hx-mem0-advisor

Deploy Mem0 OSS natively on Ubuntu Linux with systemd for long-running services, using Memory.from_config with a Qdrant vector store and Ollama LLM and embedder. Containerized deployments are rejected unless the infrastructure owner explicitly changes the architecture standard.

Configure Mem0 using Memory.from_config with a Qdrant vector store specifying collection_name and embedding_model_dims, plus an Ollama embedder with model, embedding_dims, and ollama_base_url. Keep vector dimensions aligned with the accepted embedding model and never mix embedding identities in one collection.

Mem0 Platform is classified REJECT_FOR_HX_RUNTIME because it conflicts with the native self-hosted HX-13 architecture. The hosted mcp.mem0.ai endpoint and the mem0-oss-to-platform migration skill require an explicit owner decision before any use.

Memory is the self-hosted OSS class configured with local vector stores and providers, while MemoryClient connects to the hosted Mem0 Platform. HX-13 uses the OSS Memory path; migrating to MemoryClient removes local Qdrant and Ollama configuration and contradicts current placement.

Validation follows the E3 smoke test: initialize Mem0 with accepted providers, store a synthetic memory, retrieve the exact token via semantic search with infer=False, delete it, and clean up the disposable Qdrant collection. The E4 test then validates the assigned Mem0 MCP companion separately.

Do not install Mem0 coding-assistant plugins across an agent fleet merely because the upstream repository supports them. Plugin rollout requires separate governed admission covering scope, credential model, persistence policy, and validation, distinct from the HX-13 runtime role.