feat(sqlite)!: replace Vec1 with bundled sqlite-vec

docs/en/development/memory-layer.md

@@ -90,20 +90,14 @@ the same citation fields through list and exact-read operations.
 unavailable. Explicit `vector` and `hybrid` requests fail when the configured profile does not provide that
 capability.
 
-## Enable SQLite Vec1
+## Enable SQLite vector search
 
-SQLite vector search is enabled only when both a Vec1 0.7 or newer loadable extension and an embedding model are
-supplied. PowerContext does not install or build the native extension; provide a compatible library for the target
-operating system and architecture:
+SQLite vector search is enabled when an embedding model is supplied. The `powercontext[builtin]` extra bundles
+`sqlite-vec`, so no extension path or separate native-library installation is required:
 
 ```python
-from pathlib import Path
-
 config = BuiltinConfig(
-    database=SQLiteConfig(
-        url="sqlite+aiosqlite:///powercontext.db",
-        vec1_extension=Path("/opt/sqlite-extensions/vec1"),
-    )
+    database=SQLiteConfig(url="sqlite+aiosqlite:///powercontext.db")
 )
 async with open_builtin_runtime(
     config,
@@ -112,7 +106,7 @@ async with open_builtin_runtime(
     ...
 ```
 
-The SQLite profile composes FTS5 and Vec1 strategies. It reports `fts`, `vector`, and `hybrid` through Memory
+The SQLite profile composes FTS5 and sqlite-vec strategies. It reports `fts`, `vector`, and `hybrid` through Memory
 capabilities.
 Stored projections and query vectors must use the same `EmbeddingProfile`, including model name, dimension, distance,
 and normalization. Changing that profile requires rebuilding projections before vector search resumes.
@@ -145,7 +139,7 @@ async with open_builtin_runtime(
 
 The OceanBase profile uses the same index composition as SQLite. Its full-text strategy is always available. Supplying
 an embedding model adds a `VECTOR` projection and HNSW strategy, enabling `vector` and `hybrid` modes. SQLite FTS5 and
-OceanBase FULLTEXT therefore serve the same Runtime and Server search calls; Vec1 and HNSW do the same for vector
+OceanBase FULLTEXT therefore serve the same Runtime and Server search calls; sqlite-vec and HNSW do the same for vector
 search.
 
 ## Operational checks
@@ -155,7 +149,7 @@ Before serving requests, verify:
 - the selected profile opens and initializes successfully;
 - each tenant or project maps to the intended scope ID;
 - scheduled extraction has a candidate pipeline;
-- Vec1 configuration includes a matching embedding model;
+- SQLite vector search has a matching embedding model;
 - OceanBase vector search has a matching embedding model;
 - capability responses match the indexes actually initialized;
 - database and scheduler resources close with the process lifecycle.

docs/en/development/pydantic-ai-inference.md

@@ -33,15 +33,14 @@ Vector search needs the embedding model and its complete deployment profile:
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_MODEL="provider:embedding-model"
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_PROFILE_ID="project-embedding-v1"
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_DIMENSION="1536"
-export POWERCONTEXT_SERVER_DATABASE_VEC1_EXTENSION="/opt/sqlite-extensions/vec1"
 ```
 
 Provider credentials remain in the environment variables understood by the selected Pydantic AI provider. They are
 not fields on PowerContext models.
 
 The Server rejects a partial embedding profile. `embedding_model`, `embedding_profile_id`, and `embedding_dimension`
-must be configured together. Vec1 also requires that embedding configuration because the index dimension and stored
-vectors must agree.
+must be configured together. SQLite vector search uses that embedding configuration because the index dimension and
+stored vectors must agree.
 
 ## Compose generation directly
 
@@ -99,9 +98,9 @@ embedding_model = PydanticAIEmbeddingModel(
 )
 ```
 
-Pass this adapter to `open_builtin_contexts()` or `open_builtin_runtime()` with a `SQLiteConfig` that selects the Vec1
-extension. The adapter verifies output count, order, dimension, and finite numeric values, then applies the declared
-unit normalization before vectors reach persistence.
+Pass this adapter to `open_builtin_contexts()` or `open_builtin_runtime()` with a `SQLiteConfig`. The bundled
+sqlite-vec index is enabled automatically. The adapter verifies output count, order, dimension, and finite numeric
+values, then applies the declared unit normalization before vectors reach persistence.
 
 An `EmbeddingProfile` is a deployment contract, not descriptive metadata. Stored projections and query embeddings
 must use the same profile. When the model, dimension, or normalization changes, rebuild Memory projections from the

docs/en/development/remote-access-implementation.md

@@ -56,7 +56,7 @@ export POWERCONTEXT_SERVER_DATABASE_KIND="oceanbase"
 export POWERCONTEXT_SERVER_DATABASE_URL="mysql+aoceanbase://user:password@host:2881/powercontext?charset=utf8mb4"
 ```
 
-Both database choices expose full-text search through the same Server API. With an embedding model, SQLite uses Vec1
+Both database choices expose full-text search through the same Server API. With an embedding model, SQLite uses sqlite-vec
 and OceanBase uses HNSW for `vector` and `hybrid` searches.
 
 Inference configuration is documented in [Configure Pydantic AI inference](pydantic-ai-inference.md).

docs/en/docs/reference/configuration.md

@@ -178,33 +178,29 @@ Optional settings are `POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_NORMALIZATION` an
 
 Embedding normalization defaults to `unit`.
 
-### SQLite Vec1
+### SQLite vector search
 
-SQLite vector and hybrid search additionally require a
-[SQLite Vec1](https://sqlite.org/vec1/doc/trunk/doc/vec1.md) 0.7 or newer loadable extension. PowerContext does not
-download, build, or update this native library. Obtain it for the Server's operating system and architecture, then
-set its path together with the complete embedding profile:
+SQLite vector and hybrid search use [sqlite-vec](https://alexgarcia.xyz/sqlite-vec/), which is bundled with the
+`powercontext[builtin]` dependency set. Configure the complete embedding profile; no extension path is needed:
 
 ```bash
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_MODEL=provider:embedding-model
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_PROFILE_ID=embedding-model-v1
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_DIMENSION=1024
 export POWERCONTEXT_SERVER_DATABASE_URL=sqlite+aiosqlite:////srv/powercontext/powercontext.db
-export POWERCONTEXT_SERVER_DATABASE_VEC1_EXTENSION=/opt/sqlite-extensions/vec1
 powercontext server run
 ```
 
-The extension path must identify a library that the SQLite loader can open. PowerContext loads and probes the
-extension when the Server opens the database; startup fails if the library is incompatible or older than 0.7.
+PowerContext loads and probes the bundled extension when the Server opens the database. Startup fails if the package
+does not contain a library compatible with the current platform or SQLite build.
 
 In another terminal, confirm that the initialized runtime reports vector and hybrid search:
 
 ```bash
 powercontext capabilities
 ```
 
-If Vec1 is unavailable, leave `POWERCONTEXT_SERVER_DATABASE_VEC1_EXTENSION` unset. SQLite full-text search remains
-available without an embedding model or native extension.
+SQLite full-text search remains available when no embedding model is configured.
 
 ## CLI Server connection
 

docs/zh/development/memory-layer.md

@@ -88,19 +88,14 @@ operation 返回相同的 citation 字段。
 `mode="auto"` 会选择当前可用的最强模式，并可在 query embedding 暂时不可用时回退到 FTS。显式请求 `vector`
 或 `hybrid` 时，如果 profile 没有提供相应能力，操作会失败。
 
-## 启用 SQLite Vec1
+## 启用 SQLite 向量检索
 
-只有同时提供 0.7 或更高版本的 Vec1 loadable extension 和 embedding model，SQLite 才会启用向量检索。
-PowerContext 不负责安装或构建这个 native extension；请提供适用于目标操作系统和架构的 library：
+提供 embedding model 后，SQLite 会启用向量检索。`powercontext[builtin]` 已捆绑 `sqlite-vec`，无需配置 extension
+路径或单独安装 native library：
 
 ```python
-from pathlib import Path
-
 config = BuiltinConfig(
-    database=SQLiteConfig(
-        url="sqlite+aiosqlite:///powercontext.db",
-        vec1_extension=Path("/opt/sqlite-extensions/vec1"),
-    )
+    database=SQLiteConfig(url="sqlite+aiosqlite:///powercontext.db")
 )
 async with open_builtin_runtime(
     config,
@@ -109,7 +104,7 @@ async with open_builtin_runtime(
     ...
 ```
 
-SQLite profile 会组合 FTS5 和 Vec1 strategy，并通过 Memory capabilities 报告 `fts`、`vector` 和 `hybrid`。持久化
+SQLite profile 会组合 FTS5 和 sqlite-vec strategy，并通过 Memory capabilities 报告 `fts`、`vector` 和 `hybrid`。持久化
 projection 与 query vector 必须使用同一个 `EmbeddingProfile`，包括 model name、dimension、distance 和
 normalization。更换 profile 后，应先重建 projection，再恢复 vector search。
 
@@ -141,7 +136,7 @@ async with open_builtin_runtime(
 
 OceanBase profile 与 SQLite 使用相同的 index 组合方式。全文 strategy 始终可用；提供 embedding model 后，会增加
 `VECTOR` projection 和 HNSW strategy，并启用 `vector` 与 `hybrid` mode。SQLite FTS5 与 OceanBase FULLTEXT
-服务于同一组 Runtime 和 Server search 调用，Vec1 与 HNSW 也通过同一接口提供向量检索。
+服务于同一组 Runtime 和 Server search 调用，sqlite-vec 与 HNSW 也通过同一接口提供向量检索。
 
 ## 运行检查
 
@@ -150,7 +145,7 @@ OceanBase profile 与 SQLite 使用相同的 index 组合方式。全文 strateg
 - 所选 profile 能够成功打开并完成初始化；
 - 每个 tenant 或 project 映射到预期的 scope ID；
 - 定时 extraction 已经配置 candidate pipeline；
-- Vec1 配置包含匹配的 embedding model；
+- SQLite vector search 配置了匹配的 embedding model；
 - OceanBase vector search 配置了匹配的 embedding model；
 - capability response 与实际初始化的 index 一致；
 - database 和 scheduler 资源会随进程生命周期关闭。

docs/zh/development/pydantic-ai-inference.md

@@ -32,13 +32,12 @@ vector search 需要 embedding model 和完整的 deployment profile：
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_MODEL="provider:embedding-model"
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_PROFILE_ID="project-embedding-v1"
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_DIMENSION="1536"
-export POWERCONTEXT_SERVER_DATABASE_VEC1_EXTENSION="/opt/sqlite-extensions/vec1"
 ```
 
 provider credential 仍使用所选 Pydantic AI provider 支持的环境变量，不属于 PowerContext model 字段。
 
 Server 会拒绝不完整的 embedding profile。`embedding_model`、`embedding_profile_id` 和
-`embedding_dimension` 必须一起配置。Vec1 也依赖这组配置，因为 index dimension 必须与持久化向量一致。
+`embedding_dimension` 必须一起配置。SQLite vector search 使用这组配置，因为 index dimension 必须与持久化向量一致。
 
 ## 直接组合 generation
 
@@ -95,8 +94,8 @@ embedding_model = PydanticAIEmbeddingModel(
 )
 ```
 
-将这个 adapter 传给 `open_builtin_contexts()` 或 `open_builtin_runtime()`，并通过 `SQLiteConfig` 选择 Vec1
-extension。向量进入持久化之前，adapter 会校验输出数量、顺序、dimension 和数值有效性，并执行 profile 声明的单位归一化。
+将这个 adapter 与 `SQLiteConfig` 一起传给 `open_builtin_contexts()` 或 `open_builtin_runtime()` 后，会自动启用捆绑的
+sqlite-vec index。向量进入持久化之前，adapter 会校验输出数量、顺序、dimension 和数值有效性，并执行 profile 声明的单位归一化。
 
 `EmbeddingProfile` 是 deployment contract，不是描述性 metadata。持久化 projection 和 query embedding 必须使用
 同一个 profile。model、dimension 或 normalization 发生变化后，应从权威 Memory revision 重建 projection。

docs/zh/development/remote-access-implementation.md

@@ -54,7 +54,7 @@ export POWERCONTEXT_SERVER_DATABASE_KIND="oceanbase"
 export POWERCONTEXT_SERVER_DATABASE_URL="mysql+aoceanbase://user:password@host:2881/powercontext?charset=utf8mb4"
 ```
 
-两种 database 都通过同一组 Server API 提供全文检索。配置 embedding model 后，SQLite 使用 Vec1，OceanBase
+两种 database 都通过同一组 Server API 提供全文检索。配置 embedding model 后，SQLite 使用 sqlite-vec，OceanBase
 使用 HNSW 提供 `vector` 和 `hybrid` 检索。
 
 inference 配置见[配置 Pydantic AI 推理](pydantic-ai-inference.md)。

docs/zh/docs/reference/configuration.md

@@ -172,33 +172,29 @@ export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_DIMENSION=1024
 
 Embedding normalization 默认为 `unit`。
 
-### SQLite Vec1
+### SQLite 向量检索
 
-SQLite vector 和 hybrid search 还需要 0.7 或更高版本的
-[SQLite Vec1](https://sqlite.org/vec1/doc/trunk/doc/vec1.md) loadable extension。PowerContext 不负责下载、构建或更新
-这个 native library。请先获取适用于 Server 操作系统和架构的构建产物，再同时配置 extension 路径和完整的
-embedding profile：
+SQLite vector 和 hybrid search 使用 [sqlite-vec](https://alexgarcia.xyz/sqlite-vec/)，它已包含在
+`powercontext[builtin]` 依赖中。只需配置完整的 embedding profile，无需配置 extension 路径：
 
 ```bash
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_MODEL=provider:embedding-model
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_PROFILE_ID=embedding-model-v1
 export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_DIMENSION=1024
 export POWERCONTEXT_SERVER_DATABASE_URL=sqlite+aiosqlite:////srv/powercontext/powercontext.db
-export POWERCONTEXT_SERVER_DATABASE_VEC1_EXTENSION=/opt/sqlite-extensions/vec1
 powercontext server run
 ```
 
-extension 路径必须指向 SQLite loader 可以打开的 library。Server 打开数据库时，PowerContext 会加载并探测该
-extension；如果 library 不兼容或版本低于 0.7，启动会失败。
+Server 打开数据库时，PowerContext 会加载并探测捆绑的 extension；如果当前 platform 或 SQLite build 与 package
+中的 library 不兼容，启动会失败。
 
 在另一个终端确认初始化后的 Runtime 已报告 vector 和 hybrid search：
 
 ```bash
 powercontext capabilities
 ```
 
-如果没有可用的 Vec1，请不要设置 `POWERCONTEXT_SERVER_DATABASE_VEC1_EXTENSION`。即使没有 embedding model 或
-native extension，SQLite full-text search 仍然可用。
+没有配置 embedding model 时，SQLite full-text search 仍然可用。
 
 ## CLI Server 连接
 

pyproject.toml

@@ -43,6 +43,7 @@ builtin = [
     "pydantic-settings>=2.7,<3",
     "pyobvector>=0.2.28,<0.3",
     "sqlalchemy[asyncio]>=2,<3",
+    "sqlite-vec>=0.1.9,<0.2",
 ]
 client = [
     "httpx[socks]>=0.28,<1",