Get started with Lore

Prepare a local Markdown index, search documentation in plain language, and follow the results to their source sections.

Install Lore with Cargo

Lore finds relevant passages in a project’s Markdown documentation. Use a question to find a starting point, then read the source with its headings and context.

Install the published lore-docs package from crates.io with Cargo, then prepare an index for the project you want to search.

Prepare your tools

You need Rust 1.88 or newer, Cargo, and a C/C++ toolchain. Linux x86-64 is the tested platform. Installation builds from source and needs access to Rust dependencies. The search model is a separate initial download of about 67 MB; indexing and searches work offline after setup.

Cargo installs the lore executable into $CARGO_HOME/bin, usually ~/.cargo/bin. Make sure that directory is on your PATH.

Install Lore

Run this from any directory:

cargo install lore-docs --locked
lore --help

The package name is lore-docs; the command you run is lore. Cargo downloads and builds the published package, then installs the executable.

Index your documentation

Change to the project whose documentation you want to search:

cd /path/to/project
lore model install
lore index .

Replace /path/to/project with your own directory. Lore stores the index and model under .lore/ in that directory. Each index belongs to one project. Keep running the following commands there, or pass --index-dir /absolute/path/to/project/.lore to select it from elsewhere.

Lore reads UTF-8 .md and .markdown files up to 2 MiB each. It respects .gitignore and .loreignore, skips build and cache directories, and leaves documents unchanged. Add .lore/ to your project’s .gitignore if needed.

Search and follow the source

lore brief "How do I configure this project?"
lore brief "Which environment variables are required?" --with-guidance --json
lore brief "APP_PORT" --mode keyword --scope docs --json

Results include source passages, file paths, and section references. --with-guidance adds suggested follow-up reads, --scope narrows the search, and --json returns structured output. Adapt the questions and directory to your project.

Use a returned file path to inspect its headings and text:

lore outline README.md --json
lore read README.md --json

To read a particular section, use an anchor from outline, such as lore read 'docs/configuration.md#settings' --context --json. Replace that example with a path and anchor from your own results. --context includes introductory text from enclosing headings. Long reads are paginated; when has_more is true, pass the returned next_cursor with --cursor to continue.

Connect a coding agent

After preparing the index, configure an MCP client to launch Lore over stdio:

lore --index-dir /absolute/path/to/project/.lore mcp

Use the absolute executable path in your client’s configuration if it cannot find lore on its PATH. The client starts the process and communicates through stdin and stdout. This exposes one prepared project; it does not install models or refresh the index.

Ask the agent to check lore_status, then search with lore_brief and follow the returned read targets.

Refresh after edits

From the same project directory:

lore status --check --json
lore index .

There is no automatic file watcher. Refresh after changing documents or ignore rules. Running MCP processes see the refreshed index on their next call. Search rankings help you decide what to read; inspect the source to decide whether it answers your question.