🐛 Bug: error INCOMPLETE_STREAM in langgraph agent (with orchestrator node) #3208

Activity

1. jcguimaraesnet

   added

   bugSomething isn't workingSomething isn't working

   on Feb 13, 2026
2. copilotkit-support-bot commented on Feb 13, 2026

   

   copilotkit-support-botBot

   on Feb 13, 2026 – with CopilotKit Support Bot

   More actions

   Guidance from the codebase addresses the INCOMPLETE_STREAM error with LangGraph orchestrator nodes, outlining likely causes and step-by-step resolutions. See the detailed answer below for exact instructions and code snippets.

   Details

   The error you're encountering, INCOMPLETE_STREAM ("Run ended without emitting a terminal event") and the subsequent "No active text message found" error, typically occurs when a LangGraph node uses a customized configuration (via copilotkit_customize_config) that suppresses message emission, but the overall run lifecycle fails to signal completion correctly to the CopilotKit runtime.

   2-3 Likely Causes Investigated:

   1. Inconsistent Message Emission: When emit_messages=False is used, the runtime filters out TextMessageStart but might still attempt to send a TextMessageEnd if the metadata is not consistently propagated to the final "stop" chunk of the LLM stream.
   2. Missing Terminal Event: The Python LangGraphAgent implementation (especially in older versions like 0.1.74) may not be yielding a proper terminal event (RUN_FINISHED) that the TypeScript runtime expects to close the stream gracefully.
   3. Manual Message Interference: Using copilotkit_emit_message immediately before an LLM call with emit_messages=False can sometimes lead to state overlap in the event source accumulator if the manual message is not fully processed before the LLM stream starts.

   Solution & Recommended Steps

   To resolve this, ensure that your LangGraph nodes follow the updated pattern for terminal nodes and that you are using consistent configuration across nodes:

   • Ensure Terminal Node Signaling: Instead of relying solely on copilotkit_exit, ensure your terminal nodes return the final state correctly. If you are using an orchestrator that routes to sub-nodes, make sure the sub-nodes (like left_node and right_node) are recognized as terminal by LangGraph (e.g., routing to END).

   • Fix copilotkit_customize_config usage: Ensure the config passed to copilotkit_customize_config is the "clean" base config. Avoid nesting customized configs.

   • Update CopilotKit Python SDK: There were significant improvements in event streaming and terminal event handling in later versions. It is highly recommended to upgrade to the latest version of the Python SDK:

     pip install copilotkit --upgrade

   • Explicitly handle the final state in your nodes:
     Ensure that when you call copilotkit_exit, it is the final action in a terminal node. Note that copilotkit_exit doesn't stop execution immediately; it sets a flag for the next sync event.

   async def left_node(state: WorkflowState, config: Optional[RunnableConfig] = None):
       # Ensure we use the base config for customization
       config_internal = copilotkit_customize_config(config, emit_messages=True)
       model = ChatOpenAI(model="gpt-4o")
       prompt = "Provide a point for the left-wing political spectrum."
       
       # Use config_internal for the LLM call
       response = await model.ainvoke(prompt, config=config_internal)
       
       # Signal exit on the internal config
       await copilotkit_exit(config_internal)
   
       return {
           "messages": state.get('messages', []) + [response]
       }

   Why the error occurs:

   The INCOMPLETE_STREAM error specifically means the CopilotKit Runtime reached the end of the HTTP stream from your Python backend but never saw an event of type RunFinished. The TEXT_MESSAGE_END error occurs because the runner tries to "cleanup" and close any open messages when the stream ends abruptly, but since emit_messages=False prevented the start of the message, the cleanup fails.

   Documentation & Examples:

   • LangGraph Integration Guide
   • Troubleshooting CoAgents
   • Python SDK Reference

   Confidence Level: High (This is a known behavior in older SDK versions when mixing manual emissions with suppressed LLM streams).

   ---

   Was this helpful?

   If this solution worked for you, please click on the appropriate option below to help us improve:

   ✅ This was helpful | ❌ Need more help
3. jpr5

   closed this as completedin 5884ddeon Mar 6, 2026
4. pull

   added a commit that references this issue on Mar 6, 2026

   fix(runtime): clean up stale messagesInProcess when copilotkit:emit-m…

   ...

   This issue will close once commit 5884dde is merged into the 'main' branch.

   5884dde