Notes that live where you want them, and that your AI can read

Each Korely note is a single .md file inside a folder you choose on startup. Open the folder in Finder or Explorer and you see your notes the same way you see anything else on disk. Point Obsidian, Logseq, iA Writer, Cursor, Claude Code, or VS Code at the same folder and they all just work. Apple Notes users can import and export the same Markdown back and forth. If you ever leave Korely, every byte of your knowledge comes with you. No export wizard. No migration tool. Nothing locks you in.

The Markdown follows Obsidian's conventions: wikilinks [[like this]], tags like #research, and inline images stored next to the note in an attachments/ folder. If you already keep a vault elsewhere, the importer ships with the app.

Pure keyword search misses anything you phrased differently three months ago. Pure vector search returns matches that sound right but aren't, and ranks ties poorly. Korely runs both and merges the rankings, so you find notes by the words you typed and by what you actually meant.

The keyword side runs on FTS5, the open source full text search engine built into SQLite. It indexes every word in every note and ranks results in milliseconds, with prefix matching, so typing "graph" already surfaces "GraphRAG" or "graph database" the moment you press space. The semantic side runs a local embedding model on your CPU. No API calls. No cost per token. Both indexes update as you type. Both stay on your machine unless you turn cloud sync on. Your data, your call.

The result: ask "what did we decide about the pricing rework last quarter?" and Korely finds the offsite meeting note where you wrote "team agreed to lower the entry tier by 20%". Words and meaning aligned.

Obsidian made backlinks famous, but they only work if you remember to type [[wikilinks]] while writing. Most people stop after a week. Korely doesn't need you to remember.

As you write, Korely reads each note with a local embedding model and pulls out the entities that come up across your vault: people, projects, places, concepts, and anything else that recurs. Notes that share entities become related on their own. Open a note about a customer call and the sidebar shows last year's discovery notes, the proposal draft, and the contract review you almost forgot. No tagging discipline required.

This is what makes GraphRAG sharper than plain RAG. When you ask Korely a question, or any AI tool plugged into your vault, the retrieval doesn't just look at keyword matches. It walks the graph, follows the entities, and pulls in everything related. The answer comes back with the full picture, not just the closest paragraph. More on the knowledge graph →

Korely doesn't force one filing system. Folders mirror the directory on disk, so creating them in the app or in Finder both work. Tags live inline as #research or #meetings and surface in the sidebar.

If you want structure, use folders. If you want fluidity, use tags. If you want neither, the knowledge graph still wires things together for you. Pick two, three, or none. Search and graph work the same either way.

Korely ships two ways to give your AI access to the vault. The first is a Model Context Protocol (MCP) server built into the desktop app. Add it as a source in Claude Desktop, Cursor, Zed, or any MCP aware client and the assistant can search, read, and traverse your notes directly. No file upload. No clipboard dance. No "let me paste the relevant context" workaround.

The second is the Korely CLI. Pipe queries from a terminal, a shell script, an editor command, or your own agent framework. Same five read tools as MCP: korely_search, korely_read_item, korely_get_related, korely_list_notes, and korely_list_folders. Same Markdown output. No GUI required.

Why use the CLI when you have MCP? Token economics. With MCP, your AI agent decides what to fetch and often makes five to ten exploratory tool calls per question, each one growing the context window. With the CLI, your script pulls exactly what you need and hands a compact result to the model. Same vault, same tools, a fraction of the spend. Pick MCP when you want the agent to drive. Pick the CLI when you want to drive.

Pro users also get write tools (save notes, create tasks) and a cloud MCP endpoint, so a vault that lives on your laptop can still answer questions from your phone, your CI runner, or n8n. See the full setup →