hermes-memory-setup

Run `hermes memory setup` for an interactive provider selection, or specify one directly such as `hermes memory setup mem0 --mode oss --oss-vector qdrant`. Check the active provider with `hermes memory status` and disable it with `hermes memory off`.

For free local-first setups, use Mem0 OSS with Qdrant, which needs no server. For self-hosted knowledge bases use OpenViking, and for knowledge graphs with cross-memory reasoning use Hindsight. The built-in memory suffices for 10-15 facts.

No, Mem0 OSS only supports Qdrant and PGVector as vector stores. Qdrant runs in local file mode without a server, while PGVector requires a PostgreSQL server. ChromaDB and Milvus are not supported.

The 429 errors occur when Mem0's LLM extraction points to a FreeLLM router whose upstream providers are all rate-limited. Fix it by setting `openai_base_url` in mem0.json to a direct upstream endpoint like `https://api.deepseek.com/v1` instead of the FreeLLM router.

No, DeepSeek has no embeddings API and returns 404 when used as an embedder. Use OpenAI's paid embeddings, a local Ollama model like nomic-embed-text, or a FreeLLM API endpoint that routes to free embedding providers.

Hermes has two context_length layers: the model-level `model.context_length` in config.yaml acts as a global cap that overrides per-provider settings. Raise it with `hermes config set model.context_length 1000000` to unlock the provider-level value.