NoteCast turns messy local notes into an LLM-built knowledge graph
NoteCast is an early-stage local note engine that uses LLM summaries, embeddings, theme classification, and consolidation to organize notes into a knowledge graph.
NoteCast is a local note engine that uses LLMs and embeddings to turn raw notes into an evolving knowledge graph. The GitHub repository describes a pipeline that summarizes notes, extracts keywords, embeds content, classifies notes into themes, splits overloaded themes, and consolidates cross-theme relationships. The Show HN launch confirms the same positioning and clearly labels the project as early-stage but already usable for feedback.
Key takeaways
- NoteCast is built for people who capture notes but do not want to manually categorize or maintain them.
- The pipeline has four note states: pending, processed, scanned, and organized.
- It supports multiple LLM providers, including OpenAI, Anthropic, Gemini, DeepSeek, Codex, and Ollama, but the user must explicitly choose the default provider.
- The repository describes Obsidian vault output, a CLI-first workflow, and an optional REST API server on port 3000.
- The author warns that the project is early-stage and may still contain bugs or architectural changes.
Practical LinkLoot angle
The useful idea is not simply "AI for notes." NoteCast separates note ingestion from organization proposals, so a user can capture first and review structure later. That makes it more practical than asking a chatbot to reorganize an entire vault in one prompt, especially when notes cross domains and need multi-parent relationships.
| Tool or workflow | Best use | Limitation to check | Source |
|---|---|---|---|
| NoteCast | Local note organization with summaries, embeddings, and evolving themes | Early-stage; provider setup and restart behavior need testing | GitHub repository |
| Obsidian manual folders | Durable personal knowledge management | Manual taxonomy work grows over time | Practical comparison |
| Chat-only note cleanup | Fast one-off summaries | Weak persistent graph and repeatable structure | Practical comparison |
A good small test is to create three base themes such as Work, Personal, and Research, import 20 mixed notes, and inspect whether the proposed subtopics match your actual recall needs. If the graph becomes easier to browse without hiding the original notes, the workflow is worth deeper testing.
What to verify before you act
Check where credentials are stored before connecting paid LLM providers, especially on shared machines or CI servers. Verify whether Ollama quality is acceptable if you want a more private local setup, because the repository says OpenAI is currently the most mature and best-tested provider. Also test Obsidian output on a copy of your vault first, not against your main notes.
It processes local notes with LLM summaries, keyword extraction, embeddings, and theme organization to build a knowledge graph.
For adjacent automation patterns, see LinkLoot's guide to AI workflow automation.