Frequently Asked Questions

Find concise, accurate answers to the most common questions about Zenborg.

1. Does Zenborg modify my active Git branch or working directory?

No. All loop stages execute inside temporary Git worktrees located in data/agent-loop-worktrees/<loop-id>. Your working directory and checked-out branch are never modified during a run. You merge completed changes only when you approve the diff.

2. Does Zenborg charge me for AI tokens?

No. Zenborg has no token pricing and adds no surcharge to model usage. It runs directly on the CLI tools you have already authenticated (Claude Code, Google Antigravity, Gemini CLI, Hermes, OpenCode) or your browser subscriptions.

3. Can Zenborg run offline?

Yes. License signature verification is 100% offline using Ed25519 cryptography. If you run local agents, loops can run without an internet connection.

4. Where does Zenborg store my credentials?

On macOS, Zenborg stores credentials securely inside the native macOS Keychain. On Windows, credentials are stored in restricted files (permissions mode 0600) in your user profile directory.

5. How does Zenborg prevent runaway agents or machine freezes?

Zenborg enforces strict system limits:

Per-stage resident memory cap: 6,144 MB (6 GB).
Node heap cap: 3,072 MB (3 GB).
System free memory safeguard: 8% minimum free RAM.
3-hour stage time budget.
25-minute inactivity watchdog.
Maximum 10 conversation relay turns (MAX_TURNS = 10).

6. Can I use my ChatGPT Plus or Gemini Advanced web subscriptions?

Yes. Zenborg controls an isolated Chrome browser profile (~/.zenborg-browser). You log in once, and Zenborg uses your web session to run plan drafts, reviews, and syntheses with zero API fees.

7. What happens when a web worker hits a CAPTCHA?

If you have a fallback agent configured, Zenborg automatically hands off the task without pausing and displays: "<Provider> hit a CAPTCHA; the loop continued on another agent." If no fallback agent is configured, the loop pauses and displays an "Open browser to log in" banner so you can solve the puzzle.

8. Can I resume a failed or cancelled loop?

Yes. If a loop fails due to a network glitch or provider rate limit, the "Resume Loop" button allows you to resume from the failure point as long as the initial plan draft completed.

9. Why are third-party MCP servers disabled during loops?

Loading many external MCP servers floods the agent’s context window with thousands of tokens of tool descriptions. This wastes money, slows down models, and causes tool hallucinations. Zenborg enforces strict MCP isolation (--strict-mcp-config and --ignore-user-config) so agents focus strictly on your code.

10. How does Zenborg verify the safety of agent skills?

Every installed skill is evaluated by a built-in security scanner against 15 specific rules, including prompt injection, hidden unicode, secret exfiltration, and destructive file system commands. High-risk skills are quarantined.

11. How do Agent Power settings affect my machine?

Low: Runs 1 loop and 1 task with a 0.6 CPU load factor, keeping your computer cool and responsive.
Medium: Runs up to 2 loops and 2 tasks with a 0.8 CPU load factor.
High: Maximizes concurrency (up to the larger of 2 or cores ÷ 4 loops and 3 tasks) on high-end workstations.

12. What happens when the public beta ends?

The beta period ends on 2026-11-30T23:59:59.000Z (BETA_END). After that date, your installation automatically transitions to the permanent Free tier unless you purchase a subscription or lifetime license.

13. Does Zenborg work on Windows?

Yes. Zenborg runs on Windows 10 and 11 (x64). It cleans up child processes using taskkill /T /F and monitors free memory percentage rather than POSIX load averages.

14. What are the Free plan limitations?

The Free plan allows 1 project, 1 running loop at a time, 1 automated browser tab, and locks Agent Power to Low. All core loop features (worktrees, multi-stage reviews, and fallback chains) remain available.

15. How do I check my installed version?

Open Settings (#settingsDialog) and scroll to the About section. The #aboutVersionText element displays your version number, packaging status, and build timestamp.