SuperLocalMemory has three modes. All retain the local canonical memory store; the mode selects the model/provider behavior used for configured enrichment and retrieval work. Switch at any time with slm mode a, slm mode b, or slm mode c; stored memory is not discarded by a mode change.

Mode A stores and retrieves through the configured local data root. It can use semantic, BM25, temporal, Hopfield, and spreading-activation candidates when their dependencies are healthy, followed by fusion and configured scoring.

“Local core” is not an absolute network claim. Model/dependency downloads, connectors, cloud backup, proxy providers, and other integrations can use the network when explicitly enabled. Review the complete deployment before making a privacy or compliance determination.

Mode B adds a local LLM path through Ollama. The operator chooses the endpoint and model, then verifies both with slm doctor and a representative store / recall witness. Answer and extraction quality depend on that selected model, the corpus, and the healthy retrieval channels.

# Install Ollama using reviewed platform instructions, then:
ollama pull llama3.2
slm mode b
slm doctor

Mode C enables configured provider-backed enrichment and retrieval behavior while the canonical memory database remains local. Configure the provider with slm provider set and use environment-managed credentials where appropriate. The provider’s data-processing terms and the deployment’s network controls apply to content sent through this mode.

Use slm trace "representative query" to see the evidence and channels used by an actual recall rather than inferring behavior from the selected mode.

Part of Qualixar | Created by Varun Pratap Bhardwaj