Merge branch 'main' into incremental_development

config/config.yaml

@@ -117,4 +117,6 @@ RPM: 10
 ### repair operation on the content extracted from LLM's raw output. Warning, it improves the result but not fix all cases.
 # REPAIR_LLM_OUTPUT: false
 
-# PROMPT_FORMAT: json #json or markdown
+# PROMPT_FORMAT: json #json or markdown
+
+# DISABLE_LLM_PROVIDER_CHECK: false

metagpt/actions/research.py

@@ -85,7 +85,7 @@ class CollectLinks(Action):
     llm: BaseGPTAPI = Field(default_factory=LLM)
     desc: str = "Collect links from a search engine."
     search_engine: SearchEngine = Field(default_factory=SearchEngine)
-    rank_func: Union[Callable[[list[str]], None], None] = None
+    rank_func: Optional[Callable[[list[str]], None]] = None
 
     async def run(
         self,
@@ -180,16 +180,18 @@ class WebBrowseAndSummarize(Action):
     llm: BaseGPTAPI = Field(default_factory=LLM)
     desc: str = "Explore the web and provide summaries of articles and webpages."
     browse_func: Union[Callable[[list[str]], None], None] = None
-    web_browser_engine: WebBrowserEngine = WebBrowserEngine(
-        engine=WebBrowserEngineType.CUSTOM if browse_func else None,
-        run_func=browse_func,
-    )
+    web_browser_engine: Optional[WebBrowserEngine] = None
 
     def __init__(self, **kwargs):
         super().__init__(**kwargs)
         if CONFIG.model_for_researcher_summary:
             self.llm.model = CONFIG.model_for_researcher_summary
 
+        self.web_browser_engine = WebBrowserEngine(
+            engine=WebBrowserEngineType.CUSTOM if self.browse_func else None,
+            run_func=self.browse_func,
+        )
+
     async def run(
         self,
         url: str,

metagpt/config.py

@@ -107,7 +107,9 @@ class Config(metaclass=Singleton):
         self.gemini_api_key = self._get("GEMINI_API_KEY")
         self.ollama_api_base = self._get("OLLAMA_API_BASE")
         self.ollama_api_model = self._get("OLLAMA_API_MODEL")
-        _ = self.get_default_llm_provider_enum()
+
+        if not self._get("DISABLE_LLM_PROVIDER_CHECK"):
+            _ = self.get_default_llm_provider_enum()
 
         self.openai_base_url = self._get("OPENAI_BASE_URL")
         self.openai_proxy = self._get("OPENAI_PROXY") or self.global_proxy

metagpt/llm.py

@@ -6,6 +6,8 @@
 @File    : llm.py
 """
 
+from typing import Optional
+
 from metagpt.config import CONFIG, LLMProviderEnum
 from metagpt.provider.base_gpt_api import BaseGPTAPI
 from metagpt.provider.human_provider import HumanProvider
@@ -14,6 +16,9 @@ from metagpt.provider.llm_provider_registry import LLM_REGISTRY
 _ = HumanProvider()  # Avoid pre-commit error
 
 
-def LLM(provider: LLMProviderEnum = CONFIG.get_default_llm_provider_enum()) -> BaseGPTAPI:
+def LLM(provider: Optional[LLMProviderEnum] = None) -> BaseGPTAPI:
     """get the default llm provider"""
+    if provider is None:
+        provider = CONFIG.get_default_llm_provider_enum()
+
     return LLM_REGISTRY.get_provider(provider)

metagpt/logs.py

@@ -8,6 +8,7 @@
 
 import sys
 from datetime import datetime
+from functools import partial
 
 from loguru import logger as _logger
 
@@ -26,3 +27,15 @@ def define_log_level(print_level="INFO", logfile_level="DEBUG"):
 
 
 logger = define_log_level()
+
+
+def log_llm_stream(msg):
+    _llm_stream_log(msg)
+
+
+def set_llm_stream_logfunc(func):
+    global _llm_stream_log
+    _llm_stream_log = func
+
+
+_llm_stream_log = partial(print, end="")

metagpt/provider/google_gemini_api.py

@@ -20,7 +20,7 @@ from tenacity import (
 )
 
 from metagpt.config import CONFIG, LLMProviderEnum
-from metagpt.logs import logger
+from metagpt.logs import log_llm_stream, logger
 from metagpt.provider.base_gpt_api import BaseGPTAPI
 from metagpt.provider.llm_provider_registry import register_provider
 from metagpt.provider.openai_api import CostManager, log_and_reraise
@@ -119,7 +119,7 @@ class GeminiGPTAPI(BaseGPTAPI):
         collected_content = []
         async for chunk in resp:
             content = chunk.text
-            print(content, end="")
+            log_llm_stream(content)
             collected_content.append(content)
 
         full_content = "".join(collected_content)

metagpt/provider/ollama_api.py

@@ -15,7 +15,7 @@ from tenacity import (
 
 from metagpt.config import CONFIG, LLMProviderEnum
 from metagpt.const import LLM_API_TIMEOUT
-from metagpt.logs import logger
+from metagpt.logs import log_llm_stream, logger
 from metagpt.provider.base_gpt_api import BaseGPTAPI
 from metagpt.provider.general_api_requestor import GeneralAPIRequestor
 from metagpt.provider.llm_provider_registry import register_provider
@@ -127,7 +127,7 @@ class OllamaGPTAPI(BaseGPTAPI):
             if not chunk.get("done", False):
                 content = self.get_choice_text(chunk)
                 collected_content.append(content)
-                print(content, end="")
+                log_llm_stream(content)
             else:
                 # stream finished
                 usage = self.get_usage(chunk)

metagpt/provider/openai_api.py

@@ -29,7 +29,7 @@ from tenacity import (
 )
 
 from metagpt.config import CONFIG, Config, LLMProviderEnum
-from metagpt.logs import logger
+from metagpt.logs import log_llm_stream, logger
 from metagpt.provider.base_gpt_api import BaseGPTAPI
 from metagpt.provider.constant import GENERAL_FUNCTION_SCHEMA, GENERAL_TOOL_CHOICE
 from metagpt.provider.llm_provider_registry import register_provider
@@ -222,7 +222,7 @@ class OpenAIGPTAPI(BaseGPTAPI, RateLimiter):
                 chunk_message = chunk.choices[0].delta  # extract the message
                 collected_messages.append(chunk_message)  # save the message
                 if chunk_message.content:
-                    print(chunk_message.content, end="")
+                    log_llm_stream(chunk_message.content)
         print()
 
         full_reply_content = "".join([m.content for m in collected_messages if m.content])

metagpt/provider/zhipuai_api.py

@@ -17,7 +17,7 @@ from tenacity import (
 )
 
 from metagpt.config import CONFIG, LLMProviderEnum
-from metagpt.logs import logger
+from metagpt.logs import log_llm_stream, logger
 from metagpt.provider.base_gpt_api import BaseGPTAPI
 from metagpt.provider.llm_provider_registry import register_provider
 from metagpt.provider.openai_api import CostManager, log_and_reraise
@@ -50,6 +50,8 @@ class ZhiPuAIGPTAPI(BaseGPTAPI):
         assert config.zhipuai_api_key
         zhipuai.api_key = config.zhipuai_api_key
         openai.api_key = zhipuai.api_key  # due to use openai sdk, set the api_key but it will't be used.
+        if config.openai_proxy:
+            openai.proxy = config.openai_proxy
 
     def _const_kwargs(self, messages: list[dict]) -> dict:
         kwargs = {"model": self.model, "prompt": messages, "temperature": 0.3}
@@ -94,7 +96,7 @@ class ZhiPuAIGPTAPI(BaseGPTAPI):
             if event.event == ZhiPuEvent.ADD.value:
                 content = event.data
                 collected_content.append(content)
-                print(content, end="")
+                log_llm_stream(content)
             elif event.event == ZhiPuEvent.ERROR.value or event.event == ZhiPuEvent.INTERRUPTED.value:
                 content = event.data
                 logger.error(f"event error: {content}", end="")

metagpt/roles/researcher.py

@@ -5,6 +5,7 @@
 """
 
 import asyncio
+import re
 
 from pydantic import BaseModel
 
@@ -95,9 +96,11 @@ class Researcher(Role):
         return msg
 
     def write_report(self, topic: str, content: str):
+        filename = re.sub(r'[\\/:"*?<>|]+', " ", topic)
+        filename = filename.replace("\n", "")
         if not RESEARCH_PATH.exists():
             RESEARCH_PATH.mkdir(parents=True)
-        filepath = RESEARCH_PATH / f"{topic}.md"
+        filepath = RESEARCH_PATH / f"{filename}.md"
         filepath.write_text(content)
 
 

metagpt/roles/role.py

@@ -152,7 +152,7 @@ class Role(BaseModel):
     builtin_class_name: str = ""
 
     _private_attributes = {
-        "_llm": LLM() if not is_human else HumanProvider(),
+        "_llm": None,
         "_role_id": _role_id,
         "_states": [],
         "_actions": [],

metagpt/roles/sk_agent.py

@@ -7,13 +7,13 @@
 @Modified By: mashenquan, 2023-11-1. In accordance with Chapter 2.2.1 and 2.2.2 of RFC 116, utilize the new message
         distribution feature for message filtering.
 """
+from typing import Any, Type
 
 from pydantic import Field
 from semantic_kernel import Kernel
-from semantic_kernel.orchestration.sk_function_base import SKFunctionBase
 from semantic_kernel.planning import SequentialPlanner
 from semantic_kernel.planning.action_planner.action_planner import ActionPlanner
-from semantic_kernel.planning.basic_planner import BasicPlanner, Plan
+from semantic_kernel.planning.basic_planner import BasicPlanner
 
 from metagpt.actions import UserRequirement
 from metagpt.actions.execute_task import ExecuteTask
@@ -41,13 +41,13 @@ class SkAgent(Role):
     goal: str = "Execute task based on passed in task description"
     constraints: str = ""
 
-    plan: Plan = None
-    planner_cls: BasicPlanner = BasicPlanner
-    planner: BasicPlanner = Field(default_factory=BasicPlanner)
+    plan: Any = None
+    planner_cls: Any = None
+    planner: Any = None
     llm: BaseGPTAPI = Field(default_factory=LLM)
     kernel: Kernel = Field(default_factory=Kernel)
-    import_semantic_skill_from_directory: str = ""
-    import_skill: dict[str, SKFunctionBase] = dict()
+    import_semantic_skill_from_directory: Type[Kernel.import_semantic_skill_from_directory] = None
+    import_skill: Type[Kernel.import_skill] = None
 
     def __init__(self, **kwargs) -> None:
         """Initializes the Engineer role with given attributes."""
@@ -57,8 +57,8 @@ class SkAgent(Role):
         self.kernel = make_sk_kernel()
 
         # how funny the interface is inconsistent
-        if self.planner_cls == BasicPlanner:
-            self.planner = self.planner_cls()
+        if self.planner_cls == BasicPlanner or self.planner_cls is None:
+            self.planner = BasicPlanner()
         elif self.planner_cls in [SequentialPlanner, ActionPlanner]:
             self.planner = self.planner_cls(self.kernel)
         else:
@@ -78,6 +78,7 @@ class SkAgent(Role):
 
     async def _act(self) -> Message:
         # how funny the interface is inconsistent
+        result = None
         if isinstance(self.planner, BasicPlanner):
             result = await self.planner.execute_plan_async(self.plan, self.kernel)
         elif any(isinstance(self.planner, cls) for cls in [SequentialPlanner, ActionPlanner]):

tests/metagpt/actions/test_research.py

@@ -0,0 +1,105 @@
+import pytest
+
+from metagpt.actions import research
+
+
+@pytest.mark.asyncio
+async def test_collect_links(mocker):
+    async def mock_llm_ask(self, prompt: str, system_msgs):
+        if "Please provide up to 2 necessary keywords" in prompt:
+            return '["metagpt", "llm"]'
+
+        elif "Provide up to 4 queries related to your research topic" in prompt:
+            return (
+                '["MetaGPT use cases", "The roadmap of MetaGPT", '
+                '"The function of MetaGPT", "What llm MetaGPT support"]'
+            )
+        elif "sort the remaining search results" in prompt:
+            return "[1,2]"
+
+    mocker.patch("metagpt.provider.base_gpt_api.BaseGPTAPI.aask", mock_llm_ask)
+    resp = await research.CollectLinks().run("The application of MetaGPT")
+    for i in ["MetaGPT use cases", "The roadmap of MetaGPT", "The function of MetaGPT", "What llm MetaGPT support"]:
+        assert i in resp
+
+
+@pytest.mark.asyncio
+async def test_collect_links_with_rank_func(mocker):
+    rank_before = []
+    rank_after = []
+    url_per_query = 4
+
+    def rank_func(results):
+        results = results[:url_per_query]
+        rank_before.append(results)
+        results = results[::-1]
+        rank_after.append(results)
+        return results
+
+    mocker.patch("metagpt.provider.base_gpt_api.BaseGPTAPI.aask", mock_collect_links_llm_ask)
+    resp = await research.CollectLinks(rank_func=rank_func).run("The application of MetaGPT")
+    for x, y, z in zip(rank_before, rank_after, resp.values()):
+        assert x[::-1] == y
+        assert [i["link"] for i in y] == z
+
+
+@pytest.mark.asyncio
+async def test_web_browse_and_summarize(mocker):
+    async def mock_llm_ask(*args, **kwargs):
+        return "metagpt"
+
+    mocker.patch("metagpt.provider.base_gpt_api.BaseGPTAPI.aask", mock_llm_ask)
+    url = "https://github.com/geekan/MetaGPT"
+    url2 = "https://github.com/trending"
+    query = "What's new in metagpt"
+    resp = await research.WebBrowseAndSummarize().run(url, query=query)
+
+    assert len(resp) == 1
+    assert url in resp
+    assert resp[url] == "metagpt"
+
+    resp = await research.WebBrowseAndSummarize().run(url, url2, query=query)
+    assert len(resp) == 2
+
+    async def mock_llm_ask(*args, **kwargs):
+        return "Not relevant."
+
+    mocker.patch("metagpt.provider.base_gpt_api.BaseGPTAPI.aask", mock_llm_ask)
+    resp = await research.WebBrowseAndSummarize().run(url, query=query)
+
+    assert len(resp) == 1
+    assert url in resp
+    assert resp[url] is None
+
+
+@pytest.mark.asyncio
+async def test_conduct_research(mocker):
+    data = None
+
+    async def mock_llm_ask(*args, **kwargs):
+        nonlocal data
+        data = f"# Research Report\n## Introduction\n{args} {kwargs}"
+        return data
+
+    mocker.patch("metagpt.provider.base_gpt_api.BaseGPTAPI.aask", mock_llm_ask)
+    content = (
+        "MetaGPT takes a one line requirement as input and "
+        "outputs user stories / competitive analysis / requirements / data structures / APIs / documents, etc."
+    )
+
+    resp = await research.ConductResearch().run("The application of MetaGPT", content)
+    assert resp == data
+
+
+async def mock_collect_links_llm_ask(self, prompt: str, system_msgs):
+    if "Please provide up to 2 necessary keywords" in prompt:
+        return '["metagpt", "llm"]'
+
+    elif "Provide up to 4 queries related to your research topic" in prompt:
+        return (
+            '["MetaGPT use cases", "The roadmap of MetaGPT", ' '"The function of MetaGPT", "What llm MetaGPT support"]'
+        )
+    elif "sort the remaining search results" in prompt:
+        return "[1,2]"
+
+    return ""

tests/metagpt/provider/test_zhipuai_api.py

@@ -35,3 +35,10 @@ async def test_zhipuai_acompletion(mocker):
 
     assert resp["code"] == 200
     assert "chatglm-turbo" in resp["data"]["choices"][0]["content"]
+
+def test_zhipuai_proxy(mocker):
+    import openai
+    from metagpt.config import CONFIG
+    CONFIG.openai_proxy = 'http://127.0.0.1:8080'
+    _ = ZhiPuAIGPTAPI()
+    assert openai.proxy == CONFIG.openai_proxy

tests/metagpt/roles/test_researcher.py

@@ -32,3 +32,19 @@ async def test_researcher(mocker):
         researcher.RESEARCH_PATH = Path(dirname)
         await researcher.Researcher().run(topic)
         assert (researcher.RESEARCH_PATH / f"{topic}.md").read_text().startswith("# Research Report")
+
+
+def test_write_report(mocker):
+    with TemporaryDirectory() as dirname:
+        for i, topic in enumerate(
+            [
+                ("1./metagpt"),
+                ('2.:"metagpt'),
+                ("3.*?<>|metagpt"),
+                ("4. metagpt\n"),
+            ]
+        ):
+            researcher.RESEARCH_PATH = Path(dirname)
+            content = "# Research Report"
+            researcher.Researcher().write_report(topic, content)
+            assert (researcher.RESEARCH_PATH / f"{i+1}. metagpt.md").read_text().startswith("# Research Report")