code-graph-rag 0.2.19__tar.gz → 0.3.2__tar.gz

{code_graph_rag-0.2.19/code_graph_rag.egg-info → code_graph_rag-0.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-graph-rag
-Version: 0.2.19
+Version: 0.3.2
 Summary: The ultimate RAG for your monorepo. Query, understand, and edit multi-language codebases with the power of AI and knowledge graphs
 License-Expression: MIT
 Keywords: rag,retrieval-augmented-generation,knowledge-graph,code-analysis,tree-sitter,mcp,mcp-server,llm,graph-database,semantic-search,codebase,memgraph,developer-tools,monorepo
@@ -78,6 +78,7 @@ Requires-Dist: tree-sitter-sql>=0.3.11; extra == "treesitter-full"
 Requires-Dist: tree-sitter-markdown>=0.5.1; extra == "treesitter-full"
 Provides-Extra: semantic
 Requires-Dist: qdrant-client>=1.9.0; extra == "semantic"
+Requires-Dist: sqlite-vec>=0.1.9; ((sys_platform == "linux" and (platform_machine == "x86_64" or platform_machine == "aarch64")) or sys_platform == "darwin" or (sys_platform == "win32" and platform_machine == "AMD64")) and extra == "semantic"
 Requires-Dist: torch>=2.6.0; extra == "semantic"
 Requires-Dist: transformers>=5.0.0; extra == "semantic"
 Provides-Extra: milvus
@@ -135,12 +136,12 @@ With semantic code search (UniXcoder embeddings):
 pip install 'code-graph-rag[semantic]'
 ```
 
-Qdrant is the default vector store for semantic search. To use Milvus Lite,
-install `code-graph-rag[semantic,milvus]`, then set
+The vectors go to a SQLite file beside the embedded graph
+(`~/.cgr/vectors.sqlite`), so semantic search needs no server and no Docker
+either. To use Qdrant instead, set `CGR_VECTOR_STORE_BACKEND=qdrant`. To use
+Milvus Lite, install `code-graph-rag[semantic,milvus]`, then set
 `CGR_VECTOR_STORE_BACKEND=milvus` and `MILVUS_URI=./.milvus_code_embeddings.db`
-before indexing. To keep the vectors in a SQLite file beside the embedded graph
-(`~/.cgr/vectors.sqlite`, no server), install `code-graph-rag[semantic,sqlite-vec]`
-and set `CGR_VECTOR_STORE_BACKEND=sqlite`.
+before indexing.
 
 To compute embeddings on an OpenAI-compatible endpoint (OpenAI, Ollama, vLLM)
 instead of locally, set `CGR_EMBEDDING_PROVIDER=openai` with

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/PYPI_README.md

@@ -42,12 +42,12 @@ With semantic code search (UniXcoder embeddings):
 pip install 'code-graph-rag[semantic]'
 ```
 
-Qdrant is the default vector store for semantic search. To use Milvus Lite,
-install `code-graph-rag[semantic,milvus]`, then set
+The vectors go to a SQLite file beside the embedded graph
+(`~/.cgr/vectors.sqlite`), so semantic search needs no server and no Docker
+either. To use Qdrant instead, set `CGR_VECTOR_STORE_BACKEND=qdrant`. To use
+Milvus Lite, install `code-graph-rag[semantic,milvus]`, then set
 `CGR_VECTOR_STORE_BACKEND=milvus` and `MILVUS_URI=./.milvus_code_embeddings.db`
-before indexing. To keep the vectors in a SQLite file beside the embedded graph
-(`~/.cgr/vectors.sqlite`, no server), install `code-graph-rag[semantic,sqlite-vec]`
-and set `CGR_VECTOR_STORE_BACKEND=sqlite`.
+before indexing.
 
 To compute embeddings on an OpenAI-compatible endpoint (OpenAI, Ollama, vLLM)
 instead of locally, set `CGR_EMBEDDING_PROVIDER=openai` with

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/README.md

@@ -61,8 +61,7 @@ Code-Graph-RAG parses a multi-language codebase with Tree-sitter, sharpened by c
 ## Latest News 🔥
 
 <!-- SECTION:latest_news -->
-- **Dependency Locking**: Locked `opentelemetry-api` to version 1.45.1 for stability.
-- **GHSA Security Fix**: Addressed GHSA-xcq7-p3pf-p88h vulnerability.
+- **SBOM**: CycloneDX SBOM and Docker image attestation published with each release.
 <!-- /SECTION:latest_news -->
 
 See [NEWS.md](NEWS.md) for the full history.

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2/code_graph_rag.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-graph-rag
-Version: 0.2.19
+Version: 0.3.2
 Summary: The ultimate RAG for your monorepo. Query, understand, and edit multi-language codebases with the power of AI and knowledge graphs
 License-Expression: MIT
 Keywords: rag,retrieval-augmented-generation,knowledge-graph,code-analysis,tree-sitter,mcp,mcp-server,llm,graph-database,semantic-search,codebase,memgraph,developer-tools,monorepo
@@ -78,6 +78,7 @@ Requires-Dist: tree-sitter-sql>=0.3.11; extra == "treesitter-full"
 Requires-Dist: tree-sitter-markdown>=0.5.1; extra == "treesitter-full"
 Provides-Extra: semantic
 Requires-Dist: qdrant-client>=1.9.0; extra == "semantic"
+Requires-Dist: sqlite-vec>=0.1.9; ((sys_platform == "linux" and (platform_machine == "x86_64" or platform_machine == "aarch64")) or sys_platform == "darwin" or (sys_platform == "win32" and platform_machine == "AMD64")) and extra == "semantic"
 Requires-Dist: torch>=2.6.0; extra == "semantic"
 Requires-Dist: transformers>=5.0.0; extra == "semantic"
 Provides-Extra: milvus
@@ -135,12 +136,12 @@ With semantic code search (UniXcoder embeddings):
 pip install 'code-graph-rag[semantic]'
 ```
 
-Qdrant is the default vector store for semantic search. To use Milvus Lite,
-install `code-graph-rag[semantic,milvus]`, then set
+The vectors go to a SQLite file beside the embedded graph
+(`~/.cgr/vectors.sqlite`), so semantic search needs no server and no Docker
+either. To use Qdrant instead, set `CGR_VECTOR_STORE_BACKEND=qdrant`. To use
+Milvus Lite, install `code-graph-rag[semantic,milvus]`, then set
 `CGR_VECTOR_STORE_BACKEND=milvus` and `MILVUS_URI=./.milvus_code_embeddings.db`
-before indexing. To keep the vectors in a SQLite file beside the embedded graph
-(`~/.cgr/vectors.sqlite`, no server), install `code-graph-rag[semantic,sqlite-vec]`
-and set `CGR_VECTOR_STORE_BACKEND=sqlite`.
+before indexing.
 
 To compute embeddings on an OpenAI-compatible endpoint (OpenAI, Ollama, vLLM)
 instead of locally, set `CGR_EMBEDDING_PROVIDER=openai` with

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/code_graph_rag.egg-info/requires.txt

@@ -47,6 +47,9 @@ qdrant-client>=1.9.0
 torch>=2.6.0
 transformers>=5.0.0
 
+[semantic:(sys_platform == "linux" and (platform_machine == "x86_64" or platform_machine == "aarch64")) or sys_platform == "darwin" or (sys_platform == "win32" and platform_machine == "AMD64")]
+sqlite-vec>=0.1.9
+
 [sqlite-vec]
 sqlite-vec>=0.1.9
 

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/codebase_rag/cli.py

@@ -72,7 +72,7 @@ from .services.provenance import (
 )
 from .stack import StackManager
 from .stack.cli import cli as daemon_cli
-from .stack.constants import SERVICE_QDRANT, StackState
+from .stack.constants import SERVICE_MEMGRAPH, SERVICE_QDRANT, StackState
 from .stack.manager import StackError
 from .sync_guard import RUNNING_SYNC_TOKENS as _RUNNING_SYNC_TOKENS
 from .sync_guard import OwnerCheck as _OwnerCheck
@@ -775,15 +775,25 @@ def _resolve_active_projects(projects: str | None, default_project: str) -> list
     return [default_project]
 
 
-def _maybe_start_stack() -> None:
+def _maybe_start_stack(skip_embeddings: bool = False) -> None:
     # The Docker stack serves Memgraph, and Qdrant when the vector settings
     # pick the bundled one. The embedded graph is a file and Neo4j is the
     # user's own server, so for them the stack runs only for the vectors:
     # only its Qdrant is started and checked, since every Memgraph check
-    # needs a driver those installs need not have.
+    # needs a driver those installs need not have. A run that writes no
+    # embeddings, or writes them to another backend, needs no Qdrant, so it
+    # never starts one: for the embedded graph that means no Docker at all.
+    wants_vectors = not (skip_embeddings or settings.SKIP_EMBEDDINGS)
     if settings.GRAPH_BACKEND == DIALECT_MEMGRAPH:
-        mgr = StackManager()
+        mgr = (
+            StackManager()
+            if wants_vectors
+            and settings.VECTOR_STORE_BACKEND == cs.VectorStoreBackend.QDRANT
+            else StackManager(services=(SERVICE_MEMGRAPH,))
+        )
     else:
+        if not wants_vectors:
+            return
         mgr = StackManager(services=(SERVICE_QDRANT,))
         if not mgr.serves_vectors():
             return
@@ -1479,7 +1489,7 @@ def start(
     _refuse_empty_workspace(workspace_config, update_graph, active_projects)
 
     if not no_start_stack:
-        _maybe_start_stack()
+        _maybe_start_stack(skip_embeddings=no_embeddings)
 
     effective_batch_size = settings.resolve_batch_size(batch_size)
 

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/codebase_rag/config.py

@@ -452,8 +452,11 @@ class AppConfig(BaseSettings):
     QDRANT_UPSERT_RETRIES: int = Field(default=3, gt=0)
     QDRANT_RETRY_BASE_DELAY: float = Field(default=0.5, gt=0)
     QDRANT_BATCH_SIZE: int = Field(default=50, gt=0)
+    # The `sqlite` store is the default for the reason the embedded graph is:
+    # a file under CGR_HOME needs no server and no Docker. Qdrant and Milvus
+    # stay one setting away.
     VECTOR_STORE_BACKEND: cs.VectorStoreBackend = Field(
-        cs.VectorStoreBackend.QDRANT, validation_alias="CGR_VECTOR_STORE_BACKEND"
+        cs.VectorStoreBackend.SQLITE, validation_alias="CGR_VECTOR_STORE_BACKEND"
     )
     MILVUS_URI: str = "./.milvus_code_embeddings.db"
     MILVUS_TOKEN: str | None = None
@@ -469,6 +472,18 @@ class AppConfig(BaseSettings):
     SQLITE_VECTOR_DIM: int = 768
     SQLITE_VECTOR_TOP_K: int = 5
 
+    @property
+    def qdrant_set_without_backend(self) -> bool:
+        """Whether Qdrant is configured but the vector store was left to default.
+
+        Qdrant was the default vector store before the `sqlite` one replaced
+        it, so this is how an install upgraded in place looks: its embeddings
+        live in Qdrant while cgr now opens an empty file.
+        """
+        return cs.SETTING_VECTOR_STORE_BACKEND not in self.model_fields_set and any(
+            name in self.model_fields_set for name in cs.QDRANT_CONNECTION_SETTINGS
+        )
+
     @property
     def vector_db_file(self) -> Path:
         if self.SQLITE_VECTOR_PATH is not None:

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/codebase_rag/constants/core.py

@@ -262,6 +262,9 @@ PRUNE_PENDING_FILENAME = ".cgr-prune-pending"
 # PR #3358). Not in CGR_STATE_FILENAMES: no cgr ever wrote one into the tree,
 # so there is nothing to move out of it.
 EMBEDDINGS_PENDING_PROJECT_TEMPLATE = ".cgr-embeddings-pending-{digest}"
+# The vector store backend a project's points were last settled in, one file
+# per project for the same reason; likewise never written into the tree.
+EMBEDDINGS_STORE_PROJECT_TEMPLATE = ".cgr-embeddings-store-{digest}"
 # The marker an earlier build shared between every project of a checkout: a
 # JSON list of the projects owing a pass, or an empty file. Read so that each
 # still retries once, then split into the per-project files above.

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/codebase_rag/constants/providers.py

@@ -191,6 +191,13 @@ QDRANT_DEFAULT_DB_PATH = "./.qdrant_code_embeddings"
 # The settings field that holds it. Whether it was supplied is read from the
 # fields the settings sources set, since a supplied value can equal the default.
 SETTING_QDRANT_DB_PATH = "QDRANT_DB_PATH"
+# Settings whose presence, with VECTOR_STORE_BACKEND left unset, marks an
+# install that stored its embeddings in Qdrant when Qdrant was the default.
+SETTING_VECTOR_STORE_BACKEND = "VECTOR_STORE_BACKEND"
+QDRANT_CONNECTION_SETTINGS = ("QDRANT_URL", SETTING_QDRANT_DB_PATH, "QDRANT_API_KEY")
+# The backend a project synced before cgr stamped its vector store was embedded
+# into: the default then.
+EMBEDDINGS_STORE_UNSTAMPED = VectorStoreBackend.QDRANT
 
 SEMANTIC_DEPENDENCIES = (
     MODULE_PYMILVUS,

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/codebase_rag/exceptions.py

@@ -119,12 +119,14 @@ SQLITE_VECTOR_DIM_MISMATCH = (
     "or re-index with --clean to rebuild the store with the new size."
 )
 SQLITE_VEC_NOT_INSTALLED = (
-    "CGR_VECTOR_STORE_BACKEND=sqlite needs sqlite-vec, which is not installed. "
-    "Install it with `pip install 'code-graph-rag[sqlite-vec]'` (or "
-    "`uv sync --extra sqlite-vec`)."
+    "The sqlite vector store (the default, CGR_VECTOR_STORE_BACKEND=sqlite) "
+    "needs sqlite-vec, which is not installed. Install it with "
+    "`pip install 'code-graph-rag[sqlite-vec]'` (or `uv sync --extra "
+    "sqlite-vec`); the `semantic` extra includes it."
 )
 SQLITE_VEC_EXTENSIONS_UNSUPPORTED = (
-    "CGR_VECTOR_STORE_BACKEND=sqlite loads the sqlite-vec extension, but this "
+    "The sqlite vector store (the default, CGR_VECTOR_STORE_BACKEND=sqlite) "
+    "loads the sqlite-vec extension, but this "
     "Python's sqlite3 module was built without extension loading. Use a Python "
     "that has it (uv-managed Pythons do), or set CGR_VECTOR_STORE_BACKEND=qdrant."
 )

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/codebase_rag/graph_updater.py

@@ -893,6 +893,40 @@ def _embeddings_marker(state_dir: Path, project_name: str) -> Path:
     return state_dir / cs.EMBEDDINGS_PENDING_PROJECT_TEMPLATE.format(digest=digest)
 
 
+def _embeddings_store_stamp(state_dir: Path, project_name: str) -> Path:
+    """The file naming the vector store backend `project_name`'s points are in.
+
+    A pass that settles the project writes the selected backend there. Once
+    the setting changes, as when `sqlite` replaced Qdrant as the default, the
+    points stay in a store nothing searches, and an unchanged checkout takes
+    the in-sync fast path, which never embeds: until the stamp names the
+    selected backend the project owes a pass. A checkout synced before the
+    stamp existed has none; its points went to the default then, Qdrant, so
+    only an install that now selects another backend owes that pass.
+    """
+    digest = hashlib.sha256(project_name.encode(cs.ENCODING_UTF8)).hexdigest()
+    return state_dir / cs.EMBEDDINGS_STORE_PROJECT_TEMPLATE.format(digest=digest)
+
+
+def _stamp_names_selected_store(stamp: Path) -> bool:
+    try:
+        stamped = stamp.read_text(encoding=cs.ENCODING_UTF8).strip()
+    except FileNotFoundError:
+        stamped = str(cs.EMBEDDINGS_STORE_UNSTAMPED)
+    except OSError:
+        return False
+    return stamped == str(settings.VECTOR_STORE_BACKEND)
+
+
+def _save_embeddings_store(stamp: Path) -> None:
+    try:
+        stamp.write_text(str(settings.VECTOR_STORE_BACKEND), encoding=cs.ENCODING_UTF8)
+    except OSError:
+        # Best effort, as for the pending marker: without the stamp the next
+        # sync runs the pass again, which is slower but loses nothing.
+        logger.debug(ls.EMBEDDINGS_PENDING_NOT_UPDATED)
+
+
 def _save_embeddings_owed(marker: Path, project_name: str, owed: bool) -> bool:
     """Create or remove `project_name`'s marker; say whether that went through."""
     try:
@@ -10101,6 +10135,10 @@ class GraphUpdater:
     def _owes_embeddings(self) -> bool:
         if _embeddings_marker(self.state_dir, self.project_name).exists():
             return True
+        if not _stamp_names_selected_store(
+            _embeddings_store_stamp(self.state_dir, self.project_name)
+        ):
+            return True
         shared = _load_shared_embeddings_owed(
             self.state_dir / cs.EMBEDDINGS_PENDING_FILENAME
         )
@@ -10122,6 +10160,10 @@ class GraphUpdater:
         """
         if pending:
             logger.warning(ls.EMBEDDINGS_PENDING.format(project=self.project_name))
+        else:
+            _save_embeddings_store(
+                _embeddings_store_stamp(self.state_dir, self.project_name)
+            )
         recorded = _save_embeddings_owed(
             _embeddings_marker(self.state_dir, self.project_name),
             self.project_name,

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/codebase_rag/logs.py

@@ -275,6 +275,13 @@ VECTOR_STORE_BACKEND_UNAVAILABLE = (
     "Vector store backend '{backend}' dependencies are not available"
 )
 VECTOR_STORE_BACKEND_UNKNOWN = "Unknown vector store backend '{backend}'"
+QDRANT_SET_WITHOUT_BACKEND = (
+    "Qdrant settings are set but CGR_VECTOR_STORE_BACKEND is not, so cgr is "
+    "using the sqlite vector store at {path}, now the default instead of "
+    "Qdrant. Embeddings stored in Qdrant are not searched here. To keep using "
+    "Qdrant, set CGR_VECTOR_STORE_BACKEND=qdrant; to keep the sqlite store, "
+    "set CGR_VECTOR_STORE_BACKEND=sqlite. Either one silences this notice."
+)
 VECTOR_STORE_DELETE_PROJECT = (
     "Deleting {count} {backend} vectors for project '{project}'"
 )

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/codebase_rag/parsers/go_frontend/frontend.py

@@ -373,14 +373,15 @@ def _build_tool(go: str) -> Path | None:
 
     The version floor is checked before a build so an old Go is named plainly
     instead of surfacing as `go build`'s stderr, and a failed build is not
-    retried on every sync (issue #2395).
+    retried on every sync (issue #2395). It is checked before a cached helper
+    is reused too: the helper drives the `go` on PATH to load packages, and a
+    version manager can put an older Go there after a newer one built it, so
+    a cached build must not let a Go below the floor run (PR #3410 review).
     """
     cache = _cache_dir()
     src = cache / "src"
     out = cache / "out"
     binary = out / _BINARY_NAME
-    if artifact_fresh(binary, _newest_source_mtime):
-        return binary
     version = _toolchain_version(go)
     if (required := _too_old(version)) is not None:
         logger.warning(
@@ -389,6 +390,8 @@ def _build_tool(go: str) -> Path | None:
             )
         )
         return None
+    if artifact_fresh(binary, _newest_source_mtime):
+        return binary
     marker = cache / _BUILD_FAILED_MARKER
     # The fast path: a remembered failure skips the lock, and any wait on it,
     # entirely. `_compile_and_remember` checks again once the lock is held.

code_graph_rag-0.3.2/codebase_rag/parsers/go_frontend/gotypes/go.mod

@@ -0,0 +1,10 @@
+module cgr.dev/gotypes
+
+go 1.27.2
+
+require golang.org/x/tools v0.52.0
+
+require (
+	golang.org/x/mod v0.42.0 // indirect
+	golang.org/x/sync v0.24.0 // indirect
+)

code_graph_rag-0.3.2/codebase_rag/parsers/go_frontend/gotypes/go.sum

@@ -0,0 +1,8 @@
+github.com/google/go-cmp v0.6.0 h1:ofyhxvXcZhMsU5ulbFiLKl/XBFqE1GSq7atu8tAmTRI=
+github.com/google/go-cmp v0.6.0/go.mod h1:17dUlkBOakJ0+DkrSSNjCkIjxS6bF9zb3elmeNGIjoY=
+golang.org/x/mod v0.42.0 h1:ICeL3t2L30ljSPHHxDnR4fqVcBt/zR2zJAb9juavmQg=
+golang.org/x/mod v0.42.0/go.mod h1:u5fzIrRInkL8MaO1cbRkUHT64v303YIT5KEqpNr7jRs=
+golang.org/x/sync v0.24.0 h1:iqiDrGoDgyuvAKrhLQthUEzU/R378N78gXbBjxed7oA=
+golang.org/x/sync v0.24.0/go.mod h1:sUUOizhqBxiL6pEWpqNLUiaJn1ShEbZ6BBqskPbjZm0=
+golang.org/x/tools v0.52.0 h1:ryxg4U63oaUYk7uOLTF37n43f4qgtGvqV4Ioit1BaEs=
+golang.org/x/tools v0.52.0/go.mod h1:ItcP5UNaVGG+Yc4eIYmaFLNJ4soEopkx5Y2d4lcEurs=

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/codebase_rag/utils/dependencies.py

@@ -1,7 +1,11 @@
 from __future__ import annotations
 
 import importlib.util
+from functools import cache
 
+from loguru import logger
+
+from codebase_rag import exceptions as ex
 from codebase_rag.constants import (
     MODULE_AST_GREP,
     MODULE_PYMILVUS,
@@ -45,6 +49,27 @@ def has_sqlite_vec() -> bool:
     return _check_dependency(MODULE_SQLITE_VEC)
 
 
+def sqlite_vec_loadable() -> bool:
+    """Whether this Python can load sqlite-vec, which is installed.
+
+    It loads as a SQLite extension, which some Python builds (pyenv's default
+    among them) cannot load. The `sqlite` store is then unavailable, as
+    without the package: semantic search is off rather than failing in every
+    sync and on `--clean`, and the reason is logged once.
+    """
+    from codebase_rag.sqlite_vectors import extensions_loadable
+
+    if extensions_loadable():
+        return True
+    _warn_sqlite_extensions_unsupported()
+    return False
+
+
+@cache
+def _warn_sqlite_extensions_unsupported() -> None:
+    logger.warning(ex.SQLITE_VEC_EXTENSIONS_UNSUPPORTED)
+
+
 def has_ast_grep() -> bool:
     return _check_dependency(MODULE_AST_GREP)
 
@@ -56,7 +81,7 @@ def has_vector_store_dependencies() -> bool:
     if backend == VectorStoreBackend.MILVUS:
         return has_pymilvus()
     if backend == VectorStoreBackend.SQLITE:
-        return has_sqlite_vec()
+        return has_sqlite_vec() and sqlite_vec_loadable()
     return has_qdrant_client()
 
 

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/codebase_rag/vector_store.py

@@ -5,6 +5,7 @@ import json
 import time
 import uuid
 from collections.abc import Callable, Collection, Iterable, Iterator, Mapping, Sequence
+from functools import cache
 from importlib.metadata import PackageNotFoundError, version
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, Protocol, Required, TypedDict, cast
@@ -31,7 +32,12 @@ from .sqlite_vectors import (
     open_sqlite_vector_client,
 )
 from .types_defs import EmbeddingPoint, EmbeddingSymbol
-from .utils.dependencies import has_pymilvus, has_qdrant_client, has_sqlite_vec
+from .utils.dependencies import (
+    has_pymilvus,
+    has_qdrant_client,
+    has_sqlite_vec,
+    sqlite_vec_loadable,
+)
 
 if TYPE_CHECKING:
     from qdrant_client.models import Filter, Record
@@ -233,6 +239,8 @@ def _get_vector_store() -> VectorStore | None:
         if not has_sqlite_vec():
             logger.warning(ls.VECTOR_STORE_BACKEND_UNAVAILABLE.format(backend=backend))
             return None
+        if not sqlite_vec_loadable():
+            return None
         return SqliteVectorStore()
     return None
 
@@ -486,6 +494,13 @@ def _ensure_sqlite_collection(client: SqliteVectorClient, validate: bool) -> Non
         )
 
 
+@cache
+def _note_qdrant_set_without_backend() -> None:
+    # Once per process, as the graph's own notice: the client reopens after
+    # every close, and the setting cannot change under it.
+    logger.warning(ls.QDRANT_SET_WITHOUT_BACKEND.format(path=settings.vector_db_file))
+
+
 def get_sqlite_vector_client(validate: bool = True) -> SqliteVectorClient:
     global _CLIENT, _CLIENT_BACKEND
     if not has_sqlite_vec():
@@ -493,6 +508,8 @@ def get_sqlite_vector_client(validate: bool = True) -> SqliteVectorClient:
 
     _ensure_client_backend(VectorStoreBackend.SQLITE)
     if _CLIENT is None:
+        if settings.qdrant_set_without_backend:
+            _note_qdrant_set_without_backend()
         client = open_sqlite_vector_client(settings.vector_db_file)
         try:
             _ensure_sqlite_collection(client, validate)

{code_graph_rag-0.2.19 → code_graph_rag-0.3.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "code-graph-rag"
-version = "0.2.19"
+version = "0.3.2"
 description = "The ultimate RAG for your monorepo. Query, understand, and edit multi-language codebases with the power of AI and knowledge graphs"
 readme = "PYPI_README.md"
 requires-python = ">=3.12"
@@ -193,8 +193,13 @@ treesitter-full = [
     "tree-sitter-markdown>=0.5.1",
 ]
 
+# sqlite-vec is the default vector store's, on the platforms it ships wheels
+# for (it has no sdist), which cover every one torch ships wheels for; no
+# marker tells musl from glibc, and torch has no musl wheels either.
+# qdrant-client is for CGR_VECTOR_STORE_BACKEND=qdrant, the default before it.
 semantic = [
     "qdrant-client>=1.9.0",
+    "sqlite-vec>=0.1.9; (sys_platform == 'linux' and (platform_machine == 'x86_64' or platform_machine == 'aarch64')) or sys_platform == 'darwin' or (sys_platform == 'win32' and platform_machine == 'AMD64')",
     "torch>=2.6.0",
     "transformers>=5.0.0",
 ]
@@ -203,8 +208,9 @@ milvus = [
     "pymilvus[milvus-lite]>=3.0.0",
 ]
 
-# The `sqlite` vector store (CGR_VECTOR_STORE_BACKEND=sqlite): vectors in a
-# file beside the embedded graph, so semantic search needs no server either.
+# The default `sqlite` vector store alone, for an install that computes its
+# embeddings elsewhere (CGR_EMBEDDING_PROVIDER=openai): vectors in a file
+# beside the embedded graph, so semantic search needs no server either.
 sqlite-vec = [
     "sqlite-vec>=0.1.9",
 ]

code_graph_rag-0.2.19/codebase_rag/parsers/go_frontend/gotypes/go.mod

@@ -1,10 +0,0 @@
-module cgr.dev/gotypes
-
-go 1.23.12
-
-require golang.org/x/tools v0.30.0
-
-require (
-	golang.org/x/mod v0.23.0 // indirect
-	golang.org/x/sync v0.11.0 // indirect
-)

code_graph_rag-0.2.19/codebase_rag/parsers/go_frontend/gotypes/go.sum

@@ -1,8 +0,0 @@
-github.com/google/go-cmp v0.6.0 h1:ofyhxvXcZhMsU5ulbFiLKl/XBFqE1GSq7atu8tAmTRI=
-github.com/google/go-cmp v0.6.0/go.mod h1:17dUlkBOakJ0+DkrSSNjCkIjxS6bF9zb3elmeNGIjoY=
-golang.org/x/mod v0.23.0 h1:Zb7khfcRGKk+kqfxFaP5tZqCnDZMjC5VtUBs87Hr6QM=
-golang.org/x/mod v0.23.0/go.mod h1:6SkKJ3Xj0I0BrPOZoBy3bdMptDDU9oJrpohJ3eWZ1fY=
-golang.org/x/sync v0.11.0 h1:GGz8+XQP4FvTTrjZPzNKTMFtSXH80RAzG+5ghFPgK9w=
-golang.org/x/sync v0.11.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
-golang.org/x/tools v0.30.0 h1:BgcpHewrV5AUp2G9MebG4XPFI1E2W41zU1SaqVA9vJY=
-golang.org/x/tools v0.30.0/go.mod h1:c347cR/OJfw5TI+GfX7RUPNMdDRRbjvYTS0jPyvsVtY=