rfc: agent initial tool choices

libs/langchain/langchain/agents/agent.py

@@ -925,6 +925,7 @@ class AgentExecutor(Chain):
     trim_intermediate_steps: Union[
         int, Callable[[List[Tuple[AgentAction, str]]], List[Tuple[AgentAction, str]]]
     ] = -1
+    initial_tool_choices: Sequence[Union[str, BaseTool]] = ()
 
     @classmethod
     def from_agent_and_tools(
@@ -954,6 +955,10 @@ class AgentExecutor(Chain):
                     f"Allowed tools ({allowed_tools}) different than "
                     f"provided tools ({[tool.name for tool in tools]})"
                 )
+        if values["max_iterations"] is not None:
+            values["max_iterations"] = max(
+                values["max_iterations"], len(values["initial_tool_choices"])
+            )
         return values
 
     @root_validator()
@@ -1373,10 +1378,19 @@ class AgentExecutor(Chain):
         start_time = time.time()
         # We now enter the agent loop (until it returns something).
         while self._should_continue(iterations, time_elapsed):
+            curr_inputs = inputs.copy()
+            if len(self.initial_tool_choices) > iterations:
+                tool_choice = self.initial_tool_choices[iterations]
+                tool_choice = (
+                    tool_choice.name
+                    if isinstance(tool_choice, BaseTool)
+                    else tool_choice
+                )
+                curr_inputs["tool_choice"] = tool_choice
             next_step_output = self._take_next_step(
                 name_to_tool_map,
                 color_mapping,
-                inputs,
+                curr_inputs,
                 intermediate_steps,
                 run_manager=run_manager,
             )

libs/langchain/langchain/agents/openai_functions_agent/base.py

@@ -18,7 +18,7 @@ from langchain_core.prompts.chat import (
     MessagesPlaceholder,
 )
 from langchain_core.pydantic_v1 import root_validator
-from langchain_core.runnables import Runnable, RunnablePassthrough
+from langchain_core.runnables import Runnable, RunnableLambda, RunnablePassthrough
 from langchain_core.tools import BaseTool
 
 from langchain.agents import BaseSingleActionAgent
@@ -229,6 +229,15 @@ class OpenAIFunctionsAgent(BaseSingleActionAgent):
         )
 
 
+def _configure_prompt_llm(
+    inputs: dict, prompt: BasePromptTemplate, llm_with_tools: Runnable
+) -> Runnable:
+    tool_choice = inputs.pop("tool_choice")
+    if tool_choice:
+        llm_with_tools.bind(functional_call={"name": tool_choice})
+    return prompt | llm_with_tools
+
+
 def create_openai_functions_agent(
     llm: BaseLanguageModel, tools: Sequence[BaseTool], prompt: ChatPromptTemplate
 ) -> Runnable:
@@ -307,8 +316,9 @@ def create_openai_functions_agent(
                 x["intermediate_steps"]
             )
         )
-        | prompt
-        | llm_with_tools
+        | RunnableLambda(_configure_prompt_llm).bind(
+            prompt=prompt, llm_with_tools=llm_with_tools
+        )
         | OpenAIFunctionsAgentOutputParser()
     )
     return agent