[General] When using an Ollama model deployed locally, why doesn't the message output stream?

[835479131]

Description

When using an Ollama model deployed locally, why doesn't the message output stream? use Ollama's API directly for normal streaming output.

Additional Context

OLLAMA_CHAT_MODEL = OllamaChatModel(
model_name="qwen3:30b-a3b-instruct-2507-q4_K_M",
host="http://localhost:11434",
stream=True,
)
# 创建agent实例
agent = ReActAgent(
name="friday",
sys_prompt="you are a helpfull assistant",
model=OLLAMA_CHAT_MODEL,
formatter=OllamaChatFormatter(),
toolkit=toolkit,
memory=AgentScopeSessionHistoryMemory(
service=self.session_service,
session_id=session_id,
user_id=user_id
),
)

agent.set_console_output_enabled(enabled=False)

# 恢复状态（如果存在）
if state:
    agent.load_state_dict(state)

# 流式处理消息
async for msg, last in stream_printing_messages(
        agents=[agent],
        coroutine_task=agent(msgs),
):
    yield msg, last

[rayrayraykk]

Have you tried running the agent solely (run without runtime framework)?

[835479131]

Running just the agent without using the runtime has the same problem; it won't output in a streaming manner.
`
OLLAMA_CHAT_MODEL = OllamaChatModel(
model_name="qwen3:30b-a3b-instruct-2507-q4_K_M",
host="http://localhost:11434",
stream=True,
)

async def main():
toolkit = Toolkit()
toolkit.register_tool_function(execute_python_code)
toolkit.register_tool_function(execute_shell_command)

agent = ReActAgent(
    name="Friday",
    sys_prompt="You're a helpful assistant named Friday.",
    model=OLLAMA_CHAT_MODEL,
    memory=InMemoryMemory(),
    formatter=OllamaChatFormatter(),
    toolkit=toolkit,
)

user = UserAgent(name="user")

msg = None
while True:
    msg = await agent(msg)
    msg = await user(msg)
    if msg.get_text_content() == "exit":
        break

asyncio.run(main())`

Have you tried running the agent solely (run without runtime framework)?

[rayrayraykk]

@qbc2016 Could you please have a look into this issue?

[qbc2016]

Running just the agent without using the runtime has the same problem; it won't output in a streaming manner. ` OLLAMA_CHAT_MODEL = OllamaChatModel( model_name="qwen3:30b-a3b-instruct-2507-q4_K_M", host="http://localhost:11434", stream=True, )

async def main(): toolkit = Toolkit() toolkit.register_tool_function(execute_python_code) toolkit.register_tool_function(execute_shell_command)

agent = ReActAgent(
    name="Friday",
    sys_prompt="You're a helpful assistant named Friday.",
    model=OLLAMA_CHAT_MODEL,
    memory=InMemoryMemory(),
    formatter=OllamaChatFormatter(),
    toolkit=toolkit,
)

user = UserAgent(name="user")

msg = None
while True:
    msg = await agent(msg)
    msg = await user(msg)
    if msg.get_text_content() == "exit":
        break

asyncio.run(main())`

Have you tried running the agent solely (run without runtime framework)?

Hi, thank you for pointing this out. It is indeed a bug in AgentScope, and we have fixed it in this pr.