[BUG] When building Langgraph agents from scratch, Using tools with Langgraph workers results in Message Stopped -> Cannot concat object and undefined

[zebathon]

Describe the bug
The demos seem to work ok. However, when building from scratch adding Tools to worker nodes results in this error.

• In the UX the frontend reports 'Message Stopped'
• In langsmith, the Worker node is reporting 'Aborted'
• In langsmith, the RunnableAgent is reporting the 'Cannot concat...' message.

See below

To Reproduce
Follow the most basic steps to build an Agent workflow

1. Add 1 Supervisor, 1 Worker, 1 Chat model, and 1 Tool (calculator) all with default values. (Add explicit instructions for worker to use the tool)
2. Start the conversation with a simple 'calculate 5*5'
3. See error 'Message Stopped'
4. Look in Langsmith for root cause Aborted -> Cannot concatenate etc. etc.

Expected behavior
The agent flow should be complete.

When disconnecting the tool, AgentFlow proceeds.

Screenshots

Flow
test Agents.json

Setup

• Installation npx flowise start
• Flowise Version 1.8.0
• OS: macOS
• Browser: chrome

Additional context
I've tried all types of Tools and Chat Models. Judging by the message I'm guessing it's a problem with either a null/undefinied input to the tool, or null output.

Cannot concat object and undefined

Error: Cannot concat object and undefined
    at concat (/usr/local/lib/node_modules/flowise/node_modules/@langchain/core/dist/utils/stream.cjs:165:15)
    at ChatPromptTemplate.transform (/usr/local/lib/node_modules/flowise/node_modules/@langchain/core/dist/runnables/base.cjs:381:53)
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
    at async ChatOpenAI.transform (/usr/local/lib/node_modules/flowise/node_modules/@langchain/core/dist/runnables/base.cjs:373:26)
    at async RunnableBinding.transform (/usr/local/lib/node_modules/flowise/node_modules/@langchain/core/dist/runnables/base.cjs:741:9)
    at async ToolCallingAgentOutputParser.transform (/usr/local/lib/node_modules/flowise/node_modules/@langchain/core/dist/runnables/base.cjs:373:26)
    at async RunnableSequence._streamIterator (/usr/local/lib/node_modules/flowise/node_modules/@langchain/core/dist/runnables/base.cjs:1104:30)

[HenryHengZJ]

If you try modify the Worker prompt to something related to Math so that the Supervisor can correctly use the Worker, does it work?

[zebathon]

No luck with providing more Math related prompting. FWIW I've attached a langsmith screenshot of the worker invoking a Model with valid input (from the demos). And a 2nd screenshot (from the failing test) with no input_data to the Model. Thus the model is never called by the worker without input data.

[zebathon]

I have the trace here

[brucechou1983]

Same here.

[mateusluizfb]

@HenryHengZJ same happening here. I asked the supervisor to return data, so it called a worker called "data-center" that has access to some custom tools that can perform a GET request. But I'm getting this error Cannot concat object and undefined

I'm using the docker version.

[eskimm]

Experiencing the similar issues. @HenryHengZJ can you elaborate how to craft prompts to have correct relationship between Supervisor/Workers? Or point to Langraph docs that describe correct use.

[rupjae]

reproduced.

[HenryHengZJ]

PR bugfix merged, will be pushing to new releases soon!

[HenryHengZJ]

can anyone confirm if the issue persists in v1.8.1 or resolved?

[CubanHackerAI]

issue still persists on 1.8.1

[toi500]

@HenryHengZJ still bugged #2557