We want to discover & surface connections between ideas, documents and concepts. This ability is something other tools have specialized in but each within their own silos. By creating an index of that knowledge for a user, we can give them the superpower to discover connections among their own work where ever they go.
Warning: This is an experimental piece of research software presented for transparency, and is incomplete, unsupported, and likely to stop working.
This repository is reference implementation in Rust. It exposes GraphQL API so that other tools can interface to submit / discover connections between resources identified by URLs (Local resources are represented via file:/// URLs).
At the moment no binaries are distributed, however you can build / run using cargo. To create a (debug) build run:
cargo +nightly buildBy default, the build will produce a binary at ./target/debug/knowledge-server.
You can start a knowledge-server by running:
./target/debug/knowledge-server serveOnce it's running you can explore protocol schema, execute queries / mutations using GraphQL IDE at http://localhost:8080/graphiql.
(This won't be very interesting at first, as you won't have submitted any data to it.)
You can spawn a knowledge-server as a daemon by runing:
./target/debug/knowledge-server daemonYou can ingest local markdown files into your knowledge base by running:
./target/debug/knowledge-server scan $YOUR_PATH_HEREBrowsing history can be sensitive. That's one of the reasons KSP keeps all its data on your local machine. We haven't built any tools to remove data from KSP, but it stores it data in ~/.knowledge-service/. Most of the data is stored in a .sqlite file, which you can inspect and modify at your own peril, but the full-text TF-IDF data lives in a special format used by the Tantivy library.
We ran into issue with rls-vscode extension. If you use vscode you may want to consider rust analyzer instead.