Configuration
ConfigurationPrivacy & Cloud ModeSmart RulesEffective ConfigurationPersistent DecisionsPause & ResumeProtections
Checks ReferenceShieldsSandboxSecurity PostureContent Scanner (DLP)Trusted HostsCanary CredentialsSkill PinningRepository ScanningEgress ControlBadgesHelp & Support · node9 documentation
Help & Support
Get unstuck quickly.
Frequently Asked Questions
Does Node9 send my AI agent's conversation to the cloud?
Not in local mode. Node9 evaluates the concrete tool call parameters — tool name and arguments — and does not transmit conversational context or AI-generated prose to any remote service. The Response DLP scanner does read Claude's local JSONL conversation history on disk to flag secrets in model output, but that scan stays on your machine. Cloud mode (node9 login <key>) only streams intercept metadata for tool calls, never the conversation.
What is the difference between local mode and cloud mode?
Local mode (node9 login --local) sets approvers.cloud to false, keeping all policy decisions on your machine — nothing is sent to the cloud. Cloud mode (node9 login <key>) sets approvers.cloud to true, streaming intercept events to Mission Control and allowing your policy to be managed centrally from the Cloud Policy Studio. Cloud mode is opt-in.
My agent is being blocked too aggressively. What should I do?
Check your dangerousWords list first — overly broad keywords cause false positives. Also review ignoredTools: add the tool names your agent uses for read-only operations so they pass through without inspection. Run node9 status to confirm which config file is active, and node9 explain <tool> '<args-json>' to see exactly which rule fires.
How do I use Node9 with multiple AI agents (Claude + Gemini) on the same machine?
Run node9 init once; it wires every agent it finds. To add one later, node9 agents add <agent>; node9 agents list shows what is wired. Both agents will share the same Node9 policy config. If you need different policies per agent, create separate project directories each with their own node9.config.json, and use NODE9_API_KEY or named profiles to assign different workspaces.
What is the config hierarchy? Which file takes priority?
Node9 merges per-field across layers — it does not pick a single winning file. Layers from lowest to highest precedence: 1) hardcoded defaults, 2) ~/.node9/config.json (global), 3) ./node9.config.json (project), 4) the cloud workspace (~/.node9/rules-cache.json), 5) active shields (~/.node9/shields.json). On a machine that follows a workspace, layers 2 and 3 are ignored for policy: the workspace replaces them. Most fields are merge-or-additive (smartRules, sandboxPaths, ignoredTools, toolInspection); dangerousWords replaces rather than merges. See the Configuration section for the full per-field merge table.
How do I check what Node9 is doing right now?
Run node9 status for current mode, active profile, connected workspace, and audit-event counts. node9 audit prints the formatted audit trail; node9 tail streams every tool call live; node9 report summarizes allowed/blocked/DLP/cost over a time window. The raw log is at ~/.node9/audit.log.
Can I run Node9 without a cloud account?
Yes. Run node9 login --local to set approvers.cloud to false. Node9 will enforce your local config policy, write an audit log to ~/.node9/audit.log, and never connect to the cloud. You can also run node9 init and node9 agents add without ever running node9 login.
Does Node9 work with Cursor, Windsurf, or other coding assistants?
Yes, with a difference that matters. Claude Code, Codex, Copilot CLI, Gemini CLI, Antigravity and Hermes get a pre-execution hook that sees every tool call; OpenCode and Pi get a plugin. Cursor, Windsurf, VS Code and Claude Desktop expose no hook API, so node9 governs what goes through their MCP servers and nothing the editor does on its own. Run node9 agents add <agent>, and node9 mcp gateway --all to route every MCP server through node9.