System Requirements#

This page describes runtime and tooling requirements for the Vector Retriever microservice.

Supported Platforms#

  • Linux (recommended for Docker-based deployment)

Required Software#

  • Docker 24.x or newer

  • Docker Compose (v2 plugin)

  • Python 3.11 - 3.12 for local development and tests

  • Poetry 1.8+ for dependency management

Runtime Dependencies#

  • Reachable embedding endpoint (EMBEDDINGS_ENDPOINT, or MULTIMODAL_EMBEDDING_ENDPOINT via setup.sh)

  • Embedding model name (EMBEDDING_MODEL_NAME)

Backend-specific dependencies:

  • vdms: VDMS Vector DB endpoint

  • milvus: Milvus endpoint (MILVUS_URI)

  • pgvector: PostgreSQL with pgvector and psycopg3 connection string

  • faiss: local in-process index (optional disk path for persisted index)

Minimum Resource Guidance#

For local validation:

  • CPU: 4 cores

  • Memory: 8 GB RAM

  • Disk: 5 GB free for images and logs

For production, size resources based on query volume, embedding latency, and backend throughput.

Network/Proxy Notes#

If you are behind a proxy, configure:

  • http_proxy

  • https_proxy

  • no_proxy

Validation Checklist#

  • Docker and Compose are available in shell

  • Embedding endpoint is reachable from retriever container

  • Selected backend endpoint is reachable

  • GET /health returns ok

  • GET /ready returns ready

Supporting Resources#