run-llama/llama_index · error · ValueError

Streaming is not enabled. Please use achat() instead.

Error message

Streaming is not enabled. Please use achat() instead.

What it means

Async counterpart of the sync error: CondenseQuestionChatEngine.astream_chat() raises ValueError when the underlying query engine produced a non-streaming response, i.e. it has no async_response_gen. The engine needs an async token generator to return a StreamingAgentChatResponse; without streaming enabled at query-engine construction there is nothing to iterate.

Source

Thrown at llama-index-core/llama_index/core/chat_engine/condense_question.py:374

        )

        # Record response
        if isinstance(query_response, AsyncStreamingResponse):
            # override the generator to include writing to chat history
            # TODO: query engine does not support async generator yet
            await self._memory.aput(ChatMessage(role=MessageRole.USER, content=message))
            response = StreamingAgentChatResponse(
                achat_stream=aresponse_gen_from_query_engine(
                    query_response.async_response_gen()
                ),
                sources=[tool_output],
            )
            response.awrite_response_to_history_task = asyncio.create_task(
                response.awrite_response_to_history(self._memory)
            )

        else:
            raise ValueError("Streaming is not enabled. Please use achat() instead.")
        return response

    def reset(self) -> None:
        # Clear chat history
        self._memory.reset()

    @property
    def chat_history(self) -> List[ChatMessage]:
        """Get chat history."""
        return self._memory.get_all()

View on GitHub (pinned to afd0fef371)

Solutions

1. Create the query engine with streaming enabled: index.as_query_engine(streaming=True) before constructing the chat engine
2. Or use await engine.achat('hi') which returns a complete AgentChatResponse without streaming
3. Confirm the bound LLM supports async streaming (implements astream_complete)

Example fix

# before
qe = index.as_query_engine()
chat = CondenseQuestionChatEngine.from_defaults(query_engine=qe)
resp = await chat.astream_chat('hello')  # ValueError

# after
qe = index.as_query_engine(streaming=True)
chat = CondenseQuestionChatEngine.from_defaults(query_engine=qe)
resp = await chat.astream_chat('hello')
async for token in resp.async_response_gen():
    print(token, end='')

qe = index.as_query_engine(streaming=True)
# ensure the LLM supports async streaming before astream_chat
assert hasattr(Settings.llm, 'astream_complete'), 'LLM must support astream_complete'

try:
    resp = await engine.astream_chat(msg)
except ValueError as e:
    if 'Streaming is not enabled' in str(e):
        resp = await engine.achat(msg)
    else:
        raise

When it happens

Trigger: await engine.astream_chat('hi') where the engine was built from index.as_query_engine() (default streaming=False) or any query engine that returns a plain CompletionResponse. The else-branch after the streaming check raises.

Common situations: Async web servers (FastAPI) calling astream_chat for token streaming while the query engine was built synchronously by default; mixing chat()/stream_chat() code paths and assuming streaming is a property of the chat engine rather than the query engine.

Related errors

• Streaming is not enabled. Please use chat() instead.
• response_gen is only available for streaming responses. Set
• achat_stream is None. Cannot asynchronously write to history
• achat_stream is None!
• astream_complete is not supported by default.

AI-assisted analysis of run-llama/llama_index@afd0fef371 (2026-08-15). Data as JSON: /api/errors/33c2d62d06fea355. Report an issue: GitHub.