diff --git a/src/lib.rs b/src/lib.rs
index 6e3df9d..416cf6d 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -1,4 +1,52 @@
-//! TODO: docs
+//! This crate provides a collection of different distance metrics on strings.
+//!
+//! This problem is sometimes referred to as "string similarity search", or
+//! more colloquially "fuzzy matching". Given a query string and a number of
+//! possible candidate strings, the goal is to:
+//!
+//! a) filter out the candidates that are too dissimilar from the query;
+//!
+//! b) rank the remaining candidates by their similarity to the query.
+//!
+//! Here both of these tasks are accomplished by implementing the [`Metric`]
+//! trait. This trait is at the basis of norm's design, and it is implemented
+//! by all of our metrics. Reading its documentation is a good place to start.
+//!
+//! # Performance
+//!
+//! Performance is a top priority for this crate. Our goal is to have the
+//! fastest implementation of every metric algorithm we provide, across all
+//! languages. [Here][bench] you can find a number of benchmarks comparing
+//! norm's metrics to each other, as well as to other popular libraries.
+//!
+//! # Examples
+//!
+//! ```rust
+//! use norm::fzf::{FzfParser, FzfV2};
+//! use norm::Metric;
+//!
+//! let mut fzf = FzfV2::new();
+//!
+//! let mut parser = FzfParser::new();
+//!
+//! let query = parser.parse("aa");
+//!
+//! let cities = ["Geneva", "Ulaanbaatar", "New York City", "Adelaide"];
+//!
+//! let mut results = cities
+//!     .iter()
+//!     .copied()
+//!     .filter_map(|city| {
+//!         fzf.distance(query, city).map(|mach| (city, mach.distance()))
+//!     })
+//!     .collect::<Vec<_>>();
+//!
+//! results.sort_by_key(|(_city, dist)| *dist);
+//!
+//! assert_eq!(results.len(), 2);
+//! assert_eq!(results[0].0, "Adelaide");
+//! assert_eq!(results[1].0, "Ulaanbaatar");
+//! ```
 
 #![cfg_attr(docsrs, feature(doc_cfg))]
 #![allow(clippy::needless_range_loop)]