Open-source, secure environment with real-world tools for enterprise-grade agents.
Python agentaiai-agentai-agentscode-interpretercopilotdevelopmentdevtoolsgptgpt-4javascriptllmnextjsopenaipythonreactsoftwaretypescript
commit
Vasek Mlejnskyd761f161d689
diff --git a/api-service/pyproject.toml b/api-service/pyproject.tomlindex 7a0ce6672..17d46e7aa 100644--- a/api-service/pyproject.toml+++ b/api-service/pyproject.toml@@ -8,7 +8,6 @@ authors = ["Devbook <[email protected]>"]python = ">=3.8.1,<4.0"flask = "^2.2.3"gunicorn = "^20.1.0"-openai = "^0.27.0"langchain = "^0.0.112"supabase = "^1.0.1"flask-cors = "^3.0.10"@@ -17,6 +16,7 @@ urllib3 = "^1.26.14"pydantic = "^1.10.6"aenum = "^3.1.11"python-dotenv = "^1.0.0"+openai = "^0.27.0"[build-system]requires = ["poetry-core>=1.3.0"]diff --git a/api-service/test.py b/api-service/test.pyindex f2b5836d9..50e76f5ce 100644--- a/api-service/test.py+++ b/api-service/test.py@@ -1,4 +1,8 @@+from typing import Optional, Tuple+import json+from langchain import LLMChain+from langchain.tools import BaseToolfrom langchain.agents import initialize_agent, AgentExecutorfrom langchain.agents.chat.base import ChatAgentfrom langchain.chat_models import ChatOpenAI@@ -14,8 +18,35 @@ from langchain.schema import HumanMessagefrom codegen.tools.playground import create_playground_tools-def create_prompt() -> ChatPromptTemplate:- pass+class InvalidTool(BaseTool):+ name = "InvalidTool"+ description = "indicates that the last selected tool was an invalid tool"++ def _run(self, err: str) -> str:+ return err++ async def _arun(self, err: str) -> str:+ return err+++FINAL_ANSWER_ACTION = "Final Answer:"+++class CustomChatAgent(ChatAgent):+ def _extract_tool_and_input(self, text: str) -> Optional[Tuple[str, str]]:+ if FINAL_ANSWER_ACTION in text:+ return "Final Answer", text.split(FINAL_ANSWER_ACTION)[-1].strip()+ _, action, _ = text.split("```")+ try:+ response = json.loads(action.strip())+ return response["action"], response["action_input"]+ except Exception as e:+ # input = response["action_input"]+ return (+ "InvalidTool",+ f"I just ran your response via json.loads and received this error\n{str(e)}\nPlease try again",+ )+ # raise ValueError(f"Could not parse LLM output: {text}")OPENAI_API_KEY = "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"@@ -32,9 +63,10 @@ FORMAT_INSTRUCTIONS = """"""PREFIX = """You are an AI JavaScript/Nodejs assistant.- Follow the user's requirements carefully & to the letter.+- Think step-by-step. Use each step to describe your plan on how you will go about implementing the required instructions.- Minimize any other prose.- You are building an Express server that handles your REST API and you are required to complete the code based on the provided instructions.-- You are starting with the following code snippet:+- Start with the following code snippet that runs the server```import express from 'express';const app = express();@@ -67,14 +99,17 @@ The $JSON_BLOB should only contain a SINGLE action, do NOT return a list of mult"action_input": $INPUT}}}}```-ALWAYS use the following format and don't add additional empty new lines:+ALWAYS use the following format:Instructions: the input instructions you must implementThought: you should always think about what to doAction:```-$JSON_BLOB+{{{{+ "action": $TOOL_NAME,+ "action_input": $INPUT+}}}}```Observation: the result of the action... (this Thought/Action/Observation can repeat N times)@@ -110,7 +145,8 @@ input_variables = ["method",# "instructions",]-prompt = ChatAgent.create_prompt(+# prompt = ChatAgent.create_prompt(+prompt = CustomChatAgent.create_prompt(tools=[*playground_tools,],@@ -121,30 +157,49 @@ prompt = ChatAgent.create_prompt()+invalid_tool = InvalidTool()# Create agent and it's executor-agent = ChatAgent.from_llm_and_tools(+# agent = ChatAgent.from_llm_and_tools(+agent = CustomChatAgent.from_llm_and_tools(llm=chat,- tools=playground_tools,+ tools=[+ *playground_tools,+ invalid_tool,+ ],prefix=PREFIX,suffix=SUFFIX,format_instructions=FORMAT_INSTRUCTIONS,input_variables=input_variables,verbose=True,)-ae = AgentExecutor.from_agent_and_tools(agent=agent, tools=playground_tools)-# Run-ae.run(- # "1. Check if the incoming request is POST request. If not, respond with an adequate error",- agent_scratchpad="",- input="""Here are the instructions from your boss:-1. Check if the incoming request is POST request. If not, respond with an adequate error.-2. Retrieve email from the request payload and check if it's a valid email. Respond with 'Ok' if it is, otherwise respond with the adequate error.""",- method="post",- # instructions="""// 1. Check if the incoming request is POST request. If not, respond with an adequate error.- # // 2. Retrieve email from the request payload and check if it's a valid email. Respond with 'Ok' if it is, otherwise respond with the adequate error.- # """,+ae = AgentExecutor.from_agent_and_tools(+ agent=agent, tools=[*playground_tools, invalid_tool])+try:+ # Run+ ae.run(+ # "1. Check if the incoming request is POST request. If not, respond with an adequate error",+ agent_scratchpad="",+ input="""Here are the instructions:+1. Check if the incoming request is POST request. If not, respond with an adequate error.+2. Retrieve email from the request payload and check if it's a valid email. Respond with 'Ok' if it is, otherwise respond with the adequate error.+3. Before you provide the final answer, generate the full server code and make sure it runs without any errors""",+ method="post",+ # instructions="""// 1. Check if the incoming request is POST request. If not, respond with an adequate error.+ # // 2. Retrieve email from the request payload and check if it's a valid email. Respond with 'Ok' if it is, otherwise respond with the adequate error.+ # """,+ )++except ValueError as e:+ if "Could not parse LLM output" in str(e):+ # Here we want to remind LLM that it's not properly formatting its output.+ # https://twitter.com/zachtratar/status/1633376345995739136+ # eg:+ # I just ran your response via json.loads and received this error:+ # {error}+ # Please try again+ print("FAILED TO PARSE LLM")# agent_exec = initialize_agent(# tools=playground_tools,