g3fi-.$dZddlmZddlZddlmZddlmZmZm Z m Z erddl m Z m Z ddlmZddlmZdd lmZdd lmZmZdd lmZdd lmZmZmZmZdd lmZej@e!Z"dZ#eGddZ$ddZ%ddZ&GddeZ'y)z#LLM-based tool selector middleware.) annotationsN) dataclass) TYPE_CHECKING AnnotatedLiteralUnion) AwaitableCallable)BaseTool) BaseChatModel) HumanMessage)Field TypeAdapter) TypedDict)AgentMiddlewareModelCallResult ModelRequest ModelResponse)init_chat_modelzNYour goal is to select the most relevant tools for answering the user's query.cDeZdZUdZded<ded<ded<ded <d ed <y ) _SelectionRequestz#Prepared inputs for tool selection.list[BaseTool]available_toolsstrsystem_messager last_user_messager model list[str]valid_tool_namesN__name__ __module__ __qualname____doc____annotations__h/var/www/auto_recruiter/arenv/lib/python3.12/site-packages/langchain/agents/middleware/tool_selection.pyrr"s#-#### r'rc|s d}t||Dcgc]2}tt|jt |j f4}}t t|}d}Gddt}t|Scc}w)zCreate a structured output schema for tool selection. Args: tools: Available tools to include in the schema. Returns: `TypeAdapter` for a schema where each tool name is a `Literal` with its description. z&Invalid usage: tools must be non-empty) descriptionz2Tools to use. Place the most relevant tools first.ceZdZUdZded<y)>_create_tool_selection_response..ToolSelectionResponsezUse to select relevant tools.zCAnnotated[list[selected_tool_type], Field(description=description)]toolsNr r&r'r(ToolSelectionResponser,Ds +RRr'r.) AssertionErrorrrnamerr*rtuplerr)r-msgtoolliteralsselected_tool_typer*r.s r(_create_tool_selection_responser6-s 6S!! X]OS '$))$e8H8H&IIJHuX/FKS S , --s7A;c2djd|DS)zFormat tools as markdown list. Args: tools: Tools to format. Returns: Markdown string with each tool on a new line.  c3TK|] }d|jd|j"yw)z- z: N)r0r*).0r3s r( z$_render_tool_list..Us(Ldr$))Bt'7'7&89Ls&()joinr-s r(_render_tool_listr>Ls 99LeL LLr'ceZdZdZdeddd d fdZd dZ d dZ d dZ d dZ xZ S)LLMToolSelectorMiddlewareaUses an LLM to select relevant tools before calling the main model. When an agent has many tools available, this middleware filters them down to only the most relevant ones for the user's query. This reduces token usage and helps the main model focus on the right tools. Examples: !!! example "Limit to 3 tools" ```python from langchain.agents.middleware import LLMToolSelectorMiddleware middleware = LLMToolSelectorMiddleware(max_tools=3) agent = create_agent( model="openai:gpt-4o", tools=[tool1, tool2, tool3, tool4, tool5], middleware=[middleware], ) ``` !!! example "Use a smaller model for selection" ```python middleware = LLMToolSelectorMiddleware(model="openai:gpt-4o-mini", max_tools=2) ``` N)r system_prompt max_toolsalways_includect|||_||_|xsg|_t |t tdfr||_yt||_y)aInitialize the tool selector. Args: model: Model to use for selection. If not provided, uses the agent's main model. Can be a model identifier string or `BaseChatModel` instance. system_prompt: Instructions for the selection model. max_tools: Maximum number of tools to select. If the model selects more, only the first `max_tools` will be used. If not specified, there is no limit. always_include: Tool names to always include regardless of selection. These do not count against the `max_tools` limit. N) super__init__rArBrC isinstancer typerr)selfrrArBrC __class__s r(rFz"LLMToolSelectorMiddleware.__init__usR4 *",2 emT$Z8 9/4DJ(/DJr'c<|jrt|jdk(ry|jDcgc]}t|tr|}}|jrU|Dchc]}|j }}|jDcgc] }||vs| }}|rd|dt |}t||Dcgc]}|j |jvs|}}|sy|j} |j| d|jdz } t|jD]} t| ts| } nd}t||jxs |j} |Dcgc]}|j } }t|| | | | Scc}wcc}wcc}wcc}wcc}w) zPrepare inputs for tool selection. Returns: `SelectionRequest` with prepared inputs, or `None` if no selection is needed. rNz.Tools in always_include not found in request: z. Available tools: z IMPORTANT: List the tool names in order of relevance, with the most relevant first. If you exceed the maximum number of tools, only the first z will be used.z)No user message found in request messages)rrrrr)r-lenrGdictrCr0sorted ValueErrorrArBreversedmessagesr r/rr)rIrequestr3 base_toolsavailable_tool_namesr0 missing_toolsr2rrmessagerrrs r(_prepare_selection_requestz4LLMToolSelectorMiddleware._prepare_selection_requests}}GMM 2a 7(/}}StJtT> % "#'..!1A N  0 01 &G'<0$+! & >C % % +gmm2AB$DIIBB +)/-   [T$F`6Cs/F F F  F F5FF Fcbg}g}|dD]S}||vr|j|||vs|jt||jksC|j|U|rd|}t||D cgc]} | j|vs| } } |j D cgc],} t | ts| j|jvr| .} } | j| |j D cgc]} t | ts| } } |jg| | Scc} wcc} wcc} w)zBProcess the selection response and return filtered `ModelRequest`.r-zModel selected invalid tools: r=) appendrBrLrOr0r-rGrMrCextendoverride) rIresponserrrRselected_tool_namesinvalid_tool_selections tool_namer2r3selected_toolsalways_included_toolsprovider_toolss r(_process_selection_responsez5LLMToolSelectorMiddleware._process_selection_responsesX*,"$!'* 6I 00'..y9 33&#.A*BT^^*S#**95 6 #23J2KLCS/ !-*  =P0PD* *  1 dD)dii4;N;N.N 1 1 34,3==S4JtT-12 $+/ "0*"0 "0  "0 ) "0 "0H? B(J(J((J$ (J  (J  (JT))9)  )>//D/  /r'r@)r-rryr)r-rryr)(r$ __future__rlogging dataclassesrtypingrrrrcollections.abcr r langchain.toolsr *langchain_core.language_models.chat_modelsr langchain_core.messagesr pydanticrrtyping_extensionsr!langchain.agents.middleware.typesrrrrlangchain.chat_models.baser getLoggerr!loggerr{rr6r>r@r&r'r(rs)"!;;3(D0'' 7   8 $U      .> Mh/h/r'