llama-index-pydocker

LlamaIndex store wrappers that spin up Docker containers automatically — one import swap away.

pip install llama-index-pydocker

llama-index-pydocker wraps every LlamaIndex store class that needs a running database. Pass a localhost URL and the right Docker container starts automatically via py-dockerdb. Pass a cloud URL and the wrapper is invisible — nothing Docker-related ever runs.

Switch from a local Neo4j graph store to a hosted one without touching your pipeline. Prototype a pgvector RAG index on your laptop, deploy to RDS with one URL change. Give every student an isolated vector store in seconds.

When to use this

• Local GraphRAG prototype: one import swap gives you a Neo4j container that LlamaIndex talks to directly, with automatic teardown via context manager.
• pgvector RAG on localhost: spin up Postgres + pgvector, run your retrieval pipeline, shut everything down — without ever opening a terminal.
• Qdrant / OpenSearch experiments: same pattern, same API. Swap backends by changing the URL and the docker_config type.
• Production passthrough: point any store at a cloud endpoint and llama-index-pydocker behaves exactly like the upstream LlamaIndex class.

Supported Stores

Prerequisites

• Python 3.10+ · Docker running · py-dockerdb installed

Installation

pip install llama-index-pydocker                   # core only
pip install "llama-index-pydocker[neo4j]"          # + Neo4j graph store
pip install "llama-index-pydocker[postgres]"       # + pgvector store
pip install "llama-index-pydocker[qdrant]"         # + Qdrant vector store
pip install "llama-index-pydocker[opensearch]"     # + OpenSearch vector store
pip install "llama-index-pydocker[all]"            # everything

Usage

Change one import. Everything else stays the same.

Neo4j Graph Store

# Before
from llama_index.graph_stores.neo4j import Neo4jGraphStore

# After — Docker container starts automatically on localhost URLs
from llama_index_pydocker import Neo4jGraphStore

from llama_index_pydocker import Neo4jGraphStore

store = Neo4jGraphStore(
    url="bolt://localhost:7687",
    password="test",
)
# Container is up, store is ready
store.stop()          # remove container when done

Use a context manager for automatic teardown:

with Neo4jGraphStore(url="bolt://localhost:7687", password="test") as store:
    index = KnowledgeGraphIndex.from_documents(docs, storage_context=..., graph_store=store)
    response = query_engine.query("Who founded Neo4j?")
# Container removed here

Point at a cloud instance — Docker is never touched:

store = Neo4jGraphStore(
    url="bolt://my-aura-instance.databases.neo4j.io:7687",
    password="secret",
)

pgvector Store

from llama_index_pydocker import PGVectorStore

store = PGVectorStore(
    connection_string="postgresql://user:pass@localhost:5432/vectordb",
    embed_dim=1536,
)
store.stop()

Override Docker config (volume, retries, container name, …):

from docker_db import PostgresConfig
from llama_index_pydocker import PGVectorStore

cfg = PostgresConfig(
    user="user", password="pass", database="vectordb",
    project_name="my-rag",
    retries=30,
)

with PGVectorStore(
    connection_string="postgresql://user:pass@localhost:5432/vectordb",
    embed_dim=1536,
    docker_config=cfg,
) as store:
    index = VectorStoreIndex.from_documents(docs, vector_store=store)
    print(index.as_query_engine().query("What is pgvector?"))

Qdrant Vector Store

from llama_index_pydocker import QdrantVectorStore

with QdrantVectorStore(
    collection_name="docs",
    url="http://localhost:6333",
) as store:
    index = VectorStoreIndex.from_documents(docs, vector_store=store)
    print(index.as_query_engine().query("What is Qdrant?"))

Remote cluster — no Docker:

store = QdrantVectorStore(
    collection_name="docs",
    url="https://my-cluster.qdrant.io:6333",
    api_key="sk-...",
)

OpenSearch Vector Store

from llama_index_pydocker import OpensearchVectorStore

with OpensearchVectorStore(
    index_name="docs",
    endpoint="http://localhost:9200",
    embed_dim=1536,
) as store:
    index = VectorStoreIndex.from_documents(docs, vector_store=store)
    print(index.as_query_engine().query("What is OpenSearch?"))

Routing logic

url / connection_string / endpoint
        │
        ▼
 is_localhost(host)?
   ┌────┴────┐
  YES        NO
   │          │
   ▼          ▼
 create    passthrough
 Docker    (behaves like
container  upstream class)

Port is always inferred from the URL, even when a docker_config is supplied.

More examples

Full runnable notebooks are in usage/:

Neo4j / GraphRAG · pgvector RAG · Qdrant · OpenSearch

Development

git clone https://github.com/amadou-6e/llama-index-pydocker.git
cd llama-index-pydocker
pip install -e ".[all]"

Testing

python -m pytest tests/test_utils.py
python -m pytest tests/test_neo4j.py
python -m pytest tests/test_postgres.py
python -m pytest tests/test_qdrant.py
python -m pytest tests/test_opensearch.py

Contributing

PRs welcome. Include tests for behaviour changes.

License

MIT License. See LICENSE.