Spaces:
Paused
Paused
File size: 10,117 Bytes
9792ea7 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 | # -*- coding: utf-8 -*-
"""Abstract knowledge base manager.
The manager is the **lifecycle owner** of knowledge bases:
- it creates / lists / deletes :class:`KnowledgeBaseRecord` rows in
storage,
- it allocates / drops the matching vector store collections,
- it resolves an embedding model from the record's credential and
hands a ready-to-use :class:`KnowledgeBase` runtime back to callers.
Different subclasses encode different *isolation strategies*: one
collection per knowledge base, a single shared collection scoped by
metadata, native VDB namespaces, etc. All of them share the same
:class:`KnowledgeBaseManagerBase` interface so the rest of the
application can stay strategy-agnostic.
The manager is created once at application startup and stored on
``app.state.knowledge_base_manager``. See
:func:`~agentscope.app.create_app` for the wiring.
"""
from abc import ABC, abstractmethod
from typing import TYPE_CHECKING, Self
from ._dimension_policy import DimensionPolicy
if TYPE_CHECKING:
from types import TracebackType
from ...storage import (
EmbeddingModelConfig,
KnowledgeBaseRecord,
StorageBase,
)
from ....rag import KnowledgeBase, VectorStoreBase
class KnowledgeBaseManagerBase(ABC):
"""Abstract base for knowledge base managers.
Subclasses implement a specific isolation strategy by overriding
:meth:`create_knowledge_base`, :meth:`delete_knowledge_base`, and
:meth:`get_knowledge`. The bookkeeping methods
(:meth:`get_knowledge_base`, :meth:`list_knowledge_bases`) have
default implementations that delegate to the bound storage.
"""
def __init__(
self,
storage: "StorageBase",
vector_store: "VectorStoreBase",
) -> None:
"""Initialize the manager.
Args:
storage (`StorageBase`):
The application-wide storage backend used to persist
:class:`KnowledgeBaseRecord` rows and resolve
credentials.
vector_store (`VectorStoreBase`):
The application-wide vector store instance shared by
every knowledge base allocated by this manager.
"""
self._storage = storage
self._vector_store = vector_store
# ------------------------------------------------------------------
# Lifecycle hooks
# ------------------------------------------------------------------
async def __aenter__(self) -> Self:
"""Enter the manager's lifetime.
Enters the bound vector store's async context so a single
``create_app`` parameter (the manager) covers the vector store's
lifecycle too. Subclasses that override this MUST call
``await super().__aenter__()`` first to keep the vector store
ready before subclass-specific setup runs.
"""
await self._vector_store.__aenter__()
return self
async def __aexit__(
self,
exc_type: type[BaseException] | None,
exc: BaseException | None,
tb: "TracebackType | None",
) -> None:
"""Exit the manager's lifetime, releasing the vector store.
Args:
exc_type (`type[BaseException] | None`):
The exception type raised inside the with-block, if any.
exc (`BaseException | None`):
The exception instance raised inside the with-block,
if any.
tb (`TracebackType | None`):
The traceback for the raised exception, if any.
"""
await self._vector_store.__aexit__(exc_type, exc, tb)
# ------------------------------------------------------------------
# Capability discovery
# ------------------------------------------------------------------
@abstractmethod
async def get_dimension_policy(self) -> DimensionPolicy:
"""Return the embedding-dimension policy this manager enforces.
Surfaced over HTTP so the front-end can soft-filter
incompatible models / dimensions before submission and show a
helpful banner.
Returns:
`DimensionPolicy`:
The current dimension policy.
"""
# ------------------------------------------------------------------
# CRUD
# ------------------------------------------------------------------
@abstractmethod
async def create_knowledge_base(
self,
user_id: str,
name: str,
description: str,
embedding_model_config: "EmbeddingModelConfig",
) -> "KnowledgeBaseRecord":
"""Create a new knowledge base for the given user.
Implementations must:
1. validate ``embedding_model_config.dimensions`` against
:meth:`get_dimension_policy`;
2. allocate the vector store collection (or namespace) the
strategy uses;
3. persist a :class:`KnowledgeBaseRecord` and return it.
Args:
user_id (`str`):
The owner user id.
name (`str`):
Display name.
description (`str`):
Free-form description.
embedding_model_config (`EmbeddingModelConfig`):
Embedding model configuration; pinned to the record.
Returns:
`KnowledgeBaseRecord`:
The newly persisted record.
Raises:
`DimensionPolicyError`:
If the requested dimension violates the manager's
dimension policy.
"""
async def get_knowledge_base(
self,
user_id: str,
knowledge_base_id: str,
) -> "KnowledgeBaseRecord | None":
"""Fetch a knowledge base record by id (delegates to storage).
Args:
user_id (`str`):
The owner user id.
knowledge_base_id (`str`):
The knowledge base id.
Returns:
`KnowledgeBaseRecord | None`:
The record, or ``None`` if not found / not owned by
the user.
"""
return await self._storage.get_knowledge_base(
user_id,
knowledge_base_id,
)
async def list_knowledge_bases(
self,
user_id: str,
) -> "list[KnowledgeBaseRecord]":
"""List all knowledge base records owned by the given user.
Args:
user_id (`str`):
The owner user id.
Returns:
`list[KnowledgeBaseRecord]`:
All knowledge base records belonging to the user.
"""
return await self._storage.list_knowledge_bases(user_id)
async def update_knowledge_base(
self,
user_id: str,
knowledge_base_id: str,
name: str | None = None,
description: str | None = None,
) -> "KnowledgeBaseRecord | None":
"""Update mutable fields on an existing knowledge base record.
Only ``name`` and ``description`` are mutable. The embedding
model configuration and the underlying collection are pinned
for the lifetime of the record because changing either would
invalidate every previously inserted vector.
Args:
user_id (`str`):
The owner user id.
knowledge_base_id (`str`):
The knowledge base id.
name (`str | None`, optional):
New display name; ``None`` leaves the name unchanged.
description (`str | None`, optional):
New description; ``None`` leaves the description
unchanged.
Returns:
`KnowledgeBaseRecord | None`:
The updated record, or ``None`` if the record was not
found / not owned by the user.
"""
record = await self._storage.get_knowledge_base(
user_id,
knowledge_base_id,
)
if record is None:
return None
if name is not None:
record.name = name
if description is not None:
record.description = description
return await self._storage.upsert_knowledge_base(user_id, record)
@abstractmethod
async def delete_knowledge_base(
self,
user_id: str,
knowledge_base_id: str,
) -> bool:
"""Delete a knowledge base record and its underlying storage.
Implementations must:
1. authorise the call by looking the record up first;
2. drop the vector store collection (or scope) the strategy
uses;
3. remove the :class:`KnowledgeBaseRecord` from storage.
Args:
user_id (`str`):
The owner user id.
knowledge_base_id (`str`):
The id of the knowledge base to delete.
Returns:
`bool`:
``True`` if the record existed and was deleted,
``False`` if it was not found.
"""
# ------------------------------------------------------------------
# KnowledgeBase runtime
# ------------------------------------------------------------------
@abstractmethod
async def get_knowledge(
self,
user_id: str,
knowledge_base_id: str,
) -> "KnowledgeBase":
"""Resolve a runtime :class:`KnowledgeBase` handle for one KB.
Implementations are responsible for:
- looking the record up in storage (authorisation);
- resolving the embedding model from the record's credential;
- constructing the :class:`KnowledgeBase` with the strategy's
collection name and metadata filter.
Args:
user_id (`str`):
The owner user id.
knowledge_base_id (`str`):
The knowledge base id.
Returns:
`KnowledgeBase`:
A runtime handle bound to this knowledge base.
Raises:
`KnowledgeBaseNotFoundError`:
If the record does not exist or does not belong to
the authenticated user.
"""
|