包package agentscope.app.rag

agentscope.app.rag

module agentscope.app.rag

agentscope.app.rag 属于RAG 与知识库，切分、向量化、存储与检索。该包共收录 12 个公开类型。

This subpackage groups every RAG-specific service-layer concept under one roof so the user-facing import surface stays compact:

:mod:.blob_store — backends storing uploaded document bytes;
:mod:.knowledge_base_manager — knowledge base lifecycle and the runtime :class:~agentscope.rag.KnowledgeBase handle shared between HTTP and agents;
:mod:.index_worker — out-of-process indexing worker entry point.

Commonly used symbols are re-exported here so the typical user import collapses to a single line::

from agentscope.app.rag import (
    LocalBlobStore,
    S3BlobStore,
    CollectionPerKbManager,
    run_worker,
)

For type-level imports or rare subclasses, use the submodules directly.

源码Source agentscope/app/rag/__init__.py:1

子包Subpackages

类Classes

`AsyncReadable` 类class	异步可读，把私有文档转化为可检索的知识。 Minimal async-read protocol returned by :meth:`BlobStoreBase.open`.
`BlobStoreBase` 类class	二进制对象存储基类，把私有文档转化为可检索的知识。 Abstract base class for blob storage backends.
`CollectionPerKbManager` 类class	集合每个知识库管理器，把私有文档转化为可检索的知识。 One-collection-per-KB knowledge base manager.
`DimensionPolicy` 类class	维度策略，把私有文档转化为可检索的知识。 The dimension policy a manager exposes to the front-end.
`DimensionPolicyError` 类class	维度策略错误，把私有文档转化为可检索的知识。 Raised when a create-time embedding model violates the manager's dimension policy.
`DimensionPolicyKind` 类class	维度策略种类，把私有文档转化为可检索的知识。 The kind of dimension policy the manager publishes.
`KnowledgeBaseError` 类class	知识库基类管理器错误的基类。 Base class for knowledge base manager errors.
`KnowledgeBaseManagerBase` 类class	知识库管理器基类，把私有文档转化为可检索的知识。 Abstract base for knowledge base managers.
`KnowledgeBaseNotFoundError` 类class	知识库未找到错误，把私有文档转化为可检索的知识。 Raised when a knowledge base record cannot be located.
`LocalBlobStore` 类class	本地二进制对象存储，负责持久化与读取数据。 Store blobs as files beneath a configurable root directory.
`S3BlobStore` 类class	S 3 二进制对象存储，负责持久化与读取数据。 Store blobs in an S3-compatible bucket.

函数Functions

run_worker

函数function

运行索引 worker

Run the out-of-process index worker until cancelled.

函数详情Function Details

run_worker

async def run_worker(*, storage: 'StorageBase', message_bus: 'MessageBus', blob_store: 'BlobStoreBase', knowledge_base_manager: 'KnowledgeBaseManagerBase', parsers: 'list[ParserBase] | dict[str, ParserBase]', chunkers: 'list[type[ChunkerBase]] | None' = None, node_id: str | None = None, worker_max_concurrency: int = 4, consumer_max_batch: int = 32, parser_executor: ProcessPoolExecutor | None = None, **kwargs: Any) -> None

async

持续消费索引任务，直到被取消；适合把文档向量化放到独立进程部署。

Manages the lifecycle of every passed-in backend through a single :class:AsyncExitStack.

参数Parameters

名称Name	类型Type	说明Description
`storage`kw	'StorageBase'	存储（类型 `'StorageBase'`）。Persistent storage backend. Lifecycle managed here.
`message_bus`kw	'MessageBus'	消息总线（类型 `'MessageBus'`）。Live message bus subscribed to for index-task signals. Lifecycle managed here.
`blob_store`kw	'BlobStoreBase'	二进制对象存储（类型 `'BlobStoreBase'`）。Backend the worker reads document bytes from. Lifecycle managed here.
`knowledge_base_manager`kw	'KnowledgeBaseManagerBase'	知识库管理器（类型 `'KnowledgeBaseManagerBase'`）。Resolves :class:`KnowledgeBase` runtimes for embedding + vector store writes. Lifecycle managed here.
`parsers`kw	'list[ParserBase] \| dict[str, ParserBase]'	parsers（类型 `'list[ParserBase] \| dict[str, ParserBase]'`）。Parsers used to dispatch uploads by IANA media type. Pass the same registry both to `create_app` and to the worker — list mode expands each parser's `supported_media_types` (later entries override earlier ones, with a warning); dict mode is used verbatim for explicit routing.
`chunkers`kw默认default `None`	'list[type[ChunkerBase]] \| None'	chunkers（类型 `'list[type[ChunkerBase]] \| None'`）。The chunker classes that can be rebuilt from a knowledge base's `chunker_config`. Defaults to `[ApproxTokenChunker]`.
`node_id`kw默认default `None`	str \| None	节点标识（类型 `str \| None`）。Stable identifier for this worker process used on the storage lease. Defaults to `"{hostname}:{uuid-prefix}"` — good enough for distinguishing two workers running on the same host; override when your orchestrator has a more meaningful id (Kubernetes pod name, ECS task id, ...).
`worker_max_concurrency`kw默认default `4`	int	工作进程最大 concurrency（类型 `int`）。Maximum number of documents the worker processes concurrently. See :class:`~agentscope.app._service.IndexWorker`.
`consumer_max_batch`kw默认default `32`	int	消费者最大 batch（类型 `int`）。Maximum entries the consumer drains per signal. See :class:`~agentscope.app._service.IndexTaskConsumer`.
`parser_executor`kw默认default `None`	ProcessPoolExecutor \| None	解析器执行器（类型 `ProcessPoolExecutor \| None`）。Process pool for CPU-bound parses. `None` runs parses inline (fine for text-only deployments).
`kwargs`	Any	关键字参数（类型 `Any`）。The kwargs.

返回Returns None

无返回值。No value is returned.

源码Source agentscope/app/rag/index_worker/__init__.py:80–203