NoteCast: A Local Engine That Transforms Notes Into Knowledge Graphs with LLMs

Taking notes is easy. Finding them useful three weeks later is another matter. The problem isn't capture but deferred organization: notes accumulate without structure until their implicit value simply evaporates. That exact problem is what led developer AlexWasHeree to build NoteCast, a local note engine he unveiled this week on Hacker News.

The project is early stage, as the author himself makes clear, but the core works and already proves practical from his usage experience.

How the Pipeline Works

NoteCast applies a three-stage pipeline to notes as users add them:

1. Classify: each note is categorized thematically.
2. Organize: classified notes are placed within a hierarchy of topics that emerges dynamically as volume grows. It is not a predetermined taxonomy: topics subdivide through accumulation.
3. Consolidate: the system integrates new information with what already exists in the graph, detecting relationships and avoiding duplicates.

Any change the pipeline proposes is not silently applied: it generates a proposal that users can review, edit, and explicitly confirm before incorporation into the graph. This human validation step is one of the project's most sensible design decisions, precisely because it sidesteps the usual problem of automated systems reorganizing without permission.

Obsidian Integration

NoteCast includes integration with Obsidian vaults. Simply point the configuration to the vault path (`vaultPath`) and the engine begins processing existing notes. Since Obsidian stores everything in plain Markdown files, integration is straightforward and requires no migrations or special exports.

This makes it an interesting option for those who already have hundreds or thousands of notes in Obsidian but have given up maintaining structure manually. The graph does not replace the vault: it complements it by adding a navigable semantic layer.

Why This Approach Makes Sense Now

The idea of building personal knowledge graphs is not new, PKM (Personal Knowledge Management) has been an active niche for years, but the availability of LLMs that can run locally changes the equation for cost and privacy. A pipeline like NoteCast's would have required calls to external APIs with associated token costs and exposure of potentially private notes. Run locally, no notes leave the device.

The author does not specify in the README which model or LLM runtime is used internally, something that should be clarified in upcoming versions so users know what hardware requirements they need.

Who Finds It Useful Today

In its current state, NoteCast is mainly useful for technical users who:

• Have an active, chaotic Obsidian vault.
• Feel comfortable cloning a repository and tweaking configuration by hand.
• Want a local, private solution with no cloud service dependency.

It is not yet a tool for non-technical users: documentation is sparse and the author warns that significant architectural decisions may change during 2026. Anyone coming in now does so as an early adopter willing to encounter rough edges.

Our Take

NoteCast solves a real problem with a technically coherent approach. The proposal-review-confirmation cycle is the project's most mature design decision and deserves to be maintained as the tool evolves. It is worth keeping an eye on.