
  volumes:
    - chromadb-data:/chroma/chroma
  networks:
    - hemlock-net
  healthcheck:
    test: ["CMD", "curl", "-f", "http://localhost:8000/api/v2/heartbeat"]
    interval: 10s
    timeout: 5s
    retries: 5

Management

# Check status
docker compose ps chromadb

# View logs
docker compose logs -f chromadb

# Health check
curl http://localhost:8000/api/v2/heartbeat

Collections

After seeding, ChromaDB contains:

Collection	Documents	Purpose
noise-corpus	11	Legitimate business documents

Test runs create additional temporary collections for retrieval and injection testing.

---

Ollama

LLM inference server running on the host (not in Docker), providing embeddings and text generation.

Property	Value
Port	11434
Bind address	localhost (default)
LLM model	smollm2:135m
Embedding model	nomic-embed-text

Containers reach Ollama via the extra_hosts mapping:

extra_hosts:
  - "host.docker.internal:host-gateway"

Each pipeline's OLLAMA_HOST is set to http://host.docker.internal:11434.

Models

# Pull required models
ollama pull smollm2:135m       # LLM inference
ollama pull nomic-embed-text   # Embedding model
ollama pull nomic-embed-text   # Embeddings

# List available models
curl http://localhost:11434/api/tags | jq '.models[].name'

# Test embedding generation
curl http://localhost:11434/api/embed -d '{"model":"nomic-embed-text","input":"test"}'

---

Pipeline Containers

All 5 RAG pipeline containers follow the same pattern:

• Dockerfile in docker/<framework>-rag/
• FastAPI application with 4 uniform endpoints
• depends_on ChromaDB healthcheck
• restart: unless-stopped for automatic recovery
• Environment variables for Ollama host, model, and ChromaDB connection

LangChain Pipeline

Property	Value
Version	0.3.35
Port	8100
Dockerfile	docker/langchain-rag/Dockerfile

Extraction approach: Maps file extensions to loaders:

Extension	Loader
.html	BSHTMLLoader
.pdf	PyPDFLoader
.docx	Docx2txtLoader
.md	UnstructuredMarkdownLoader
.txt	Plain open() read

RAG chain: RetrievalQA with ChromaDB retriever + Ollama LLM.

---

LlamaIndex Pipeline

Property	Value
Version	0.12.33
Port	8101
Dockerfile	docker/llamaindex-rag/Dockerfile

Extraction approach: Writes uploaded file to a temp directory, then uses SimpleDirectoryReader which auto-detects format.

RAG chain: VectorStoreIndex with ChromaDB vector store + Ollama via Settings.

---

Unstructured Pipeline

Property	Value
Version	0.17.2
Port	8102
Dockerfile	docker/unstructured-rag/Dockerfile

Extraction approach: Uses partition() which returns typed Element objects. Text content is joined from all elements.

RAG chain: Manual embedding + ChromaDB search + Ollama prompt (no wrapper chain).

---

Haystack Pipeline

Property	Value
Version	2.12.1
Port	8103
Dockerfile	docker/haystack-rag/Dockerfile

Extraction approach: Maps file extensions to converters:

Extension	Converter
.html	HTMLToDocument
.pdf	PyPDFToDocument
.docx	DocxToDocument
.md	MarkdownToDocument
.txt	TextFileToDocument

RAG chain: Document objects + ChromaDB retriever + Ollama generator.

---

ColPALI Pipeline

Property	Value
Port	8104
Dockerfile	docker/colpali-rag/Dockerfile

Extraction approach: Vision-based multimodal extraction — renders document pages as images and uses ColPALI embeddings rather than text-based parsing.

RAG chain: ColPALI embeddings + ChromaDB retriever + Ollama generator.

---

Container Environment

All pipeline containers receive the same environment variables:

environment:
  CHROMADB_HOST: chromadb
  CHROMADB_PORT: "8000"
  OLLAMA_HOST: http://host.docker.internal:11434
  OLLAMA_MODEL: ${OLLAMA_MODEL:-smollm2:135m}
  OLLAMA_EMBED_MODEL: ${OLLAMA_EMBED_MODEL:-nomic-embed-text}
  SYSTEM_PROMPT_FILE: ${SYSTEM_PROMPT_FILE:-}

Managing All Pipelines

# Status of all containers
docker compose ps

# View logs for a specific pipeline
docker compose logs -f langchain-rag

# Restart a single pipeline
docker compose restart haystack-rag

# Restart all pipelines
docker compose restart langchain-rag llamaindex-rag unstructured-rag haystack-rag colpali-rag

---

Harness Container

The test harness runs as a separate container with the test profile:

Property	Value
Dockerfile	docker/harness/Dockerfile
Network mode	host (accesses all services via localhost)
Reports	Mounted at ../reports:/app/reports

# Run full test suite
docker compose --profile test run --rm harness

---

Next Steps

• Inventory — Configuration reference
• Pipelines — Detailed framework comparison
• API Reference — Endpoint documentation