vexor 0.19.0a1__py3-none-any.whl → 0.21.0__py3-none-any.whl

vexor/services/search_service.py

@@ -7,12 +7,15 @@ from functools import lru_cache
 from pathlib import Path
 import json
 import re
+import numpy as np
 from typing import Sequence, TYPE_CHECKING
 from urllib import error as urlerror
 from urllib import request as urlrequest
 
 from ..config import (
     DEFAULT_EMBED_CONCURRENCY,
+    DEFAULT_EXTRACT_BACKEND,
+    DEFAULT_EXTRACT_CONCURRENCY,
     DEFAULT_FLASHRANK_MAX_LENGTH,
     DEFAULT_FLASHRANK_MODEL,
     DEFAULT_RERANK,
@@ -45,7 +48,11 @@ class SearchRequest:
     exclude_patterns: tuple[str, ...]
     extensions: tuple[str, ...]
     auto_index: bool = True
+    temporary_index: bool = False
+    no_cache: bool = False
     embed_concurrency: int = DEFAULT_EMBED_CONCURRENCY
+    extract_concurrency: int = DEFAULT_EXTRACT_CONCURRENCY
+    extract_backend: str = DEFAULT_EXTRACT_BACKEND
     rerank: str = DEFAULT_RERANK
     flashrank_model: str | None = None
     remote_rerank: RemoteRerankConfig | None = None
@@ -105,6 +112,20 @@ def _normalize_by_max(scores: Sequence[float]) -> list[float]:
     return [score / max_score for score in scores]
 
 
+def _resolve_rerank_candidates(top_k: int) -> int:
+    candidate = int(top_k * 2)
+    return max(20, min(candidate, 150))
+
+
+def _top_indices(scores: np.ndarray, limit: int) -> list[int]:
+    if limit <= 0:
+        return []
+    if limit >= scores.size:
+        return sorted(range(scores.size), key=lambda idx: (-scores[idx], idx))
+    indices = np.argpartition(-scores, limit - 1)[:limit]
+    return sorted(indices.tolist(), key=lambda idx: (-scores[idx], idx))
+
+
 def _bm25_scores(
     query_tokens: Sequence[str],
     documents: Sequence[Sequence[str]],
@@ -336,9 +357,13 @@ def _apply_remote_rerank(
 def perform_search(request: SearchRequest) -> SearchResponse:
     """Execute the semantic search flow and return ranked results."""
 
+    if request.temporary_index or request.no_cache:
+        return _perform_search_with_temporary_index(request)
+
     from ..cache import (  # local import
         embedding_cache_key,
         list_cache_entries,
+        load_chunk_metadata,
         load_embedding_cache,
         load_index_vectors,
         load_query_vector,
@@ -375,12 +400,15 @@ def perform_search(request: SearchRequest) -> SearchResponse:
             model_name=request.model_name,
             batch_size=request.batch_size,
             embed_concurrency=request.embed_concurrency,
+            extract_concurrency=request.extract_concurrency,
+            extract_backend=request.extract_backend,
             provider=request.provider,
             base_url=request.base_url,
             api_key=request.api_key,
             local_cuda=request.local_cuda,
             exclude_patterns=request.exclude_patterns,
             extensions=request.extensions,
+            no_cache=request.no_cache,
         )
         if result.status == IndexStatus.EMPTY:
             return SearchResponse(
@@ -435,6 +463,7 @@ def perform_search(request: SearchRequest) -> SearchResponse:
 
     file_snapshot = metadata.get("files", [])
     chunk_entries = metadata.get("chunks", [])
+    chunk_ids = metadata.get("chunk_ids", [])
     stale = bool(file_snapshot) and not is_cache_current(
         request.directory,
         request.include_hidden,
@@ -455,12 +484,15 @@ def perform_search(request: SearchRequest) -> SearchResponse:
             model_name=request.model_name,
             batch_size=request.batch_size,
             embed_concurrency=request.embed_concurrency,
+            extract_concurrency=request.extract_concurrency,
+            extract_backend=request.extract_backend,
             provider=request.provider,
             base_url=request.base_url,
             api_key=request.api_key,
             local_cuda=request.local_cuda,
             exclude_patterns=index_excludes,
             extensions=index_extensions,
+            no_cache=request.no_cache,
         )
         if result.status == IndexStatus.EMPTY:
             return SearchResponse(
@@ -529,7 +561,6 @@ def perform_search(request: SearchRequest) -> SearchResponse:
             index_empty=True,
         )
 
-    from sklearn.metrics.pairwise import cosine_similarity  # local import
     from ..search import SearchResult, VexorSearcher  # local import
     searcher = VexorSearcher(
         model_name=request.model_name,
@@ -542,9 +573,9 @@ def perform_search(request: SearchRequest) -> SearchResponse:
     )
     query_vector = None
     query_hash = None
-    query_text_hash = embedding_cache_key(request.query)
+    query_text_hash = None
     index_id = metadata.get("index_id")
-    if index_id is not None:
+    if index_id is not None and not request.no_cache:
         query_hash = query_cache_key(request.query, request.model_name)
         try:
             query_vector = load_query_vector(int(index_id), query_hash)
@@ -554,7 +585,8 @@ def perform_search(request: SearchRequest) -> SearchResponse:
         if query_vector is not None and query_vector.size != file_vectors.shape[1]:
             query_vector = None
 
-    if query_vector is None:
+    if query_vector is None and not request.no_cache:
+        query_text_hash = embedding_cache_key(request.query)
         cached = load_embedding_cache(request.model_name, [query_text_hash])
         query_vector = cached.get(query_text_hash)
         if query_vector is not None and query_vector.size != file_vectors.shape[1]:
@@ -562,25 +594,57 @@ def perform_search(request: SearchRequest) -> SearchResponse:
 
     if query_vector is None:
         query_vector = searcher.embed_texts([request.query])[0]
-        try:
-            store_embedding_cache(
-                model=request.model_name,
-                embeddings={query_text_hash: query_vector},
-            )
-        except Exception:  # pragma: no cover - best-effort cache storage
-            pass
-    if query_vector is not None and index_id is not None and query_hash is not None:
+        if not request.no_cache:
+            if query_text_hash is None:
+                query_text_hash = embedding_cache_key(request.query)
+            try:
+                store_embedding_cache(
+                    model=request.model_name,
+                    embeddings={query_text_hash: query_vector},
+                )
+            except Exception:  # pragma: no cover - best-effort cache storage
+                pass
+    if (
+        not request.no_cache
+        and query_vector is not None
+        and index_id is not None
+        and query_hash is not None
+    ):
         try:
             store_query_vector(int(index_id), query_hash, request.query, query_vector)
         except Exception:  # pragma: no cover - best-effort cache storage
             pass
-    similarities = cosine_similarity(
-        query_vector.reshape(1, -1),
-        file_vectors,
-    )[0]
-    scored = []
-    for idx, (path, score) in enumerate(zip(paths, similarities)):
-        chunk_meta = chunk_entries[idx] if idx < len(chunk_entries) else {}
+    reranker = None
+    rerank = (request.rerank or DEFAULT_RERANK).strip().lower()
+    use_rerank = rerank in {"bm25", "flashrank", "remote"}
+    if use_rerank:
+        candidate_limit = _resolve_rerank_candidates(request.top_k)
+    else:
+        candidate_limit = request.top_k
+    candidate_count = min(len(paths), candidate_limit)
+
+    query_vector = np.asarray(query_vector, dtype=np.float32).ravel()
+    similarities = np.asarray(file_vectors @ query_vector, dtype=np.float32)
+    top_indices = _top_indices(similarities, candidate_count)
+    chunk_meta_by_id: dict[int, dict] = {}
+    if chunk_ids:
+        candidate_ids = [
+            chunk_ids[idx] for idx in top_indices if idx < len(chunk_ids)
+        ]
+        if candidate_ids:
+            try:
+                chunk_meta_by_id = load_chunk_metadata(candidate_ids)
+            except Exception:  # pragma: no cover - best-effort metadata lookup
+                chunk_meta_by_id = {}
+    scored: list[SearchResult] = []
+    for idx in top_indices:
+        path = paths[idx]
+        score = similarities[idx]
+        chunk_meta = {}
+        if chunk_ids and idx < len(chunk_ids):
+            chunk_meta = chunk_meta_by_id.get(chunk_ids[idx], {})
+        elif idx < len(chunk_entries):
+            chunk_meta = chunk_entries[idx]
         start_line = chunk_meta.get("start_line")
         end_line = chunk_meta.get("end_line")
         scored.append(
@@ -593,12 +657,138 @@ def perform_search(request: SearchRequest) -> SearchResponse:
                 end_line=int(end_line) if end_line is not None else None,
             )
         )
-    scored.sort(key=lambda item: item.score, reverse=True)
+    if use_rerank:
+        candidates = scored
+        if rerank == "bm25":
+            candidates = _apply_bm25_rerank(request.query, candidates)
+            reranker = "bm25"
+        elif rerank == "flashrank":
+            candidates = _apply_flashrank_rerank(
+                request.query,
+                candidates,
+                request.flashrank_model,
+            )
+            reranker = "flashrank"
+        else:
+            candidates = _apply_remote_rerank(
+                request.query,
+                candidates,
+                request.remote_rerank,
+            )
+            reranker = "remote"
+        results = candidates[: request.top_k]
+    else:
+        results = scored[: request.top_k]
+    return SearchResponse(
+        base_path=request.directory,
+        backend=searcher.device,
+        results=results,
+        is_stale=stale,
+        index_empty=False,
+        reranker=reranker,
+    )
+
+
+def _perform_search_with_temporary_index(request: SearchRequest) -> SearchResponse:
+    from .index_service import build_index_in_memory  # local import
+
+    paths, file_vectors, metadata = build_index_in_memory(
+        request.directory,
+        include_hidden=request.include_hidden,
+        respect_gitignore=request.respect_gitignore,
+        mode=request.mode,
+        recursive=request.recursive,
+        model_name=request.model_name,
+        batch_size=request.batch_size,
+        embed_concurrency=request.embed_concurrency,
+        extract_concurrency=request.extract_concurrency,
+        extract_backend=request.extract_backend,
+        provider=request.provider,
+        base_url=request.base_url,
+        api_key=request.api_key,
+        local_cuda=request.local_cuda,
+        exclude_patterns=request.exclude_patterns,
+        extensions=request.extensions,
+        no_cache=request.no_cache,
+    )
+
+    if not len(paths):
+        return SearchResponse(
+            base_path=request.directory,
+            backend=None,
+            results=[],
+            is_stale=False,
+            index_empty=True,
+        )
+
+    from ..search import SearchResult, VexorSearcher  # local import
+
+    searcher = VexorSearcher(
+        model_name=request.model_name,
+        batch_size=request.batch_size,
+        embed_concurrency=request.embed_concurrency,
+        provider=request.provider,
+        base_url=request.base_url,
+        api_key=request.api_key,
+        local_cuda=request.local_cuda,
+    )
+    query_vector = None
+    query_text_hash = None
+    if not request.no_cache:
+        from ..cache import embedding_cache_key, load_embedding_cache, store_embedding_cache
+
+        query_text_hash = embedding_cache_key(request.query)
+        cached = load_embedding_cache(request.model_name, [query_text_hash])
+        query_vector = cached.get(query_text_hash)
+        if query_vector is not None and query_vector.size != file_vectors.shape[1]:
+            query_vector = None
+
+    if query_vector is None:
+        query_vector = searcher.embed_texts([request.query])[0]
+        if not request.no_cache:
+            if query_text_hash is None:
+                from ..cache import embedding_cache_key, store_embedding_cache
+
+                query_text_hash = embedding_cache_key(request.query)
+            try:
+                store_embedding_cache(
+                    model=request.model_name,
+                    embeddings={query_text_hash: query_vector},
+                )
+            except Exception:  # pragma: no cover - best-effort cache storage
+                pass
     reranker = None
     rerank = (request.rerank or DEFAULT_RERANK).strip().lower()
-    if rerank in {"bm25", "flashrank", "remote"}:
-        candidate_count = min(len(scored), request.top_k * 2)
-        candidates = scored[:candidate_count]
+    use_rerank = rerank in {"bm25", "flashrank", "remote"}
+    if use_rerank:
+        candidate_limit = _resolve_rerank_candidates(request.top_k)
+    else:
+        candidate_limit = request.top_k
+    candidate_count = min(len(paths), candidate_limit)
+
+    query_vector = np.asarray(query_vector, dtype=np.float32).ravel()
+    similarities = np.asarray(file_vectors @ query_vector, dtype=np.float32)
+    top_indices = _top_indices(similarities, candidate_count)
+    chunk_entries = metadata.get("chunks", [])
+    scored: list[SearchResult] = []
+    for idx in top_indices:
+        path = paths[idx]
+        score = similarities[idx]
+        chunk_meta = chunk_entries[idx] if idx < len(chunk_entries) else {}
+        start_line = chunk_meta.get("start_line")
+        end_line = chunk_meta.get("end_line")
+        scored.append(
+            SearchResult(
+                path=path,
+                score=float(score),
+                preview=chunk_meta.get("preview"),
+                chunk_index=int(chunk_meta.get("chunk_index", 0)),
+                start_line=int(start_line) if start_line is not None else None,
+                end_line=int(end_line) if end_line is not None else None,
+            )
+        )
+    if use_rerank:
+        candidates = scored
         if rerank == "bm25":
             candidates = _apply_bm25_rerank(request.query, candidates)
             reranker = "bm25"
@@ -623,7 +813,7 @@ def perform_search(request: SearchRequest) -> SearchResponse:
         base_path=request.directory,
         backend=searcher.device,
         results=results,
-        is_stale=stale,
+        is_stale=False,
         index_empty=False,
         reranker=reranker,
     )
@@ -764,6 +954,7 @@ def _filter_index_by_extensions(
     ext_set = {ext.lower() for ext in extensions if ext}
     if not ext_set:
         return list(paths), file_vectors, metadata
+    chunk_ids = metadata.get("chunk_ids")
     keep_indices: list[int] = []
     filtered_paths: list[Path] = []
     for idx, path in enumerate(paths):
@@ -778,6 +969,8 @@ def _filter_index_by_extensions(
             ext_set,
         )
         filtered_metadata["chunks"] = []
+        if chunk_ids is not None:
+            filtered_metadata["chunk_ids"] = []
         return [], filtered_vectors, filtered_metadata
     filtered_vectors = file_vectors[keep_indices]
     chunk_entries = metadata.get("chunks", [])
@@ -790,6 +983,10 @@ def _filter_index_by_extensions(
         ext_set,
     )
     filtered_metadata["chunks"] = filtered_chunks
+    if chunk_ids is not None:
+        filtered_metadata["chunk_ids"] = [
+            chunk_ids[idx] for idx in keep_indices if idx < len(chunk_ids)
+        ]
     return filtered_paths, filtered_vectors, filtered_metadata
 
 
@@ -802,6 +999,7 @@ def _filter_index_by_exclude_patterns(
 ) -> tuple[list[Path], Sequence[Sequence[float]], dict]:
     if exclude_spec is None:
         return list(paths), file_vectors, metadata
+    chunk_ids = metadata.get("chunk_ids")
     keep_indices: list[int] = []
     filtered_paths: list[Path] = []
     root_resolved = root.resolve()
@@ -822,6 +1020,8 @@ def _filter_index_by_exclude_patterns(
             exclude_spec,
         )
         filtered_metadata["chunks"] = []
+        if chunk_ids is not None:
+            filtered_metadata["chunk_ids"] = []
         return [], filtered_vectors, filtered_metadata
     filtered_vectors = file_vectors[keep_indices]
     chunk_entries = metadata.get("chunks", [])
@@ -834,6 +1034,10 @@ def _filter_index_by_exclude_patterns(
         exclude_spec,
     )
     filtered_metadata["chunks"] = filtered_chunks
+    if chunk_ids is not None:
+        filtered_metadata["chunk_ids"] = [
+            chunk_ids[idx] for idx in keep_indices if idx < len(chunk_ids)
+        ]
     return filtered_paths, filtered_vectors, filtered_metadata
 
 
@@ -850,6 +1054,7 @@ def _filter_index_by_directory(
         relative_dir = directory.resolve().relative_to(index_root.resolve())
     except ValueError:
         return list(paths), file_vectors, metadata
+    chunk_ids = metadata.get("chunk_ids")
     keep_indices: list[int] = []
     filtered_paths: list[Path] = []
     for idx, path in enumerate(paths):
@@ -870,6 +1075,8 @@ def _filter_index_by_directory(
             recursive=recursive,
         )
         filtered_metadata["chunks"] = []
+        if chunk_ids is not None:
+            filtered_metadata["chunk_ids"] = []
         filtered_metadata["root"] = str(directory)
         return [], filtered_vectors, filtered_metadata
     filtered_vectors = file_vectors[keep_indices]
@@ -884,6 +1091,10 @@ def _filter_index_by_directory(
         recursive=recursive,
     )
     filtered_metadata["chunks"] = filtered_chunks
+    if chunk_ids is not None:
+        filtered_metadata["chunk_ids"] = [
+            chunk_ids[idx] for idx in keep_indices if idx < len(chunk_ids)
+        ]
     filtered_metadata["root"] = str(directory)
     return filtered_paths, filtered_vectors, filtered_metadata
 

vexor/text.py

@@ -19,6 +19,7 @@ class Messages:
     HELP_SEARCH_FORMAT = (
         "Output format (rich=table, porcelain=tab-separated for scripts, porcelain-z=NUL-delimited)."
     )
+    HELP_NO_CACHE = "Disable all disk caches (index + embedding/query)."
     HELP_INCLUDE_HIDDEN = "Use the index built with hidden files included."
     HELP_INDEX_PATH = "Root directory to scan for indexing."
     HELP_INDEX_INCLUDE = "Include hidden files and directories when building the index."
@@ -58,6 +59,8 @@ class Messages:
     HELP_SET_MODEL = "Set the default embedding model."
     HELP_SET_BATCH = "Set the default batch size (0 = single request)."
     HELP_SET_EMBED_CONCURRENCY = "Set the number of concurrent embedding requests."
+    HELP_SET_EXTRACT_CONCURRENCY = "Set the number of concurrent file extraction workers."
+    HELP_SET_EXTRACT_BACKEND = "Set the extraction backend (auto, thread, process)."
     HELP_SET_PROVIDER = "Set the default embedding provider (e.g., gemini, openai, custom, or local)."
     HELP_SET_BASE_URL = "Override the provider's base URL (leave unset for official endpoints)."
     HELP_CLEAR_BASE_URL = "Remove the custom base URL override."
@@ -116,6 +119,10 @@ class Messages:
     ERROR_EMPTY_QUERY = "Query text must not be empty."
     ERROR_BATCH_NEGATIVE = "Batch size must be >= 0"
     ERROR_CONCURRENCY_INVALID = "Embedding concurrency must be >= 1"
+    ERROR_EXTRACT_CONCURRENCY_INVALID = "Extraction concurrency must be >= 1"
+    ERROR_EXTRACT_BACKEND_INVALID = (
+        "Unsupported extraction backend '{value}'. Allowed values: {allowed}."
+    )
     ERROR_MODE_INVALID = "Unsupported mode '{value}'. Allowed values: {allowed}."
     ERROR_PROVIDER_INVALID = "Unsupported provider '{value}'. Allowed values: {allowed}."
     ERROR_RERANK_INVALID = "Unsupported rerank value '{value}'. Allowed values: {allowed}."
@@ -265,6 +272,8 @@ class Messages:
     INFO_MODEL_SET = "Default model set to {value}."
     INFO_BATCH_SET = "Default batch size set to {value}."
     INFO_EMBED_CONCURRENCY_SET = "Embedding concurrency set to {value}."
+    INFO_EXTRACT_CONCURRENCY_SET = "Extraction concurrency set to {value}."
+    INFO_EXTRACT_BACKEND_SET = "Extraction backend set to {value}."
     INFO_PROVIDER_SET = "Default provider set to {value}."
     INFO_BASE_URL_SET = "Base URL override set to {value}."
     INFO_BASE_URL_CLEARED = "Base URL override cleared."
@@ -299,12 +308,16 @@ class Messages:
     ERROR_CONFIG_EDITOR_NOT_FOUND = "Unable to determine a text editor. Set $VISUAL or $EDITOR, or install nano/vi."
     ERROR_CONFIG_EDITOR_FAILED = "Editor exited with status {code}."
     ERROR_CONFIG_EDITOR_LAUNCH = "Failed to launch editor: {reason}."
+    ERROR_CONFIG_JSON_INVALID = "Config JSON must be an object."
+    ERROR_CONFIG_VALUE_INVALID = "Config JSON has invalid value for {field}."
     INFO_CONFIG_SUMMARY = (
         "API key set: {api}\n"
         "Default provider: {provider}\n"
         "Default model: {model}\n"
         "Default batch size: {batch}\n"
         "Embedding concurrency: {concurrency}\n"
+        "Extract concurrency: {extract_concurrency}\n"
+        "Extract backend: {extract_backend}\n"
         "Auto index: {auto_index}\n"
         "Rerank: {rerank}\n"
         "{flashrank_line}"
@@ -315,6 +328,7 @@ class Messages:
     INFO_FLASHRANK_MODEL_SUMMARY = "FlashRank model: {value}"
     INFO_REMOTE_RERANK_SUMMARY = "Remote rerank: {value}"
     INFO_SEARCH_RUNNING = "Searching cached index under {path}..."
+    INFO_SEARCH_RUNNING_NO_CACHE = "Searching in-memory index under {path}..."
     INFO_DOCTOR_CHECKING = "Checking if `vexor` is on PATH..."
     INFO_DOCTOR_FOUND = "`vexor` command is available at {path}."
     ERROR_DOCTOR_MISSING = (

{vexor-0.19.0a1.dist-info → vexor-0.21.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vexor
-Version: 0.19.0a1
+Version: 0.21.0
 Summary: A vector-powered CLI for semantic search over files.
 Project-URL: Repository, https://github.com/scarletkc/vexor
 Author: scarletkc
@@ -69,9 +69,8 @@ Description-Content-Type: text/markdown
 
 ---
 
-**Vexor** is a vector-powered CLI and desktop app for semantic file search. It uses configurable embedding models and ranks results by cosine similarity.
-
-![GUI](https://raw.githubusercontent.com/scarletkc/vexor/refs/heads/main/assets/gui_demo.png)
+**Vexor** is a semantic search engine that builds reusable indexes over files and code.
+It supports configurable embedding and reranking providers, and exposes the same core through a Python API, a CLI tool, and an optional desktop frontend.
 
 <video src="https://github.com/user-attachments/assets/4d53eefd-ab35-4232-98a7-f8dc005983a9" controls="controls" style="max-width: 600px;">
       Vexor Demo Video
@@ -98,18 +97,13 @@ vexor init
 ```
 The wizard also runs automatically on first use when no config exists.
 
-### 1. Configure API Key
-```bash
-vexor config --set-api-key "YOUR_KEY"
-```
-Or via environment: `VEXOR_API_KEY`, `OPENAI_API_KEY`, or `GOOGLE_GENAI_API_KEY`.
-
-### 2. Search
+### 1. Search
 ```bash
-vexor "api client config"  # defaults to search
-vexor search "api client config"  # searches current directory
+vexor "api client config"  # defaults to search current directory
 # or explicit path:
 vexor search "api client config" --path ~/projects/demo --top 5
+# in-memory search only:
+vexor search "api client config" --no-cache 
 ```
 
 Vexor auto-indexes on first search. Example output:
@@ -122,7 +116,7 @@ Vexor semantic file search results
 3   0.809        ./tests/test_config_loader.py   -       tests for config loader
 ```
 
-### 3. Explicit Index (Optional)
+### 2. Explicit Index (Optional)
 ```bash
 vexor index  # indexes current directory
 # or explicit path:
@@ -130,6 +124,15 @@ vexor index --path ~/projects/demo --mode code
 ```
 Useful for CI warmup or when `auto_index` is disabled.
 
+## Desktop App (Experimental)
+
+> The desktop app is experimental and not actively maintained.
+> It may be unstable. For production use, prefer the CLI.
+
+![GUI](https://raw.githubusercontent.com/scarletkc/vexor/refs/heads/main/assets/gui_demo.png)
+
+Download the desktop app from [releases](https://github.com/scarletkc/vexor/releases).
+
 ## Python API
 
 Vexor can also be imported and used directly from Python:
@@ -144,8 +147,19 @@ for hit in response.results:
     print(hit.path, hit.score)
 ```
 
-By default it reads `~/.vexor/config.json`. To ignore config and pass everything explicitly,
-set `use_config=False`.
+By default it reads `~/.vexor/config.json`. For runtime config overrides, cache
+controls, and per-call options, see [`docs/api/python.md`](https://github.com/scarletkc/vexor/tree/main/docs/api/python.md).
+
+## AI Agent Skill
+
+This repo includes a skill for AI agents to use Vexor effectively:
+
+```bash
+vexor install --skills claude  # Claude Code
+vexor install --skills codex   # Codex
+```
+
+Skill source: [`plugins/vexor/skills/vexor-cli`](https://github.com/scarletkc/vexor/raw/refs/heads/main/plugins/vexor/skills/vexor-cli/SKILL.md)
 
 ## Configuration
 
@@ -153,7 +167,9 @@ set `use_config=False`.
 vexor config --set-provider openai          # default; also supports gemini/custom/local
 vexor config --set-model text-embedding-3-small
 vexor config --set-batch-size 0             # 0 = single request
-vexor config --set-embed-concurrency 2       # parallel embedding requests
+vexor config --set-embed-concurrency 4       # parallel embedding requests
+vexor config --set-extract-concurrency 4     # parallel file extraction workers
+vexor config --set-extract-backend auto      # auto|thread|process (default: auto)
 vexor config --set-auto-index true          # auto-index before search (default)
 vexor config --rerank bm25                  # optional BM25 rerank for top-k results
 vexor config --rerank flashrank             # FlashRank rerank (requires optional extra)
@@ -175,10 +191,16 @@ FlashRank requires `pip install "vexor[flashrank]"` and caches models under `~/.
 
 Config stored in `~/.vexor/config.json`.
 
+### Configure API Key
+```bash
+vexor config --set-api-key "YOUR_KEY"
+```
+Or via environment: `VEXOR_API_KEY`, `OPENAI_API_KEY`, or `GOOGLE_GENAI_API_KEY`.
+
 ### Rerank
 
-Rerank reorders the semantic results with a secondary ranker. It uses 2x the requested
-`--top` as candidates (e.g., top 10 reranked to show 5).
+Rerank reorders the semantic results with a secondary ranker. Candidate sizing uses
+`clamp(int(--top * 2), 20, 150)`.
 
 Recommended defaults:
 - Keep `off` unless you want extra precision.
@@ -285,20 +307,10 @@ Re-running `vexor index` only re-embeds changed files; >50% changes trigger full
 | `--no-respect-gitignore` | Include gitignored files |
 | `--format porcelain` | Script-friendly TSV output |
 | `--format porcelain-z` | NUL-delimited output |
+| `--no-cache` | In-memory only; do not read/write index cache |
 
 Porcelain output fields: `rank`, `similarity`, `path`, `chunk_index`, `start_line`, `end_line`, `preview` (line fields are `-` when unavailable).
 
-## AI Agent Skill
-
-This repo includes a skill for AI agents to use Vexor effectively:
-
-```bash
-vexor install --skills claude  # Claude Code
-vexor install --skills codex   # Codex
-```
-
-Skill source: [`plugins/vexor/skills/vexor-cli`](https://github.com/scarletkc/vexor/raw/refs/heads/main/plugins/vexor/skills/vexor-cli/SKILL.md)
-
 ## Documentation
 
 See [docs](https://github.com/scarletkc/vexor/tree/main/docs) for more details.

vexor-0.21.0.dist-info/RECORD

@@ -0,0 +1,33 @@
+vexor/__init__.py,sha256=i0ly8cFA4N_PEQ_rhYgoLp2NPRQc3_ln8Gfi8QWjXSQ,441
+vexor/__main__.py,sha256=ZFzom1wCfP6TPXe3aoDFpNcUgjbCZ7Quy_vfzNsH5Fw,426
+vexor/api.py,sha256=YCHpiydbPbRJUqdQYrpwe1JrRI-w_7LRuyZDGBP1_d4,11506
+vexor/cache.py,sha256=3i9FKFLSyZ1kx-w1apc12umPaQxWqMP-P8_lvo67hBw,52832
+vexor/cli.py,sha256=M9GKdD_mJ068Zpm62znTp0KhhKp1dkh_WHmfJHR9hwU,68094
+vexor/config.py,sha256=CiPfEH7Ilt6XepEx4p02qfW5HfkpNDBjhEMyckbSWaA,17413
+vexor/modes.py,sha256=N_wAWoqbxmCfko-v520p59tpAYvUwraCSSQRtMaF4ac,11549
+vexor/output.py,sha256=iooZgLlK8dh7ajJ4XMHUNNx0qyTVtD_OAAwrBx5MeqE,864
+vexor/search.py,sha256=MSU4RmH6waFYOofkIdo8_ElTiz1oNaKuvr-3umif7Bs,6826
+vexor/text.py,sha256=2aK5nJHkosmbmyzp9o_Tzb3YlmVnju_IX8BcEPUdhTA,24794
+vexor/utils.py,sha256=GzfYW2rz1-EuJjkevqZVe8flLRtrQ60OWMmFNbMh62k,12472
+vexor/providers/__init__.py,sha256=kCEoV03TSLKcxDUYVNjXnrVoLU5NpfNXjp1w1Ak2imE,92
+vexor/providers/gemini.py,sha256=IWHHjCMJC0hUHQPhuaJ_L_97c_mnOXkPkCVdrIR6z-g,5705
+vexor/providers/local.py,sha256=5X_WYCXgyBGIVvvVLgMnDjTkPR4GBF0ksNPyviBlB7w,4838
+vexor/providers/openai.py,sha256=YnJDY9gJW7RfGGdkgswVHvmOKNvgLRQUsbpA1MUuLPg,5356
+vexor/services/__init__.py,sha256=dA_i2N03vlYmbZbEK2knzJLWviunkNWbzN2LWPNvMk0,160
+vexor/services/cache_service.py,sha256=ywt6AgupCJ7_wC3je4znCMw5_VBouw3skbDTAt8xw6o,1639
+vexor/services/config_service.py,sha256=PojolfbSKh9pW8slF4qxCOs9hz5L6xvjf_nB7vfVlsU,5039
+vexor/services/content_extract_service.py,sha256=zdhLxpNv70BU7irLf3Uc0ou9rKSvdjtrDcHkgRKlMn4,26421
+vexor/services/index_service.py,sha256=FXf1bBoqj4-K1l38ItxHf6Oh7QHVIdNAdVY2kg_Zoq8,32265
+vexor/services/init_service.py,sha256=3D04hylGA9FRQhLHCfR95nMko3vb5MNBcRb9nWWaUE8,26863
+vexor/services/js_parser.py,sha256=eRtW6KlK4JBYDGbyoecHVqLZ0hcx-Cc0kx6bOujHPAQ,16254
+vexor/services/keyword_service.py,sha256=vmke8tII9kTwRDdBaLHBc6Hpy_B3p98L65iGkCQgtMU,2211
+vexor/services/search_service.py,sha256=K7SiAuMA7bGeyPWOHPMKpFFvzzkj5kHWwa3p94NakJs,38663
+vexor/services/skill_service.py,sha256=Rrgt3OMsKPPiXOiRhSNAWjBM9UNz9qmSWQe3uYGzq4M,4863
+vexor/services/system_service.py,sha256=KPlv83v3rTvBiNiH7vrp6tDmt_AqHxuUd-5RI0TfvWs,24638
+vexor/_bundled_skills/vexor-cli/SKILL.md,sha256=m3FlyqgHBdRwyGPEp8PrUS21K0G2jEl88tRvhSPta08,2798
+vexor/_bundled_skills/vexor-cli/references/install-vexor.md,sha256=IUBShLI1mAxugwUIMAJQ5_j6KcaPWfobe0gSd6MWU7w,1245
+vexor-0.21.0.dist-info/METADATA,sha256=Lc5PHY_Ir3F56ILYe6IBlkwhN6gMQGZvf48f7x_uVDg,13494
+vexor-0.21.0.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+vexor-0.21.0.dist-info/entry_points.txt,sha256=dvxp6Q1R1d6bozR7TwmpdJ0X_v83MkzsLPagGY_lfr0,40
+vexor-0.21.0.dist-info/licenses/LICENSE,sha256=wP7TAKRll1t9LoYGxWS9NikPM_0hCc00LmlLyvQBsL8,1066
+vexor-0.21.0.dist-info/RECORD,,