Middleware and External Services
中文 · English
NexusDesk requires PostgreSQL for business data and task queues. Other services depend on the enabled capabilities:
| Service | Purpose | Requirement |
|---|---|---|
| PostgreSQL | Business data, runtime and task queues, and optional pgvector retrieval for customer memory | Required; deployment templates use PostgreSQL 17 |
| Milvus | Vector indexing and retrieval for the knowledge base | Required for knowledge vector/hybrid search, not keyword-only search |
| MinIO / S3 | Original document storage | Optional; PostgreSQL is used when object storage is not configured |
| Model APIs | Chat, Embedding, Rerank, OCR, and speech operations | Configure when using the corresponding real model capability |
| External document parser | OCR and specialized document parsing | Configure when local parsing is insufficient |
Task scheduling does not require Redis, RabbitMQ, Kafka, or Celery. Knowledge vector/hybrid search uses Milvus; customer-memory vector retrieval can use PostgreSQL pgvector when the extension and embedding profile are available, which does not require Milvus.
See the deployment guide, production environment template, and backend environment template. Keep credentials in environment configuration rather than source code or images. The Chinese middleware guide lists all settings and Compose boundaries; the environment-specific server record is retained only in the historical archive.