Tagging and linking with AI (Napkin.one)

Napkin turns the most tedious part of personal knowledge management—manual tagging—into an AI-assisted workflow that automatically assigns tags and then uses those tags to resurface related ideas. The payoff is practical: instead of spending time labeling every note, users can dump thoughts (or Readwise highlights) into Napkin and let a GPT-3–based system generate “magic tags,” then guide review sessions and idea discovery through similarity.

Napkin is built around three functions. First, it collects thoughts sent by the user. Second, it uses AI (specifically a fork of GPT-3) to automatically tag each thought. Third, it helps users create by resurfacing similar notes based on those tags. In practice, the app’s strongest entry point is its integration with Readwise: users can import Readwise highlights into Napkin via an “import Readwise highlights” button, with syncing happening incrementally for performance. That means Napkin becomes a staging layer for the flood of unprocessed highlights many people already store, rather than a place where every item must be manually curated.

Tagging and review happen through a “daily mix” flow. Napkin prioritizes older thoughts first, shows existing AI-generated tags alongside user-added ones, and lets users quickly refine keywords with keyboard shortcuts. Users can add or remove tags on the fly—for example, swapping a generic tag for a more precise one like “self-organization”—and can archive items that no longer fit. The interface also supports exploration: clicking tags filters what’s shown, while related items appear on the left and right, connected by shared tags (such as “relationship”). As users add more manual tags and feed more highlights into Napkin, the system’s tag suggestions and similarity surfacing improve.

For creation, Napkin supports an ideation-to-drafting workflow using “stacks,” a pile-like structure for organizing related thoughts into sections. A talk-planning example shows how a user can search by tag (e.g., “note taking”), pull relevant thoughts into a stack, arrange them under headings like “problem” and “PKM,” and then copy the assembled stack into Obsidian. The copied content includes the stack name, section structure, thought text, source citations, tags, and a pointer back to the Readwise item.

Data portability is a key reassurance. Napkin can export connections and tags as JSON or CSV, including fields like source, author, and tags—so users aren’t locked into the app. The main limitation is the lack of a native Obsidian integration for now; Napkin offers an API that supports one-way pushing into Napkin, while a two-way tagging workflow would require careful design. Even without that, Napkin is positioned as an “initial ideation exploration layer” that raises the bar for what gets turned into Obsidian notes, because resurfacing in Napkin can reveal ideas that users might not have searched for—or had the energy to process—at the time they were first captured.

The result is a workflow that makes tagging feel less like bookkeeping and more like interactive discovery, while still feeding structured outputs into Obsidian when it’s time to write.