Neboy72

Self-hosted long-term memory on Qdrant with automatic per-turn prefetch and write-back, plus recall/remember/forget/guardrail tools and a knowledge graph. Runs fully local with Ollama or sentence-transformers embeddings; cloud embedders (Voyage, OpenAI, Google, Jina) are opt-in via their own API keys, and query rewriting plus paid consolidation stations stay off until enabled. Note that an embedding key already present in the environment is auto-detected, so the README's egress section is the authoritative list of what can leave the machine. Hermes and any MCP-connected agent share one collection, so they read and write the same memories. Disclosure — an embedding key already in Hermes' environment (VOYAGE/OPENAI/GOOGLE/JINA_API_KEY) is auto-detected and sends turn text to that provider for embedding; session-end fact extraction sends conversation text to the model endpoint in Hermes config.yaml; reads ~/.hermes/.env and config.yaml; state and daily backups under ~/.nexus-memory; an opt-in NEXUS_UPDATE_CHECK=1 adds a GitHub release check and an upgrade nudge to the system prompt.