unique-toolkit 2026.30.0.dev7__tar.gz → 2026.30.0.dev8__tar.gz

{unique_toolkit-2026.30.0.dev7 → unique_toolkit-2026.30.0.dev8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: unique-toolkit
-Version: 2026.30.0.dev7
+Version: 2026.30.0.dev8
 Summary: 
 Author: Cedric Klinkert, Martin Fadler, Sadique Sheik, Fabian Schläpfer, Pascal Hauri, Ejiroghene Oki-Peter, Rami Azouz, Ricardo Visini, Sanzio Monti
 Author-email: Cedric Klinkert <cedric.klinkert@unique.ch>, Martin Fadler <martin.fadler@unique.ch>, Sadique Sheik <sadique@unique.ch>, Fabian Schläpfer <fabian@unique.ch>, Pascal Hauri <pascal@unique.ch>, Ejiroghene Oki-Peter <oki-peter.ejiroghene@unique.ai>, Rami Azouz <rami.ext@unique.ch>, Ricardo Visini <ricardo.visini@unique.ch>, Sanzio Monti <sanzio.monti@unique.ai>
@@ -32,6 +32,7 @@ Requires-Dist: pypandoc>=1.16.2,<2
 Requires-Dist: tenacity>=9.1.4
 Requires-Dist: pydantic-core>=2.41.5
 Requires-Dist: cachetools>=5.0,<6
+Requires-Dist: rapidfuzz>=3.0,<4
 Requires-Dist: fastapi>=0.121.0,<0.122 ; extra == 'fastapi'
 Requires-Dist: starlette>=0.49.1 ; extra == 'fastapi'
 Requires-Dist: uvicorn>=0.37.0,<0.38 ; extra == 'fastapi'

{unique_toolkit-2026.30.0.dev7 → unique_toolkit-2026.30.0.dev8}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "unique_toolkit"
-version = "2026.30.0.dev7"
+version = "2026.30.0.dev8"
 description = ""
 readme = "README.md"
 requires-python = ">=3.12"
@@ -44,6 +44,7 @@ dependencies = [
     "tenacity>=9.1.4",
     "pydantic-core>=2.41.5",
     "cachetools>=5.0,<6",
+    "rapidfuzz>=3.0,<4",
 ]
 
 [project.optional-dependencies]

{unique_toolkit-2026.30.0.dev7 → unique_toolkit-2026.30.0.dev8}/unique_toolkit/experimental/components/content_tree/service.py

@@ -19,12 +19,14 @@ on its own.
 from __future__ import annotations
 
 import asyncio
-import difflib
 import functools
 import json
+import re
 from collections.abc import Callable
 from typing import TYPE_CHECKING, Any, Protocol, Self, overload
 
+from rapidfuzz import fuzz
+
 from unique_toolkit._common.validate_required_values import validate_required_values
 from unique_toolkit.app.unique_settings import UniqueSettings
 from unique_toolkit.content.schemas import ContentInfo
@@ -44,6 +46,32 @@ if TYPE_CHECKING:
     from unique_toolkit.app.unique_settings import UniqueContext
 
 
+_NON_ALNUM_RE = re.compile(r"[^a-zA-Z0-9]+")
+_TRAILING_EXTENSION_RE = re.compile(r"\.[A-Za-z0-9]{1,8}$")
+
+
+def _tokenize_for_fuzzy_scoring(
+    value: str, case_sensitive: bool, *, strip_extension: bool = False
+) -> str:
+    """Normalize ``value`` into whitespace-separated tokens for
+    :func:`rapidfuzz.fuzz.token_set_ratio`, which only tokenizes on
+    whitespace — filenames/paths use ``_``/``-``/``/`` (and other
+    punctuation) as their real word separators, so those need converting to
+    spaces first or every candidate is scored as a single opaque token (no
+    better than a plain char-ratio). Mirrors what
+    :func:`rapidfuzz.utils.default_process` does, minus the unconditional
+    lowercasing it bundles in — kept as a local regex rather than that
+    helper (via ``fuzz.token_set_ratio``'s ``processor=`` argument) so
+    ``case_sensitive=True`` can still skip lowercasing while keeping
+    separator normalization.
+    """
+    if strip_extension:
+        value = _TRAILING_EXTENSION_RE.sub("", value)
+    if not case_sensitive:
+        value = value.lower()
+    return _NON_ALNUM_RE.sub(" ", value).strip()
+
+
 class _CachedResolveTaskFactory(Protocol):
     """Structural type for :func:`functools.cache`-wrapped resolve-task factory.
 
@@ -340,16 +368,24 @@ class ContentTree:
     ) -> list[FuzzyMatch]:
         """Fuzzy-match ``query`` against visible file names and/or paths.
 
-        Scoring uses :class:`difflib.SequenceMatcher` (stdlib), which returns a
-        ratio in ``[0.0, 1.0]``. Matching is case-insensitive by default since
-        file names in a knowledge base tend to be noisy.
+        Scoring uses :func:`rapidfuzz.fuzz.token_set_ratio` over each string's
+        tokens (split on whitespace, ``_``, ``-``, and ``/``, extension
+        stripped), scaled to ``[0.0, 1.0]``. Token-set scoring means a short
+        query that's a complete word inside a much longer filename scores
+        highly (ideally ``1.0`` for an exact token match) instead of being
+        penalized for the filename's unrelated length — a plain character-level
+        ratio (e.g. :class:`difflib.SequenceMatcher`, used here previously)
+        scores "alpensys" vs. "alpensys_budget_vs_actual_q3_2024.docx" at only
+        ~0.35, indistinguishable from genuinely unrelated files, because it
+        penalizes every unmatched character in the (much longer) candidate.
+        Matching is case-insensitive by default since file names in a
+        knowledge base tend to be noisy.
 
         Args:
             query: The search string (typically a fragment of a filename or path).
             limit: Maximum number of matches to return, after sorting by score
                 descending.
-            min_score: Drop matches scoring below this threshold. ``0.6`` is
-                :mod:`difflib`'s own rule of thumb for "reasonably close".
+            min_score: Drop matches scoring below this threshold.
             match_on: Score against the basename (``"key"``), the joined folder
                 path (``"path"``), or take the max of both (``"both"``).
             case_sensitive: If ``False`` (default) both sides are lowercased
@@ -369,25 +405,26 @@ class ContentTree:
             metadata_filter=metadata_filter,
             max_concurrent_scope_lookups=max_concurrent_scope_lookups,
         )
-        normalized_query = query if case_sensitive else query.lower()
+        normalized_query = _tokenize_for_fuzzy_scoring(query, case_sensitive)
 
         matches: list[FuzzyMatch] = []
         for content_info, segments in rows:
-            key_candidate = content_info.key
-            path_candidate = "/".join(segments)
-            if not case_sensitive:
-                key_candidate = key_candidate.lower()
-                path_candidate = path_candidate.lower()
+            key_candidate = _tokenize_for_fuzzy_scoring(
+                content_info.key, case_sensitive, strip_extension=True
+            )
+            path_candidate = _tokenize_for_fuzzy_scoring(
+                "/".join(segments), case_sensitive
+            )
 
             score_key = match_on in ("key", "both")
             score_path = match_on in ("path", "both")
             key_score = (
-                difflib.SequenceMatcher(None, normalized_query, key_candidate).ratio()
+                fuzz.token_set_ratio(normalized_query, key_candidate) / 100.0
                 if score_key
                 else 0.0
             )
             path_score = (
-                difflib.SequenceMatcher(None, normalized_query, path_candidate).ratio()
+                fuzz.token_set_ratio(normalized_query, path_candidate) / 100.0
                 if score_path
                 else 0.0
             )

{unique_toolkit-2026.30.0.dev7 → unique_toolkit-2026.30.0.dev8}/unique_toolkit/services/knowledge_base.py

@@ -230,7 +230,7 @@ class KnowledgeBaseService:
             searches = search_content_chunks(
                 user_id=self._user_id,
                 company_id=self._company_id,
-                chat_id="",
+                chat_id=None,
                 search_string=search_string,
                 search_type=search_type,
                 limit=limit,
@@ -327,7 +327,7 @@ class KnowledgeBaseService:
             searches = await search_content_chunks_async(
                 user_id=self._user_id,
                 company_id=self._company_id,
-                chat_id="",
+                chat_id=None,
                 search_string=search_string,
                 search_type=search_type,
                 limit=limit,
@@ -363,7 +363,7 @@ class KnowledgeBaseService:
         return search_contents(
             user_id=self._user_id,
             company_id=self._company_id,
-            chat_id="",
+            chat_id=None,
             where=where,
             include_failed_content=include_failed_content,
         )
@@ -387,7 +387,7 @@ class KnowledgeBaseService:
         return await search_contents_async(
             user_id=self._user_id,
             company_id=self._company_id,
-            chat_id="",
+            chat_id=None,
             where=where,
             include_failed_content=include_failed_content,
         )
@@ -429,7 +429,7 @@ class KnowledgeBaseService:
             content_name=content_name,
             mime_type=mime_type,
             scope_id=scope_id,
-            chat_id="",
+            chat_id=None,
             skip_ingestion=skip_ingestion,
             ingestion_config=ingestion_config,
             metadata=metadata,
@@ -470,7 +470,7 @@ class KnowledgeBaseService:
             content_name=content_name,
             mime_type=mime_type,
             scope_id=scope_id,
-            chat_id="",
+            chat_id=None,
             skip_ingestion=skip_ingestion,
             ingestion_config=ingestion_config,
             metadata=metadata,
@@ -511,7 +511,7 @@ class KnowledgeBaseService:
             content_name=content_name,
             mime_type=mime_type,
             scope_id=scope_id,
-            chat_id="",
+            chat_id=None,
             skip_ingestion=skip_ingestion,
             skip_excel_ingestion=skip_excel_ingestion,
             ingestion_config=ingestion_config,
@@ -544,7 +544,7 @@ class KnowledgeBaseService:
             user_id=self._user_id,
             company_id=self._company_id,
             content_id=content_id,
-            chat_id="",
+            chat_id=None,
             filename=output_filename,
             tmp_dir_path=output_dir_path,
         )