Cannot switch Google accounts: login reuses cached browser profile + crashes on new account

[Halmurat-Uyghur]

Bug Summary

notebooklm login cannot switch to a different Google account. Two issues:

1. Cached browser profile auto-logs in: Even after deleting ~/.notebooklm/storage_state.json, running notebooklm login automatically logs into the previously authenticated account because the browser profile at ~/.notebooklm/browser_profile/ retains Google session cookies.

2. Crash when adding new account: When attempting to click "Use another account" on the Google sign-in page, the browser closes and throws a TargetClosedError.

Steps to Reproduce

1. Authenticate with Account A: notebooklm login
2. Delete auth state: rm ~/.notebooklm/storage_state.json
3. Run notebooklm login again — it auto-logs into Account A (no account picker shown)
4. If you manually navigate to "Use another account" before it auto-redirects, the browser closes with:

playwright._impl._errors.TargetClosedError: Page.goto: Target page, context or browser has been closed

Expected Behavior

• A notebooklm logout command (or notebooklm auth logout) that clears both storage_state.json and the browser profile session
• OR notebooklm login --new / notebooklm login --account flag that opens a fresh browser context without cached Google sessions

Workaround

Manually delete the entire browser profile directory before re-login:

rm -rf ~/.notebooklm/browser_profile ~/.notebooklm/storage_state.json
notebooklm login

Environment

• notebooklm-py: v0.3.4
• Python: 3.14
• OS: macOS (Darwin 24.6.0)
• Playwright browser: Chromium (bundled)

Interested in Contributing

Happy to submit a PR for this if the maintainers agree on the approach (e.g., adding a logout command or a --fresh flag to login).

[teng-lin]

Root Cause Analysis

Thanks for the detailed report! I've traced through the code and identified the root causes for both issues.

Bug 1: Login reuses cached browser profile

Root cause: session.py:265 — the login command uses Playwright's launch_persistent_context(user_data_dir=browser_profile), which stores Google session cookies in a Chromium user data directory (~/.notebooklm/profiles/<profile>/browser_profile/).

Deleting storage_state.json only removes the exported cookie snapshot used by the Python HTTP client. The actual Google session cookies remain in the Chromium profile's cookie database (browser_profile/Default/Cookies). On the next notebooklm login, Playwright reopens that same profile, Google sees valid session cookies, and auto-redirects past the account picker.

There is currently no mechanism to clear the persistent browser context — no logout command, and the clear command (session.py:547) only clears notebook context (context.json), not auth state.

Bug 2: TargetClosedError when clicking "Use another account"

Root cause: When using launch_persistent_context with an existing Google session, clicking "Use another account" triggers a sign-out + re-authentication flow. This can destroy and recreate pages within the browser context, making the page reference at session.py:267 stale. The error handling in the retry loop (session.py:270-304) only catches RETRYABLE_CONNECTION_ERRORS (ERR_CONNECTION_CLOSED, ERR_CONNECTION_RESET), not TargetClosedError.

This bug is a consequence of Bug 1 — users only encounter it because they're forced to manually navigate to the account switcher within a persistent context that was never designed for that flow.

Recommended Fix

1. Add --fresh flag to login (addresses Bug 1 directly)

When --fresh is passed, delete browser_profile/ directory (via shutil.rmtree) before calling launch_persistent_context. This gives users a clean browser context with no cached Google sessions.

notebooklm login --fresh   # Start with clean browser, no cached session

2. Add notebooklm auth logout command

Delete both storage_state.json and browser_profile/ for the active profile. Provides the expected CLI symmetry (login / logout).

3. Handle TargetClosedError in login flow (addresses Bug 2)

In the navigation/wait section, catch TargetClosedError and recover by getting a fresh page from context.pages or context.new_page().

Files to modify

• src/notebooklm/cli/session.py — add --fresh flag, auth logout subcommand, TargetClosedError handling
• tests/unit/cli/test_session.py — tests for new functionality

Edge cases to handle

• --fresh on first login (no browser_profile/ exists yet) — guard with if exists()
• logout when already logged out — no-op with friendly message
• Windows: shutil.rmtree may fail if browser profile is locked — catch PermissionError with helpful message
• Legacy layout: check both legacy and profile-based paths

---

@teng-lin/notebooklm-py maintainers — the contributor has offered to submit a PR. I think the approach above is solid and the scope is well-contained. Welcome the PR! 🎉