[BUG] MRKLAgent is not compatible with Custom Tool

[zhengxiang5965]

Describe the bug MRKLAgent is not compatible with Custom Tool

To Reproduce Simultaneously using MRKLAgent and Custom Tool to form a stream

Expected behavior A clear and concise description of what you expected to happen.

Screenshots

Flow Setup

• Installation [e.g. docker, npx flowise start, yarn start]
• Flowise Version [e.g. 1.2.11]
• OS: [e.g. macOS, Windows, Linux]
• Browser [e.g. chrome, safari]

Additional context Add any other context about the problem here.

2023-08-14 05:14:53 [ERROR]: [server]: Error: [ { "code": "invalid_type", "expected": "object", "received": "string", "path": [], "message": "Expected object, received string" } ] ZodError: [ { "code": "invalid_type", "expected": "object", "received": "string", "path": [], "message": "Expected object, received string" } ] at get error [as error] (/usr/local/lib/node_modules/flowise/node_modules/zod/lib/types.js:43:31) at ZodObject.parseAsync (/usr/local/lib/node_modules/flowise/node_modules/zod/lib/types.js:164:22) at process.processTicksAndRejections (node:internal/process/task_queues:95:5) at async DynamicStructuredTool.call (/usr/local/lib/node_modules/flowise/node_modules/langchain/dist/tools/base.cjs:29:24) at async /usr/local/lib/node_modules/flowise/node_modules/langchain/dist/agents/executor.cjs:104:23 at async Promise.all (index 0) at async AgentExecutor._call (/usr/local/lib/node_modules/flowise/node_modules/langchain/dist/agents/executor.cjs:100:30) at async AgentExecutor.call (/usr/local/lib/node_modules/flowise/node_modules/langchain/dist/chains/base.cjs:98:28) at async MRKLAgentLLM_Agents.run (/usr/local/lib/node_modules/flowise/node_modules/flowise-components/dist/nodes/agents/MRKLAgentLLM/MRKLAgentLLM.js:42:24) at async App.processPrediction (/usr/local/lib/node_modules/flowise/dist/index.js:873:23) error: unexpected EOF

[chungyau97]

Hi @zhengxiang5965,

Try using OpenAI Function Agent node.

[zhengxiang5965]

Hi @zhengxiang5965,

Try using OpenAI Function Agent node.

Why are you doing it so rigidly？ @chungyau97 Langflow's PythonFunctionTool can support multiple agents

[HenryHengZJ]

Hi @zhengxiang5965, Try using OpenAI Function Agent node.

Why are you doing it so rigidly？ @chungyau97 Langflow's PythonFunctionTool can support multiple agents

Custom Tool is meant to work with Conversation Agent or better with OpenAI Function Agent. Why? Because underlying its a Zod Schema similar to Pydantic, and works best with OpenAI Function Calling. That's why it's recommended to use OpenAI Function Agent

[zhengxiang5965]

The react prompt agent needs to be used in combination with "custom tools", and I believe it can be fully implemented technically, which has already been implemented in Langflow

@HenryHengZJ

and there will be no problem with "expected object, received string". This problem seems to be easy to solve

https://python.langchain.com/docs/modules/agents/tools/custom_tools

Obviously, ZERO SHOT REACT_ DESCRIPTION can be combined with 'customers tools'