Docs/Security & Privacy

Security & Privacy

You're always in control. Pinchr gives you powerful tools to monitor, restrict, and approve your AI assistant's actions.

🔒

Required Reading — OpenClaw Security

Pinchr is powered by OpenClaw, an open-source AI agent engine. Pinchr inherits OpenClaw's security model — understanding it is essential before granting your agent access to your system.

Found a vulnerability? Report it at trust.openclaw.ai.

🛡️

Your Mac, Your Rules

Pinchr runs entirely on your Mac. No data is sent to our servers. Your conversations, files, and API keys stay local. Your agent only has the permissions you grant, and you can revoke them anytime. Pinchr's security model is built on top of OpenClaw's security framework.

Permission Scopes

Your agent requests permission before performing sensitive actions. You control what it can access:

📁
Filesystem Access

Control which folders your agent can read, write, or execute files in. You can whitelist specific directories or require approval for each operation.

✓ Allow: ~/Documents/Projects
✓ Allow: ~/Downloads
✗ Deny: ~/Documents/Personal
💻
Shell & Terminal

Your agent can run terminal commands for git, npm, docker, and more. You can require approval for destructive commands or allow trusted operations automatically.

✓ Auto-allow: git status, ls, cat
⚠️ Ask first: npm install, rm -rf
✗ Block: sudo commands
🌐
Network Access

Control which domains and APIs your agent can contact. Useful for preventing accidental data leaks or restricting internet access entirely.

✓ Allow: api.github.com
✓ Allow: *.anthropic.com
⚠️ Ask first: unknown domains
💬
Messaging & Communication

Permissions for sending messages via Slack, email, iMessage, or Discord. Require approval before sending, or trust your agent to respond on your behalf.

✓ Auto-send: internal Slack channels
⚠️ Approve first: emails to external contacts
🖥️
Browser & Desktop Control

Let your agent interact with desktop apps — clicking, typing, opening URLs. Requires macOS Accessibility and Screen Recording permissions.

⚠️ Approval required for all browser interactions

Configure permission scopes from Settings → Security → Permissions.

Approval Mode

In Approval Mode, your agent asks permission before every action:

Agent wants to:
Run command:
git push origin main

Approval Mode is great when you're learning how Pinchr works or working on sensitive tasks. You can enable it from Settings → Security.

Emergency Kill Switch

If your agent does something unexpected or you want to stop all activity immediately, use the Kill Switch:

🛑
Emergency Stop

Press ⌘⇧K or click the kill switch icon in the menu bar to:

  • Immediately halt all agent actions
  • Cancel pending tasks and automations
  • Disable all permissions until you re-enable them

The kill switch doesn't delete any data or history — it just pauses all agent activity. You can resume by re-enabling permissions in Settings.

Audit Log

Every action your agent takes is logged with full details. View the audit log from Settings → Security → Audit Log:

Recent Activity
Read file
success
~/Documents/Projects/api/server.js
2 minutes ago
Shell command
success
git commit -m "Fix auth bug"
5 minutes ago
Network request
success
GET https://api.github.com/user
12 minutes ago
Shell command
denied
rm -rf node_modules
20 minutes ago

Logs are stored locally and encrypted. You can export or clear your audit log anytime.

Session Monitoring

Watch what your agent is doing in real-time from the Activity Monitor:

📊
Live Activity Feed

See every action your agent takes as it happens — file reads, commands, network requests.

🔍
Resource Usage

Monitor CPU, memory, and API usage. Set limits to prevent runaway costs or resource spikes.

⏱️
Task Duration

See how long tasks take. Identify slow operations or bottlenecks in your workflows.

🚨
Anomaly Detection

Get alerts if your agent behaves unusually — excessive API calls, unexpected network access, etc.

Data Privacy

Here's what happens to your data:

All data stays local

Conversations, files, and agent actions are stored on your Mac. Nothing is sent to Pinchr servers.

API keys are encrypted

Your Anthropic or OpenAI API keys are encrypted using macOS Keychain. They never touch our servers.

LLM providers see only what you allow

Your agent sends context to Anthropic/OpenAI APIs for reasoning. You can exclude sensitive files or folders.

Telemetry is minimal and opt-in

We collect anonymized crash reports and feature usage (if you opt in). No personal data, file contents, or conversations.

Security Best Practices

  • Start with Approval Mode enabled while you get comfortable with Pinchr
  • Review the audit log weekly to understand what your agent is doing
  • Use folder whitelists to restrict filesystem access to work directories only
  • Set API rate limits to prevent accidental cost spikes
  • Never share your API keys or Pinchr data folder with others
  • Keep Pinchr updated — we ship security improvements regularly

Security Concerns?

We take security seriously. Reach out if you have questions or need help.