Notebook Binding + NotebookLM (Web)

This guide covers the user-facing Web flow for connecting NotebookLM and choosing which notebooks CCCC should use.

The Web UI is intentionally minimal:

1. connect Google
2. choose the Work Notebook
3. choose the Memory Notebook

Actual NotebookLM operations such as query, ingest, source management, artifacts, and job handling are handled by agents through MCP / CLI surfaces, not by the normal user settings page.

1. Provider Activation

Connecting Google in CCCC Web stores the provider credential and activates the real NotebookLM path. CCCC does not require an environment toggle.

If you expose Web outside localhost, first create an Admin Access Token in Settings > Web Access and keep the service behind a network boundary until that token exists.

2. Open Notebook Settings

1. Open a target group in Web.
2. Open Settings.
3. Open the Notebook tab.

3. Connect Google

In Google Account:

1. Click Connect Google.
2. Complete sign-in in the interactive browser view shown inside CCCC Web.
3. Wait until the account status becomes connected.

Notes:

• If a valid credential is already stored, reconnect may complete without a full browser login.
• The default Web page does not expose manual credential editing anymore.
• The Web flow uses a projected sign-in browser so Docker / remote deployments do not need a local desktop browser on the daemon host.
• The projected sign-in browser now runs in headed mode for better Google compatibility. In server/container environments without a native display, CCCC uses Xvfb automatically.
• The Docker image installs system Chromium and its display dependencies during the image build; projected sign-in does not perform a lazy browser download.

4. Bind the Work Notebook

In Work Notebook:

1. Choose an existing notebook from the selector, or
2. Click Create and bind new.

Use Work Notebook for shared project knowledge and working materials.

Expected result:

• Work binding becomes Bound.
• The current notebook title/id updates immediately.

5. Bind the Memory Notebook

In Memory Notebook:

1. Choose an existing notebook from the selector, or
2. Click Create and bind new.

Use Memory Notebook for finalized memory recall.

Expected result:

• Memory binding becomes Bound.
• The current notebook title/id updates immediately.

6. Connection Summary

Use Connection Summary only as a lightweight status snapshot:

1. Google connected or not
2. Work notebook bound or not
3. Memory notebook bound or not
4. a short warning message if something is degraded

The summary is intentionally human-oriented and does not expose internal queue/job/runtime details.

7. What the Web Page No Longer Does

The normal user-facing Web settings page no longer exposes these agent/developer operations:

1. Notebook query
2. ingest submission
3. source management
4. artifact generation/download
5. job queue operations
6. manual credential write/clear
7. provider health check

That is by design.

8. Agent-Side Usage Still Exists

NotebookLM usage still exists through agent-facing surfaces:

1. MCP tools
2. CLI
3. prompt/help-guided agent workflows

The Web page is now only for account connection and notebook binding.

9. Disconnect

Use Disconnect Google in the Notebook settings when this machine should no longer use the stored NotebookLM credential.

10. Explicit source ingestion and legacy sync state

For 0.4.36, files are added explicitly from the attached project scope. For example, a caller can ingest <scope_root>/space/spec.md; CCCC validates the resolved path and uploads it without turning the entire directory into an automatic mirror.

<scope_root>/space/

Homes upgraded from Python 0.4.35 may still contain these legacy status files:

• <scope_root>/space/.space-index.json
• <scope_root>/space/.space-sync-state.json
• <scope_root>/space/.space-status.json
• <scope_root>/space/.sync/remote-sources/*.json
• <scope_root>/space/artifacts/notebooklm/...

The native Rust adapter follows the notebooklm-py v0.8.1 protocol baseline for the current personal-app host, source requests, artifact status values, and audio output format. It supports attached-scope local files, pasted text, Web URL, YouTube, and Google Drive Docs/Slides/Sheets ingestion.

Each explicit ingest is recorded before the provider write. A normal request makes one provider attempt and settles the job; failed jobs are retried only by an explicit space jobs retry, not by a hidden background loop. A job left running after a process interruption or an ambiguous provider response represents an uncertain provider result. Repeating the same ingest is deduped to that durable job, and direct retry is blocked until the user inspects and reconciles the remote notebook or cancels the job. Retrying a terminal old job is also refused if the Group's current work binding no longer matches the notebook saved on that job.

Automatic work- and memory-lane mirroring is retired in 0.4.36. Rust reads the same canonical metadata so existing status survives an upgrade, but it never resumes the old remote mutation loop. The CLI, Web API, and MCP surface no longer advertise a sync action.

These implementation details matter for agent/developer workflows, but they are not part of the normal user-facing binding flow.