mirror of
https://github.com/langchain-ai/langgraph-codeact.git
synced 2026-07-19 21:33:32 -04:00
Harrison Chase
85 lines
3.1 KiB
Python
85 lines
3.1 KiB
Python
import inspect
|
|
from collections import ChainMap
|
|
from typing import Any, Callable, Optional, Sequence, Callable, Union
|
|
|
|
from langchain_core.language_models import BaseChatModel
|
|
from langchain_core.messages import AIMessage, BaseMessage, MessageLikeRepresentation
|
|
from langchain_core.tools import BaseTool
|
|
from langchain_core.tools import tool as create_tool
|
|
|
|
from langgraph.checkpoint.base import BaseCheckpointSaver
|
|
from langgraph.func import entrypoint, task
|
|
from langgraph.store.base import BaseStore
|
|
from langgraph.graph import StateGraph, MessagesState
|
|
from langgraph.types import Command
|
|
from typing import Literal, Tuple
|
|
|
|
|
|
class CodeActState(MessagesState):
|
|
"""state for codeact agent."""
|
|
|
|
script: str
|
|
context: dict
|
|
|
|
|
|
def create_codeact(
|
|
model: BaseChatModel,
|
|
tools: Sequence[Union[BaseTool, Callable]],
|
|
eval_fn: Callable[[str, dict[str, Callable]], Tuple[str, dict]],
|
|
*,
|
|
prompt: Optional[str] = None,
|
|
):
|
|
_tools = [t if isinstance(t, BaseTool) else create_tool(t) for t in tools]
|
|
# create the prompt
|
|
prompt = f"""
|
|
{prompt or ""}
|
|
|
|
You will be given a task to perform. You should output either
|
|
- a Python code snippet that provides the solution to the task, or a step towards the solution. Any output you want to extract from the code should be printed to the console. Code should be output in a fenced code block.
|
|
- text to be shown directly to the user, if you want to ask for more information or provide the final answer.
|
|
|
|
In addition to the Python Standard Library, you can use the following functions:
|
|
"""
|
|
|
|
for tool in _tools:
|
|
prompt += f'''
|
|
def {tool.name}{str(inspect.signature(tool.func))}:
|
|
"""{tool.description}"""
|
|
...
|
|
'''
|
|
|
|
prompt += """
|
|
|
|
Variables defined at the top level of previous code snippets can be referenced in your code.
|
|
|
|
Reminder: use python code snippets to call tools"""
|
|
|
|
def call_model(state: CodeActState) -> Command[Literal["__end__", "sandbox"]]:
|
|
messages = [{"role": "system", "content": prompt}] + state["messages"]
|
|
msg = model.invoke(messages)
|
|
if "```" in msg.content:
|
|
# get content between fences
|
|
code = msg.content.split("```")[1]
|
|
# remove first line, which is the language or empty string
|
|
code = "\n".join(code.splitlines()[1:])
|
|
return Command(goto="sandbox", update={"messages": [msg], "script": code})
|
|
else:
|
|
# no code block, return None
|
|
return Command(goto="__end__", update={"messages": [msg], "script": ""})
|
|
|
|
def sandbox(state: CodeActState):
|
|
script = state["script"]
|
|
old_context = state.get("context", {})
|
|
context = {**old_context, **{tool.name: tool.func for tool in _tools}}
|
|
# execute the script
|
|
output, new_vars = eval_fn(script, context)
|
|
new_context = {**old_context, **new_vars}
|
|
return {"messages": [{"role": "user", "content": output}], "context": new_context}
|
|
|
|
agent = StateGraph(CodeActState)
|
|
agent.add_node(call_model)
|
|
agent.add_node(sandbox)
|
|
agent.add_edge("__start__", "call_model")
|
|
agent.add_edge("sandbox", "call_model")
|
|
return agent
|