NotebookLM
NotebookLM is an optional recall layer for Agenties. It lets you upload curated session summaries, specs, and research notes into a Google NotebookLM notebook, then query them semantically from Agenties. It is not a replacement for local project state.
Role in the memory stack
| Layer | Best for | Not for |
|---|
| .agenties/ | Exact state, issues, agents, sessions, reports, graph files, and continuity journal. | Fuzzy semantic search. |
| Obsidian | Human-readable project knowledge and manual review. | Replacing local operational state. |
| NotebookLM | Semantic questions over curated sources and session material. | Storing every raw agent log or acting as the only memory source. |
Note:NotebookLM is active only when configured with a notebook ID and the local Python CLI is installed and authenticated. If the CLI is not found, all NotebookLM operations gracefully report "not found" without breaking the app.
How it works
Agenties integrates with NotebookLM through the open-source notebooklm Python CLI tool, installed in a local virtual environment at ~/.notebooklm-venv/. All NotebookLM operations are subprocess calls to this CLI. There is no official Google API — this is a community CLI that automates the NotebookLM web interface.
CLI location
Windows: %USERPROFILE%\.notebooklm-venv\Scripts\notebooklm.exe
macOS/Linux: ~/.notebooklm-venv/bin/notebooklm
Setup
NotebookLM login must be done in a terminal — it requires interactive browser authentication that cannot be started from inside Agenties. The Settings panel shows instructions and copy-pasteable commands.
| Step | Action |
|---|
| 1 | Install Python 3.10+ and create the venv: python -m venv ~/.notebooklm-venv |
| 2 | Install the CLI: ~/.notebooklm-venv/Scripts/pip install notebooklm |
| 3 | Authenticate in a terminal: ~/.notebooklm-venv/Scripts/notebooklm login |
| 4 | In Agenties, go to Settings → Integrations → NotebookLM and verify login status. |
| 5 | Create or select a notebook for the project. |
Warning:Authentication tokens expire. If Agenties shows "not authenticated", run notebooklm login again in a terminal. The keepalive check runs every 8 minutes to detect expiry and update the UI badge.
Available operations
| Operation | What it does |
|---|
| List notebooks | Fetches all notebooks from the connected account. |
| Create notebook | Creates a new named notebook and returns its ID. |
| Verify notebook | Confirms a notebook ID is still valid. |
| Add source | Uploads a local Markdown or text file as a source to a notebook. |
| List sources | Shows all sources currently uploaded to a notebook. |
| Ask | Asks a natural language question against the notebook sources. |
Workflow
| Moment | What Agenties does |
|---|
| Project setup | Creates or selects a NotebookLM notebook for the project when configured. |
| Session wrapup | Writes a Markdown summary to a temp file and adds it as a source to the notebook via source add. |
| Research recall | Uses the ask command for fuzzy recall over curated project material and returns concise context to the orchestrator. |
Tip:Upload summaries, specs, and decisions — not raw conversation logs. NotebookLM works best with curated, human-readable source material. Keep exact operational state in .agenties/.
Limitations
| Limitation | Detail |
|---|
| Community CLI | The notebooklm CLI is not an official Google product. It may break when Google updates the NotebookLM web interface. |
| Interactive login | Login cannot be automated from Agenties. You must run the login command in a terminal. |
| No real-time sync | Sources are uploaded on wrapup, not live. The notebook reflects your last uploaded summary. |
| Not a database | NotebookLM is a semantic recall tool over uploaded documents — it is not a queryable operational store. |