Since: NeoMutt 2026-02-11
This feature is under active development. Its API and behaviour may change in future releases.
The Fuzzy Search feature lets you find items — such as mailboxes, aliases, or commands — by typing just a few characters from anywhere in the name. You don't need to remember the exact spelling or the full path. For example, typing “mlnd” is enough to match “mailinglists/neomutt-dev”.
The matcher works through the whole list of candidates in a single pass and ranks every match by quality. Matches at the start of a word, at a path separator, or in consecutive characters score higher than scattered matches, so the best result naturally rises to the top.
Searching is case-insensitive by default. When smart case is enabled, typing an all-lowercase pattern still matches regardless of case, but including even one uppercase letter makes the whole search case-sensitive. This means you can find “INBOX” with “inbox”, but typing “Inbox” will only match strings that contain a capital “I”.
Fuzzy search quality depends on ranking, not just yes/no matching. If multiple candidates match, the one with the best score is shown first. This usually favors short, compact, boundary-aligned matches.
The following NeoMutt features use Fuzzy Search internally:
Sidebar Search— dynamically filter the Sidebar mailbox list as you type
Developers can benchmark the matcher using
fuzzy/fuzzy-benchmark. See
fuzzy/BENCHMARK.md for usage details.
The library currently provides one algorithm:
subsequence matching
(
FUZZY_ALGO_SUBSEQ ), an FZF-style approach in which every character of the pattern must appear in the candidate string in order, but not necessarily consecutively. The architecture is pluggable, so additional algorithms (edit distance, token-based) can be added without changing the public API.
Every match is given a numeric score. A higher score means a better match. A score of
-1 means no match at all.
Table 18.2. Bonuses
| Condition | Points |
|---|---|
| Match at position 0 (start of string) | +30 |
Match after a separator (
/
.
-
_ ) |
+15 |
| Each consecutive matched character | +15 |
| CamelCase boundary (lowercase then uppercase ASCII) | +10 |
Match at position 0 with
prefer_prefix enabled |
+40 |
Table 18.3. Penalties
| Condition | Points |
|---|---|
| Gap between two matched characters | −2 per skipped character |
| Total match span (last position − first position + 1) | −1 per character |
| Candidate string length | −(length / 4) |
After all bonuses and penalties are applied, the final score is clamped to a minimum of
0 for any valid match.
By default, matching ignores case for ASCII letters (A–Z). When
smart_case is enabled in
FuzzyOptions:
An all-lowercase pattern → case-insensitive matching (ASCII)
A pattern containing any uppercase ASCII letter → case-sensitive matching
Non-ASCII characters (e.g. accented letters, CJK) are always matched as-is, byte-for-byte, since only ASCII A–Z are case-folded.
Matching is hybrid:
ASCII pattern characters are matched with ASCII case folding (A–Z only).
Non-ASCII pattern characters are matched as complete UTF-8 codepoints (their byte sequences must stay contiguous).
This avoids false positives where bytes from one UTF-8 character are accidentally matched across different characters. Boundary bonuses (separator and CamelCase) are still ASCII-only.
The algorithm is designed for interactive use with thousands of candidates:
O(n) time — single forward pass through the candidate
O(m) stack space — fixed 256-byte array for match positions; no heap allocation
No backtracking, no recursion
Throughput depends on CPU, compiler flags, input shape and iteration count. Use the bundled benchmark tool to measure your environment rather than relying on a fixed ops/sec number.
