Shelfmark: A private semantic library for the web

This analysis is based entirely on the self-reported description provided by the author. No independent verification, revenue data, customer names, or third-party sources are available. All claims are treated as stated by the author and not proven.

The description states that Shelfmark is a native macOS and iOS application designed to build a private, searchable library of web resources. It allows users to save links with only a URL, then fetches page information within size limits, extracts useful text, and generates local summaries and topic tags.

Key technical components include:

• Swift, SwiftUI, SwiftData
• Apple’s Natural Language framework for semantic search
• Private CloudKit database
• Local vector snapshots for semantic indexing

The application stores data locally and does not send saved content to remote providers in v1. It supports both exact and related (semantic) search.

Inference: The product appears to be a personal knowledge management tool focused on organizing web content through local processing and semantic understanding.

The author positions Shelfmark as a solution to the problem of saving information without being able to find it later. It aims to turn saved links into organized, searchable knowledge by understanding meaning rather than just filenames or URLs.

Claims made:

• The app remembers meaning instead of only filenames and URLs.
• Semantic search is useful when the library understands what was saved.
• Saved resources can be explored as lists or through shelves built from topics, sources, and relationships.
• Exact search finds known words; related search uses Apple’s local Natural Language sentence embeddings.

Inference: The positioning has evolved from a personal hackathon project into a focused tool for private semantic knowledge organization. The evolution reflects an emphasis on usability and reliability over feature bloat.

The description does not name specific customers or personas. However, the author describes the need as one of constant information gathering — such as notes, bookmarks, “read it later” lists, and browser tabs — where finding previously saved items becomes difficult.

It is implied that the target user is someone who:

• Saves many web resources regularly
• Values privacy and local processing
• Needs to retrieve content based on meaning rather than title or URL

Inference: The ICP likely includes individuals who manage large volumes of digital information and value control over their data, particularly those using macOS and iOS devices.

There is no evidence of pricing, monetization strategy, or business model in the description. The author emphasizes that v1 does not require an AI account or remote LLM integration and keeps all derived indexes local.

Not evidenced: No indication of revenue streams, subscriptions, or paid features.

The project is built using:

• Swift, SwiftUI, SwiftData
• Apple’s Natural Language framework
• Private CloudKit for synchronization
• Spec Kit for feature definition
• Codex and GPT-5.6 for implementation support

Key technical decisions include:

• Rejection of Core Spotlight in favor of Apple’s NLP embeddings
• Fixed English policy for v1
• Local-only semantic search
• Bounded enrichment and recovery mechanisms
• Separation between manual and generated tags

Inference: The team prioritized reliability, privacy, and deterministic workflows over scalability or external integrations.

The description indicates that the project was developed during OpenAI Build Week and is not yet released broadly. It mentions:

• Successful builds on macOS and iOS
• Focused deterministic coverage for strong workflows
• Recovery of previously stuck live resources with real summaries and tags

However, there is no evidence of:

• User adoption or feedback
• Customer base or usage metrics
• Public release or distribution channels

Not evidenced: No signs of traction or market validation beyond the author’s own use.

The description does not mention competitors. However, it implies a niche in private, semantic knowledge management tools for macOS and iOS users.

Inference: Shelfmark competes with other personal knowledge tools (e.g., Notion, Obsidian) but focuses on web content capture and local processing, distinguishing itself through privacy and semantic search capabilities.

• No commercial traction or user base: The project is described as a personal tool without evidence of adoption.
• Single-person team: The entire development effort is attributed to one individual (David Navalho).
• Limited scope: V1 excludes remote LLM integration and multilingual support, which may limit future growth.
• Unproven market demand: No external validation or user testing beyond the author’s own experience.

Inference: Without evidence of users or revenue, the risk of failure in a commercial context is high.

1. What specific problems do you observe in how people currently organize and retrieve web-based information?
2. Have you tested Shelfmark with others outside your own workflow? If so, what feedback did you receive?
3. How do you plan to scale beyond a single developer’s effort?
4. Are there any plans for monetization or broader distribution beyond personal use?
5. What are the key assumptions behind the current architecture and feature set?