Searching Logseq by Concept, Not Keystrokes

Logseq is great for dumping daily notes, but finding them again later can be a pain. If you’re looking for notes on a “connection timeout” but originally wrote “increasing the socket keepalive”, a standard keyword search will give you nothing. You end up having to guess the exact phrasing your past self used.

I wanted a way to search my graph by concept rather than exact text matches. That’s why I put together the Logseq Semantic Search plugin.

The upcoming database version of Logseq actually has semantic search built-in. But since I’m still using the standard Markdown version for my day-to-day workflow, I wanted to get that capability right now.

Indexing the hierarchy

The plugin uses text embeddings to find conceptually similar blocks. But just embedding individual bullet points doesn’t work well for outliners. A block that just says “needs refactoring” is useless on its own.

If you’ve seen my Logsqueak project, you’ll recognise the indexing approach here. Every block is indexed along with its complete structural lineage—the page name, properties, and the full chain of parent blocks above it.

Because it captures this nested context, the search index knows that a vague bullet point nested under billing-service → Database Connection Pool is actually about your Postgres setup. Searching for “optimizing billing db” will pull that specific child block right to the top of the results.

Running it locally

Since a Logseq graph is essentially a private brain dump, I wanted this to run entirely locally. By default, the plugin connects to Ollama using the lightweight nomic-embed-text model. It’s smart enough to only re-embed blocks that have changed, so it’s relatively fast even without a GPU. (If you prefer, you can also point it at any OpenAI-compatible endpoint in the settings).

I run Fedora Workstation and prefer to keep my host system clean, so I run Ollama via Podman. It’s incredibly straightforward to set up:

# Start the Ollama container, exposing the default port and persisting data
podman run -d -v ollama:/root/.ollama -p 11434:11434 --name ollama docker.io/ollama/ollama

# Pull the lightweight embedding model
podman exec -it ollama ollama pull nomic-embed-text

Because we mapped port 11434, the Semantic Search plugin can talk to the container seamlessly at http://localhost:11434 right out of the box. No dependency issues, just a private embedding server ready to run in the background.

Usage

You can grab the plugin directly from the Logseq Marketplace. Once it’s installed, hit Alt+K (or click the toolbar icon) to open the search modal. Try typing a natural language query—like “notes about debugging pipeline failures”—and it will surface the relevant blocks even if you didn’t use the word “debugging.”

The source code is up on GitHub if you want to poke around or contribute.