Polywise

Polywise: An Open-Source Alternative to NotebookLM

Productivity

Open Source

Artificial Intelligence

GitHub

Visit WebsiteSee on Product HuntGithub ⧉Twitter ⧉

The concept originated from an idea I had earlier this year: "Could the human brain's memory mechanisms be integrated with AI to endow an 'Agent' with true 'Agency'?" Specifically, could we simulate a neural network by treating the knowledge nodes and edges within a Graph RAG framework as neural nodes and synaptic connections? My goal was to provide Agents with an evolutionary environment that fosters "self-evolution." Concurrently, I observed that many people were using a combination of Obsidian and Claude for knowledge management. However, this approach requires extensive configuration and results in a rather disjointed user experience—it simply isn't "AI-native" enough. (Conversely, tools like NotebookLM and Ima, while functional, lean too heavily toward traditional paradigms and also lack that truly AI-native feel.) This sparked an idea: could I build a knowledge base application—usable directly on the client side—that possesses "self-learning" capabilities? Such an application could serve as a superior alternative to NotebookLM. (After all, storing work-related proprietary knowledge in NotebookLM or Ima often raises compliance and security concerns; conversely, storing it in Obsidian typically necessitates constant manual organization of directories and bidirectional links—and even with AI assistance, the process still falls short of being truly AI-native.) I spent a month conducting research and practical experiments using a "vibe coding" approach—exploring the boundaries of pure coding capabilities while simultaneously validating the technical feasibility of my implementation path. This process confirmed that the technical approach was sound. My timing also proved fortuitous, as Kapasi’s "LLM-Wiki" concept suddenly went viral, thereby validating the existence of a genuine user demand for such a solution. Thus, Polywise was born. It adopts the "Agent" as its fundamental unit—replacing the traditional "Notebook" concept found in tools like NotebookLM and Ima. Furthermore, it empowers each Agent to autonomously create its own skills and tools, and to engage in self-directed learning via Graph RAG, thereby enabling the Agent to evolve its own intelligence and develop true "agency."