Skip to main content
Privacy mode forces every LLM request to local Ollama regardless of the configured provider — a single switch, not a per-component opt-in.

​
Enable

Via a slash command in any channel:
Or via the REST API:
This is a runtime toggle on the live ModelRouter — it isn’t written back to config.toml, so it resets to false on the next gateway restart. Set it again after restarting if you want it always on.

​
What changes

STT (faster-whisper) and embeddings (sentence-transformers) already run fully locally regardless of privacy mode.

​
Local model requirements

  1. Ollama running locally: ollama serve
  2. A local model pulled: ollama pull llama3.2:1b (or your configured models.local ID)

​
Verifying zero external calls

The privacy audit log records the destination host of every outbound HTTP call and persists across restarts (SQLite at ~/.neuralcleave/privacy_audit.db, 90-day retention). After enabling privacy mode, a fresh session’s report should show no external destinations:
See Privacy API for the full report/clear reference.

​
Competitive context

OpenClaw’s equivalent privacy controls are opt-in per-module. NeuralCleave’s /privacy on (or privacy_mode: true) is a single switch that enforces the LLM-traffic constraint and gives an auditable, persistent proof via the report endpoint.