Linkit: Open-Source AI Bookmark Manager With Semantic Search and Knowledge Graph Hits Hacker News

A bookmark manager that treats links as knowledge

Linkit, a free and open-source desktop bookmark manager, reached Hacker News's front page on October 21 as a Show HN post. According to the project's GitHub README, it is aimed at knowledge workers who want saved links to work as retrievable assets rather than a flat list of URLs.

The app pairs a conventional bookmark manager — categories, tags, notes, star ratings and reading status — with a set of AI-driven features: automatic page summarization, semantic search across the library, AI-curated collections built from natural language prompts, and a knowledge graph that visualizes bookmarks as a network connected by shared tags, collections and semantic similarity.

AI features, but you bring the model

Notably, Linkit does not ship with its own LLM service or API keys. The README states that users must connect their own model: any OpenAI- or DeepSeek-compatible API, or a local LLM such as Ollama, LocalAI or vLLM. Once configured, the model powers summaries and key takeaways, tag and category recommendations, semantic vector search, and duplicate detection that flags not only exact-URL matches but also near-duplicates pointing at the same content under different URLs.

API keys are stored in the operating system's credential store — the macOS Keychain or Windows Credential Manager — rather than in the app's database, and are never uploaded, according to the project.

Local-first by default, Supabase sync as an option

Linkit stores everything in a local SQLite database by default, with no account required and no telemetry, the README says. Optional cloud sync runs through Supabase, with PostgreSQL row-level security binding each record to the signed-in user's ID. The project claims this means even its own developers cannot access synced bookmarks, and that revision tracking with automatic merging handles conflicts between devices.

Import paths cover browser HTML exports from Chrome, Firefox, Safari and Edge, Raindrop.io CSV/JSON files, and clipboard URLs. A Spotlight-style search bar triggered with Cmd/Ctrl+K doubles as a quick-capture field for pasting and saving a link without leaving the current tab.

Housekeeping features include background link health checks that surface 404s and silently changed pages, a periodic insights report summarizing curation habits and reading progress, and AI-curated collections whose suggestions remain editable previews until the user explicitly confirms them.

Built with Wails, available on all three desktop platforms

The technology stack is Wails (Go) for the desktop shell, React with Vite and TypeScript on the frontend, Tailwind CSS for styling, and GORM over SQLite for local storage. The app runs on macOS, Windows and Linux, with macOS users able to install via Homebrew. It ships with twelve themes and localization in English and Simplified Chinese.

One caveat worth flagging: the README acknowledges that the macOS build is not notarized by Apple, and that the Homebrew cask clears the Gatekeeper quarantine attribute automatically on install — a workaround that users with strict security policies may want to weigh before installing.

The project positions itself explicitly against paid bookmarking services, arguing that sync, full-text search and backups should not sit behind monthly subscriptions. Everything, including cross-device sync, is free with no paywalls, according to the README.

Why it matters

Bookmark managers are a crowded category, and the pile of saved-but-never-read links is a familiar problem. Linkit's angle is combining three currents in one package: local-first data ownership, user-supplied AI that can run entirely on a local model, and open-source licensing with no subscription tier. For readers wary of sending their browsing archive to a third-party AI service, the ability to run summarization and semantic search against a local model over a local database is the distinguishing proposition — the AI augments data the user keeps, rather than being the reason to hand it over. The unnotarized macOS build and the bring-your-own-model requirement are rough edges, but the overall shape — a free tool that turns a passive link pile into something searchable and connected — is a sensible answer to how much reading now happens across dozens of tabs and devices.