diff --git a/kitty/search_query_parser.py b/kitty/search_query_parser.py
new file mode 100644
index 000000000..0f64d7a9f
--- /dev/null
+++ b/kitty/search_query_parser.py
@@ -0,0 +1,275 @@
+#!/usr/bin/env python
+# License: GPLv3 Copyright: 2022, Kovid Goyal <kovid at kovidgoyal.net>
+import re
+from enum import Enum
+from functools import lru_cache
+from gettext import gettext as _
+from typing import (
+    Callable, Iterator, List, NamedTuple, Optional, Sequence, Set, Tuple,
+    TypeVar, Union
+)
+
+
+class ParseException(Exception):
+
+    @property
+    def msg(self) -> str:
+        if len(self.args) > 0:
+            return str(self.args[0])
+        return ""
+
+
+class ExpressionType(Enum):
+    OR = 1
+    AND = 2
+    NOT = 3
+    TOKEN = 4
+
+
+class TokenType(Enum):
+    OPCODE = 1
+    WORD = 2
+    QUOTED_WORD = 3
+    EOF = 4
+
+
+T = TypeVar('T')
+GetMatches = Callable[[str, str, Set[T]], Set[T]]
+
+
+class SearchTreeNode:
+    type = ExpressionType.OR
+
+    def __init__(self, type: ExpressionType) -> None:
+        self.type = type
+
+    def search(self, universal_set: Set[T], get_matches: GetMatches[T]) -> Set[T]:
+        return self(universal_set, get_matches)
+
+    def __call__(self, candidates: Set[T], get_matches: GetMatches[T]) -> Set[T]:
+        ...
+
+    def iter_token_nodes(self) -> Iterator['TokenNode']:
+        ...
+
+
+class OrNode(SearchTreeNode):
+
+    def __init__(self, lhs: SearchTreeNode, rhs: SearchTreeNode) -> None:
+        self.lhs = lhs
+        self.rhs = rhs
+
+    def __call__(self, candidates: Set[T], get_matches: GetMatches[T]) -> Set[T]:
+        lhs = self.lhs(candidates, get_matches)
+        return lhs.union(self.rhs(candidates.difference(lhs), get_matches))
+
+    def iter_token_nodes(self) -> Iterator['TokenNode']:
+        yield from self.lhs.iter_token_nodes()
+        yield from self.rhs.iter_token_nodes()
+
+
+class AndNode(SearchTreeNode):
+    type = ExpressionType.AND
+
+    def __init__(self, lhs: SearchTreeNode, rhs: SearchTreeNode) -> None:
+        self.lhs = lhs
+        self.rhs = rhs
+
+    def __call__(self, candidates: Set[T], get_matches: GetMatches[T]) -> Set[T]:
+        lhs = self.lhs(candidates, get_matches)
+        return self.rhs(lhs, get_matches)
+
+    def iter_token_nodes(self) -> Iterator['TokenNode']:
+        yield from self.lhs.iter_token_nodes()
+        yield from self.rhs.iter_token_nodes()
+
+
+class NotNode(SearchTreeNode):
+    type = ExpressionType.NOT
+
+    def __init__(self, rhs: SearchTreeNode) -> None:
+        self.rhs = rhs
+
+    def __call__(self, candidates: Set[T], get_matches: GetMatches[T]) -> Set[T]:
+        return candidates.difference(self.rhs(candidates, get_matches))
+
+    def iter_token_nodes(self) -> Iterator['TokenNode']:
+        yield from self.rhs.iter_token_nodes()
+
+
+class TokenNode(SearchTreeNode):
+    type = ExpressionType.TOKEN
+
+    def __init__(self, location: str, query: str) -> None:
+        self.location = location
+        self.query = query
+
+    def __call__(self, candidates: Set[T], get_matches: GetMatches[T]) -> Set[T]:
+        return get_matches(self.location, self.query, candidates)
+
+    def iter_token_nodes(self) -> Iterator['TokenNode']:
+        yield self
+
+
+class Token(NamedTuple):
+    type: TokenType
+    val: str
+
+
+lex_scanner = getattr(re, 'Scanner')([
+        (r'[()]', lambda x, t: Token(TokenType.OPCODE, t)),
+        (r'@.+?:[^")\s]+', lambda x, t: Token(TokenType.WORD, str(t))),
+        (r'[^"()\s]+', lambda x, t: Token(TokenType.WORD, str(t))),
+        (r'".*?((?<!\\)")', lambda x, t: Token(TokenType.QUOTED_WORD, t[1:-1])),
+        (r'\s+',              None)
+], flags=re.DOTALL)
+REPLACEMENTS = tuple(('\\' + x, chr(i + 1)) for i, x in enumerate('\\"()'))
+
+
+class Parser:
+
+    def __init__(self, allow_no_location: bool = False) -> None:
+        self.current_token = 0
+        self.tokens: List[Token] = []
+        self.allow_no_location = allow_no_location
+
+    def token(self, advance: bool = False) -> Optional[str]:
+        if self.is_eof():
+            return None
+        res = self.tokens[self.current_token].val
+        if advance:
+            self.current_token += 1
+        return res
+
+    def lcase_token(self, advance: bool = False) -> Optional[str]:
+        if self.is_eof():
+            return None
+        res = self.tokens[self.current_token].val
+        if advance:
+            self.current_token += 1
+        return res.lower()
+
+    def token_type(self) -> TokenType:
+        if self.is_eof():
+            return TokenType.EOF
+        return self.tokens[self.current_token].type
+
+    def is_eof(self) -> bool:
+        return self.current_token >= len(self.tokens)
+
+    def advance(self) -> None:
+        self.current_token += 1
+
+    def tokenize(self, expr: str) -> List[Token]:
+        # Strip out escaped backslashes, quotes and parens so that the
+        # lex scanner doesn't get confused. We put them back later.
+        for k, v in REPLACEMENTS:
+            expr = expr.replace(k, v)
+        tokens = lex_scanner.scan(expr)[0]
+
+        def unescape(x: str) -> str:
+            for k, v in REPLACEMENTS:
+                x = x.replace(v, k[1:])
+            return x
+
+        return [
+            Token(tt, unescape(tv) if tt in (TokenType.WORD, TokenType.QUOTED_WORD) else tv)
+            for tt, tv in tokens
+        ]
+
+    def parse(self, expr: str, locations: Sequence[str]) -> SearchTreeNode:
+        self.locations = locations
+        self.tokens = self.tokenize(expr)
+        self.current_token = 0
+        prog = self.or_expression()
+        if not self.is_eof():
+            raise ParseException(_('Extra characters at end of search'))
+        return prog
+
+    def or_expression(self) -> SearchTreeNode:
+        lhs = self.and_expression()
+        if self.lcase_token() == 'or':
+            self.advance()
+            return OrNode(lhs, self.or_expression())
+        return lhs
+
+    def and_expression(self) -> SearchTreeNode:
+        lhs = self.not_expression()
+        if self.lcase_token() == 'and':
+            self.advance()
+            return AndNode(lhs, self.and_expression())
+
+        # Account for the optional 'and'
+        if ((self.token_type() in (TokenType.WORD, TokenType.QUOTED_WORD) or self.token() == '(') and self.lcase_token() != 'or'):
+            return AndNode(lhs, self.and_expression())
+        return lhs
+
+    def not_expression(self) -> SearchTreeNode:
+        if self.lcase_token() == 'not':
+            self.advance()
+            return NotNode(self.not_expression())
+        return self.location_expression()
+
+    def location_expression(self) -> SearchTreeNode:
+        if self.token_type() == TokenType.OPCODE and self.token() == '(':
+            self.advance()
+            res = self.or_expression()
+            if self.token_type() != TokenType.OPCODE or self.token(advance=True) != ')':
+                raise ParseException(_('missing )'))
+            return res
+        if self.token_type() not in (TokenType.WORD, TokenType.QUOTED_WORD):
+            raise ParseException(_('Invalid syntax. Expected a lookup name or a word'))
+
+        return self.base_token()
+
+    def base_token(self) -> SearchTreeNode:
+        tt = self.token(advance=True)
+        assert tt is not None
+        if self.token_type() is TokenType.QUOTED_WORD:
+            if self.allow_no_location:
+                return TokenNode('all', tt)
+            raise ParseException(f'No location specified before {tt}')
+
+        words = tt.split(':')
+        # The complexity here comes from having colon-separated search
+        # values. That forces us to check that the first "word" in a colon-
+        # separated group is a valid location. If not, then the token must
+        # be reconstructed. We also have the problem that locations can be
+        # followed by quoted strings that appear as the next token. and that
+        # tokens can be a sequence of colons.
+
+        # We have a location if there is more than one word and the first
+        # word is in locations. This check could produce a "wrong" answer if
+        # the search string is something like 'author: "foo"' because it
+        # will be interpreted as 'author:"foo"'. I am choosing to accept the
+        # possible error. The expression should be written '"author:" foo'
+        if len(words) > 1 and words[0].lower() in self.locations:
+            loc = words[0].lower()
+            words = words[1:]
+            if len(words) == 1 and self.token_type() == TokenType.QUOTED_WORD:
+                tt = self.token(advance=True)
+                assert tt is not None
+                return TokenNode(loc, tt)
+            return TokenNode(loc.lower(), ':'.join(words))
+
+        if self.allow_no_location:
+            return TokenNode('all', ':'.join(words))
+        raise ParseException(f'No location specified before {tt}')
+
+
+@lru_cache(maxsize=64)
+def build_tree(query: str, locations: Union[str, Tuple[str, ...]], allow_no_location: bool = False) -> SearchTreeNode:
+    if isinstance(locations, str):
+        locations = tuple(locations.split())
+    p = Parser(allow_no_location)
+    try:
+        return p.parse(query, locations)
+    except RuntimeError as e:
+        raise ParseException(f'Failed to parse {query!r}, too much recursion required') from e
+
+
+def search(
+    query: str, locations: Union[str, Tuple[str, ...]], universal_set: Set[T], get_matches: GetMatches[T],
+    allow_no_location: bool = False,
+) -> Set[T]:
+    return build_tree(query, locations, allow_no_location).search(universal_set, get_matches)
diff --git a/kitty_tests/search_query_parser.py b/kitty_tests/search_query_parser.py
new file mode 100644
index 000000000..19c55e621
--- /dev/null
+++ b/kitty_tests/search_query_parser.py
@@ -0,0 +1,27 @@
+#!/usr/bin/env python
+# License: GPLv3 Copyright: 2022, Kovid Goyal <kovid at kovidgoyal.net>
+
+
+from . import BaseTest
+
+
+class TestSQP(BaseTest):
+
+    def test_search_query_parser(self):
+        from kitty.search_query_parser import search
+        locations = 'id'
+        universal_set = {1, 2, 3, 4, 5}
+
+        def get_matches(location, query, candidates):
+            return {x for x in candidates if query == str(x)}
+
+        def t(q, expected):
+            actual = search(q, locations, universal_set, get_matches)
+            self.ae(actual, expected)
+
+        t('id:1', {1})
+        t('id:1 and id:1', {1})
+        t('id:1 or id:2', {1, 2})
+        t('id:1 and id:2', set())
+        t('not id:1', universal_set - {1})
+        t('(id:1 or id:2) and id:1', {1})