Agent Sentry
Support
Something wrong, missing or confusing, or an idea for the app? Tell me: I read every message.
Report from the app
Open Agent Sentry's Settings › Support, say what happened, and click Send Report. It includes the log I need to find the bug, and you can preview it first.
Write to piotr@codice.pl. Mention your Mac model and macOS version, and the agent involved, or attach a report copied from Settings › Support.
Ideas
Missing a feature, or have an idea for Agent Sentry? Send it by email. Suggestions and comments help decide what comes next.
Questions
Do I need Agent Sentry Guard?
No. Without it, Agent Sentry recognizes your agents, shows what they do, and decides which servers they may reach. The Guard adds rules for files and programs, and follows programs that detach from their agent. Read what the Guard adds.
Which agents does it recognize?
Claude Code, Codex, Gemini CLI, Cursor and Cursor Agent, GitHub Copilot, Antigravity, Aider, OpenCode, Goose, Amp, Qwen Code, Crush and others, the Claude, ChatGPT and Gemini apps, editors such as Visual Studio Code and Zed, and model servers such as Ollama and LM Studio. If one you use isn't recognized, send a report while it runs: it says which agents Agent Sentry saw.
What do “Unconfirmed” and “Inferred” mean?
“Unconfirmed”: Agent Sentry knows the agent by its name or where it's installed, but not by its maker's signature, as with agents that run on Node or Python. Its rules still apply, and stricter ones where they differ. With the Guard installed, agents the App Store's sandbox keeps Agent Sentry from checking are confirmed by the Guard instead.
“Inferred”: Agent Sentry knows the agent only from a variable the process inherited, such as Claude Code's CLAUDECODE. The same stricter rules apply.
An agent can't reach a server it needs.
Open the agent's card in the menu bar panel to see which host was blocked and by which rule. Add an Allow rule for that host in Settings › Network › Rules, for that agent or for all of them.
Was a connection opened before network protection started checked?
No. macOS shows a network filter only the connections that start after it does, so a connection an agent already had open, such as a long-lived streaming link, goes on unchecked until it closes and the agent connects again. Restart agents that were running when you installed or turned on network protection, or when Agent Sentry was updated. For a minute after the filter starts, Settings › General says so.
An agent can't sign in after I turned on blocking.
Some agents read their sign-in from your keychain with macOS's security command, which the built-in Keychain access rule blocks. An agent confirmed by its maker's signature may still read its own sign-in. If the agent isn't confirmed, for example because it was installed with npm, limit Keychain access to the agents you don't trust, or turn blocking off, in Settings › Files › Rules. The Keychains rule only records which agents open your keychain, so it never stops a sign-in.
Can an agent turn Agent Sentry off?
Not by running commands. The rules are kept by macOS system extensions that only answer Agent Sentry, and a built-in rule keeps every agent out of Agent Sentry's own settings. Removing the extensions takes an administrator's password, and with the Guard installed an agent can't quit Agent Sentry either.
How do I uninstall Agent Sentry?
Quit Agent Sentry and move it to the Trash. macOS asks for an administrator's password, then removes its network filter with it. Your rules and settings stay on the Mac. To remove those too, along with Agent Sentry Guard, use Uninstall Everything in Agent Sentry Guard. How it works.
Which Macs are supported?
Macs with Apple silicon, running macOS 26 or later.