steven_li
5ba5c7e4c1
集成新的Beaver后端服务到应用实例中，替换原有的nanobot实现。 主要变更包括： - 在Dockerfile和环境配置中添加Beaver相关路径和配置变量 - 更新工作目录结构从.nanobot到.beaver - 实现Beaver引擎加载器，支持配置文件加载和工具组装 - 添加内置工具如ListDirectoryTool、ReadFileTool、SearchFilesTool - 更新消息处理流程，支持通道适配器和网关模式 - 重构技能系统，支持显式工具提示和嵌入式检索 - 改进错误处理和生命周期管理 此变更使应用实例能够使用统一的Beaver后端进行AI代理运行时管理。
107 lines
3.7 KiB
Python
107 lines
3.7 KiB
Python
"""Task-driven tool assembler.
|
|
|
|
这层和 SkillAssembler 的位置类似：它不执行工具，只决定本轮 run 应该把哪些
|
|
tool schema 暴露给模型。
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from collections.abc import Sequence
|
|
from typing import TYPE_CHECKING
|
|
|
|
from beaver.engine.context import SkillContext
|
|
from beaver.foundation.embedding import EmbeddingRetriever
|
|
from beaver.tools.base import ToolSpec
|
|
from beaver.tools.registry import ToolRegistry
|
|
|
|
if TYPE_CHECKING:
|
|
from beaver.engine.providers.runtime import ProviderRuntime
|
|
from beaver.skills.catalog.loader import SkillsLoader
|
|
|
|
|
|
class ToolAssembler:
|
|
"""Use skill hints and embedding retrieval to select run-scoped tools."""
|
|
|
|
def __init__(
|
|
self,
|
|
*,
|
|
retriever: EmbeddingRetriever | None = None,
|
|
always_tool_names: Sequence[str] | None = None,
|
|
) -> None:
|
|
self.retriever = retriever or EmbeddingRetriever()
|
|
self.always_tool_names = tuple(always_tool_names or ("memory", "session_search", "skill_view"))
|
|
|
|
async def assemble(
|
|
self,
|
|
*,
|
|
task_description: str,
|
|
registry: ToolRegistry,
|
|
skills_loader: SkillsLoader | None = None,
|
|
activated_skills: Sequence[SkillContext] | None = None,
|
|
embedding_runtime: ProviderRuntime | None = None,
|
|
top_k: int = 10,
|
|
) -> list[ToolSpec]:
|
|
"""Return selected tool specs for the current run.
|
|
|
|
Selection order is intentionally deterministic:
|
|
1. always tools from config/spec
|
|
2. tools explicitly declared by activated skills
|
|
3. embedding top-k tools for the task
|
|
"""
|
|
|
|
selected: list[ToolSpec] = []
|
|
selected_names: set[str] = set()
|
|
|
|
def add_specs(specs: Sequence[ToolSpec]) -> None:
|
|
for spec in specs:
|
|
if spec.name in selected_names:
|
|
continue
|
|
selected.append(spec)
|
|
selected_names.add(spec.name)
|
|
|
|
add_specs(registry.list_always_specs())
|
|
add_specs(registry.get_specs(self.always_tool_names))
|
|
|
|
skill_tool_names = self._collect_skill_tool_names(
|
|
skills_loader=skills_loader,
|
|
activated_skills=activated_skills or (),
|
|
)
|
|
add_specs(registry.get_specs(skill_tool_names))
|
|
|
|
candidates = [
|
|
spec.to_embedding_candidate()
|
|
for spec in registry.list_specs()
|
|
if spec.name not in selected_names
|
|
]
|
|
retrieved = await self.retriever.retrieve(
|
|
query=task_description,
|
|
candidates=candidates,
|
|
top_k=top_k,
|
|
api_key=embedding_runtime.api_key if embedding_runtime is not None else None,
|
|
api_base=embedding_runtime.api_base if embedding_runtime is not None else None,
|
|
model=embedding_runtime.model if embedding_runtime is not None else None,
|
|
extra_headers=embedding_runtime.extra_headers if embedding_runtime is not None else None,
|
|
timeout_seconds=(
|
|
embedding_runtime.request_timeout_seconds if embedding_runtime is not None else None
|
|
),
|
|
fallback_top_k=top_k,
|
|
)
|
|
add_specs(registry.get_specs([item["name"] for item in retrieved]))
|
|
return selected
|
|
|
|
@staticmethod
|
|
def _collect_skill_tool_names(
|
|
*,
|
|
skills_loader: SkillsLoader | None,
|
|
activated_skills: Sequence[SkillContext],
|
|
) -> list[str]:
|
|
if skills_loader is None or not activated_skills:
|
|
return []
|
|
|
|
result: list[str] = []
|
|
for skill in activated_skills:
|
|
for name in skills_loader.get_skill_tool_hints(skill.name):
|
|
if name not in result:
|
|
result.append(name)
|
|
return result
|