Claude does the thinking.
NotebookLM does the remembering.

You give Claude a second brain by pairing it with a Google NotebookLM vault. Claude stays the reasoning engine, NotebookLM holds the project context, and every new session pulls what it needs instead of you re-briefing from scratch.

The architecture: reasoning versus memory

The setup splits one job across two tools. Claude is the reasoning engine: logic, problem-solving, code generation, complex writing. NotebookLM is the memory vault: past conversation notes, project specifications, reference documents. The link between them is the part that saves you anything: Claude pulls context from NotebookLM on demand, instead of you loading thousands of tokens into the prompt window every time you open a new chat.

Step 1: Set up your central NotebookLM vault

Go to Google NotebookLM and create a new notebook dedicated to the project, or to your overall personal memory: something you can name plainly, like Claude Project Memory. Add your base context sources to it: project SOPs, tech stack details, strategy documents, master guidelines. This notebook is what Claude reads from for the rest of the project's life, so it is worth filling properly once.

Step 2: Connect Claude to NotebookLM

Link Claude to the notebook through a Model Context Protocol (MCP) connector, and confirm the connection actually reads as established before you rely on it. Then set the standing rule in your system prompt: before starting a cold project context, query my NotebookLM vault to recall past decisions, code patterns, and project rules. That single line is what turns a connected vault into an automatic one, Claude checks its memory without being asked.

Step 3: Run the wrap-up at the end of every session

Memory only carries forward if something writes it down. At the end of a working session, ask Claude to summarise the session's key takeaways, API patterns, code decisions and unsolved tasks into a concise memory brief. Send that brief into your connected NotebookLM vault: pasted in, or synced automatically through the integration, then close the chat knowing the context is stored rather than lost with the tab.

Step 4: Resume without the re-briefing

Open a fresh Claude chat and ask a question about the ongoing work: how did we structure that auth flow? Claude queries NotebookLM, retrieves the specific memory it needs, and answers with full context behind it. You never re-explain the project. The tokens that used to go on setup go on the actual work instead, and the answers stop drifting from decisions you already made.

What the vault unlocks on top

Keeping project history in NotebookLM buys more than recall, because the same knowledge base drives its multi-modal tools. Infographics auto-design charts and diagrams from your notes. Cinematic video and audio overviews turn project research into storyboarded clips or spoken summaries. Deep research searches across hundreds of past notes and cites its answers back to the documents you put in, so the output stays anchored to your own material.

Drafted with AI under human editorial review. No third-party sources were used. · TALC AI