No-leak LLM mode

All requests route to local models. Cloud providers blocked at the network layer.

What it is

When enabled, no-leak mode routes every model request to a local model on your device. Cloud provider hostnames (api.openai.com, claude.ai, etc.) are blocked at the network layer — even if a tool or plugin tries to call them, the request fails.

This is the strictest privacy mode in the platform. Combined with local-first storage, you get a workspace where no piece of your data has ever touched anyone else's server.

Key capabilities

Network-layer block on cloud LLM endpoints
Visible "no-leak" indicator in the chat
Audit log shows every blocked attempt
Per-workspace policy — admins enforce, users can't opt out
Works with llama.cpp, Ollama-compatible runtimes, MLX

Use cases

Legal and healthcare workflows under strict PII rules
Defence and government workspaces
Compliance teams during audit windows