Devika is the first open-source implementation of an Agentic Software Engineer. Initially started as an open-source alternative to Devin.
Python
Fixing playwright using sync API (#298)
Samuel Birocchi744eff7d60b8
diff --git a/devika.py b/devika.pyindex c03350d..6c63aec 100644--- a/devika.py+++ b/devika.py@@ -2,8 +2,10 @@DO NOT REARRANGE THE ORDER OF THE FUNCTION CALLS AND VARIABLE DECLARATIONSAS IT MAY CAUSE IMPORT ERRORS AND OTHER ISSUES"""-import eventlet-eventlet.monkey_patch()+# import eventlet+# eventlet.monkey_patch()+from gevent import monkey+monkey.patch_all()from src.init import init_devikainit_devika()@@ -91,7 +93,7 @@ def handle_message(data):thread.start()if action == 'execute_agent':- thread = Thread(target=lambda: agent.execute(message, project_name, search_engine))+ thread = Thread(target=lambda: agent.execute(message, project_name))thread.start()diff --git a/src/agents/agent.py b/src/agents/agent.pyindex 92fa935..d8241fb 100644--- a/src/agents/agent.py+++ b/src/agents/agent.py@@ -28,10 +28,13 @@ import jsonimport timeimport platformimport tiktoken+import asyncio++from src.socket_instance import emit_agentclass Agent:- def __init__(self, base_model: str, search_engine: str):+ def __init__(self, base_model: str, search_engine: str, browser: Browser = None):if not base_model:raise ValueError("base_model is required")@@ -63,23 +66,33 @@ class Agent:self.engine = search_engineself.tokenizer = tiktoken.get_encoding("cl100k_base")- def search_queries(self, queries: list, project_name: str, engine: str) -> dict:+ async def open_page(self, project_name, pdf_download_url):+ browser = await Browser().start()++ await browser.go_to(pdf_download_url)+ _, raw = await browser.screenshot(project_name)+ data = await browser.extract_text()+ await browser.close()++ return browser, raw, data++ def search_queries(self, queries: list, project_name: str) -> dict:results = {}+knowledge_base = KnowledgeBase()web_search = None- if engine == "bing":+ if self.engine == "bing":web_search = BingSearch()- elif engine == "google":+ elif self.engine == "google":web_search = GoogleSearch()else:web_search = DuckDuckGoSearch()- self.logger.info(f"Search : {engine}")+ self.logger.info(f"\nSearch Engine :: {self.engine}")for query in queries:- browser = Browser()query = query.strip().lower()# knowledge = knowledge_base.get_knowledge(tag=query)@@ -87,18 +100,19 @@ class Agent:# results[query] = knowledge# continue+ loop = asyncio.new_event_loop()+ asyncio.set_event_loop(loop)+web_search.search(query)link = web_search.get_first_link()print("\nLink :: ", link, '\n')- browser.go_to(link)- browser.screenshot(project_name)+ browser, raw, data = loop.run_until_complete(self.open_page(project_name, link))+ emit_agent("screenshot", {"data": raw, "project_name": project_name}, False)+ results[query] = self.formatter.execute(data, project_name)- results[query] = self.formatter.execute(browser.extract_text(), project_name)- # browser.close()self.logger.info(f"got the search results for : {query}")-# knowledge_base.add_knowledge(tag=query, contents=results[query])return results@@ -137,8 +151,7 @@ class Agent:project_name_space_url)response = f"I have generated the PDF document. You can download it from here: {pdf_download_url}"- Browser().go_to(pdf_download_url)- Browser().screenshot(project_name)+ asyncio.run(self.open_page(project_name, pdf_download_url))self.project_manager.add_message_from_devika(project_name, response)@@ -244,16 +257,15 @@ class Agent:project_name_space_url)response = f"I have generated the PDF document. You can download it from here: {pdf_download_url}"- browser = Browser()- browser.go_to(pdf_download_url)- browser.screenshot(project_name)+ self.browser.go_to(pdf_download_url)+ self.browser.screenshot(project_name)self.project_manager.add_message_from_devika(project_name, response)self.agent_state.set_agent_active(project_name, False)self.agent_state.set_agent_completed(project_name, True)- def execute(self, prompt: str, project_name_from_user: str = None, engine: str = None) -> str:+ def execute(self, prompt: str, project_name_from_user: str = None) -> str:"""Agentic flow of execution"""@@ -333,7 +345,7 @@ class Agent:self.agent_state.set_agent_active(project_name, True)if queries and len(queries) > 0:- search_results = self.search_queries(queries, project_name, engine)+ search_results = self.search_queries(queries, project_name)else:search_results = {}@@ -351,6 +363,6 @@ class Agent:self.agent_state.set_agent_active(project_name, False)self.agent_state.set_agent_completed(project_name, True)self.project_manager.add_message_from_devika(project_name,- "I have completed the my task. 🚀 \n"+ "I have completed the my task. \n""if you would like me to do anything else, please let me know. \n")diff --git a/src/agents/coder/coder.py b/src/agents/coder/coder.pyindex 6584644..324e261 100644--- a/src/agents/coder/coder.py+++ b/src/agents/coder/coder.py@@ -7,6 +7,7 @@ from typing import List, Dict, Unionfrom src.config import Configfrom src.llm import LLMfrom src.state import AgentState+from src.logger import LoggerPROMPT = open("src/agents/coder/prompt.jinja2", "r").read().strip()@@ -14,7 +15,7 @@ class Coder:def __init__(self, base_model: str):config = Config()self.project_dir = config.get_projects_dir()-+ self.logger = Logger()self.llm = LLM(model_id=base_model)def render(@@ -31,6 +32,11 @@ class Coder:def validate_response(self, response: str) -> Union[List[Dict[str, str]], bool]:response = response.strip()+ self.logger.debug(f"Response from the model: {response}")++ if "~~~" not in response:+ return False+response = response.split("~~~", 1)[1]response = response[:response.rfind("~~~")]response = response.strip()diff --git a/src/agents/feature/feature.py b/src/agents/feature/feature.pyindex 59dab56..811e31f 100644--- a/src/agents/feature/feature.py+++ b/src/agents/feature/feature.py@@ -85,13 +85,13 @@ class Feature:def emulate_code_writing(self, code_set: list, project_name: str):for file in code_set:- file = file["file"]+ filename = file["file"]code = file["code"]new_state = AgentState().new_state()new_state["internal_monologue"] = "Writing code..."- new_state["terminal_session"]["title"] = f"Editing {file}"- new_state["terminal_session"]["command"] = f"vim {file}"+ new_state["terminal_session"]["title"] = f"Editing {filename}"+ new_state["terminal_session"]["command"] = f"vim {filename}"new_state["terminal_session"]["output"] = codeAgentState().add_to_current_state(project_name, new_state)time.sleep(1)diff --git a/src/browser/browser.py b/src/browser/browser.pyindex 11f5645..3cf6ef3 100644--- a/src/browser/browser.py+++ b/src/browser/browser.py@@ -1,51 +1,61 @@+import asyncio+import base64import os-from playwright.sync_api import sync_playwright, TimeoutError+from playwright.sync_api import sync_playwright, TimeoutError, Page+from playwright.async_api import async_playwright, TimeoutErrorfrom markdownify import markdownify as mdfrom pdfminer.high_level import extract_text-+from src.socket_instance import emit_agentfrom src.config import Configfrom src.state import AgentStateclass Browser:def __init__(self):- self.playwright = sync_playwright().start()- chromium = self.playwright.chromium- self.browser = chromium.launch()- self.page = self.browser.new_page()+ self.playwright = None+ self.browser = None+ self.page = None+ self.agent = AgentState()++ async def start(self):+ self.playwright = await async_playwright().start()+ self.browser = await self.playwright.chromium.launch(headless=True)+ self.page = await self.browser.new_page()+ return self- def new_page(self):- return self.browser.new_page()+ # def new_page(self):+ # return self.browser.new_page()- def go_to(self, url):+ async def go_to(self, url):try:- self.page.goto(url, timeout=20000)+ await self.page.goto(url, timeout=20000)except TimeoutError as e:print(f"TimeoutError: {e} when trying to navigate to {url}")return Falsereturn True- def screenshot(self, project_name):+ async def screenshot(self, project_name):screenshots_save_path = Config().get_screenshots_dir()- page_metadata = self.page.evaluate("() => { return { url: document.location.href, title: document.title } }")+ page_metadata = await self.page.evaluate("() => { return { url: document.location.href, title: document.title } }")page_url = page_metadata['url']random_filename = os.urandom(20).hex()filename_to_save = f"{random_filename}.png"path_to_save = os.path.join(screenshots_save_path, filename_to_save)- self.page.emulate_media(media="screen")- self.page.screenshot(path=path_to_save)-- new_state = AgentState().new_state()+ await self.page.emulate_media(media="screen")+ await self.page.screenshot(path=path_to_save)+ screenshot = await self.page.screenshot()+ screenshot_bytes = base64.b64encode(screenshot).decode()+ new_state = self.agent.new_state()new_state["internal_monologue"] = "Browsing the web right now..."new_state["browser_session"]["url"] = page_urlnew_state["browser_session"]["screenshot"] = path_to_save- AgentState().add_to_current_state(project_name, new_state)-- return path_to_save+ self.agent.add_to_current_state(project_name, new_state)+ # self.close()+ return path_to_save, screenshot_bytesdef get_html(self):return self.page.content()@@ -74,7 +84,6 @@ class Browser:def extract_text(self):return self.page.evaluate("() => document.body.innerText")- def close(self):- self.page.close()- self.browser.close()- self.playwright.stop()+ async def close(self):+ await self.page.close()+ await self.browser.close()diff --git a/src/browser/search.py b/src/browser/search.pyindex 6b140ff..e624b09 100644--- a/src/browser/search.py+++ b/src/browser/search.py@@ -98,13 +98,13 @@ class DuckDuckGoSearch:def _get_url(self, method, url, data):try:resp = self.asession.request(method, url, data=data)+ if resp.status_code == 200:+ return resp.content+ if resp.status_code == (202, 301, 403):+ raise Exception(f"Error: {resp.status_code} rate limit error")except Exception as error:if "timeout" in str(error).lower():raise TimeoutError("Duckduckgo timed out error")- if resp.status_code == 200:- return resp.content- if resp.status_code == (202, 301, 403):- raise Exception(f"Error: {resp.status_code} rate limit error")def duck(self, query):resp = self._get_url("POST", "https://duckduckgo.com/", data={"q": query})diff --git a/src/llm/llm.py b/src/llm/llm.pyindex 023e70b..f040c38 100644--- a/src/llm/llm.py+++ b/src/llm/llm.py@@ -78,6 +78,7 @@ class LLM:self.update_global_token_usage(prompt, project_name)model_enum = self.model_id_to_enum_mapping().get(self.model_id)+ print(f"Model: {self.model_id}, Enum: {model_enum}")if model_enum is None:raise ValueError(f"Model {self.model_id} not supported")diff --git a/src/socket_instance.py b/src/socket_instance.pyindex f10ece0..51f9d0b 100644--- a/src/socket_instance.py+++ b/src/socket_instance.py@@ -1,14 +1,17 @@# socketio_instance.pyfrom flask_socketio import SocketIOfrom src.logger import Logger-socketio = SocketIO(cors_allowed_origins="*", async_mode="eventlet")+socketio = SocketIO(cors_allowed_origins="*", async_mode="gevent")logger = Logger()-def emit_agent(channel, content):+def emit_agent(channel, content, log=True):try:socketio.emit(channel, content)- logger.info(f"SOCKET {channel} MESSAGE: {content}")+ if log:+ logger.info(f"SOCKET {channel} MESSAGE: {content}")+ return Trueexcept Exception as e:logger.error(f"SOCKET {channel} ERROR: {str(e)}")+ return Falsediff --git a/src/state.py b/src/state.pyindex 80b03dd..4984353 100644--- a/src/state.py+++ b/src/state.py@@ -63,6 +63,7 @@ class AgentState:agent_state = AgentStateModel(project=project, state_stack_json=json.dumps(state_stack))session.add(agent_state)session.commit()+ emit_agent("agent-state", state_stack)def get_current_state(self, project: str):with Session(self.engine) as session:@@ -84,6 +85,7 @@ class AgentState:agent_state = AgentStateModel(project=project, state_stack_json=json.dumps(state_stack))session.add(agent_state)session.commit()+ emit_agent("agent-state", state_stack)def get_latest_state(self, project: str):with Session(self.engine) as session:diff --git a/ui/src/lib/components/BrowserWidget.svelte b/ui/src/lib/components/BrowserWidget.svelteindex 00bf982..6c543e1 100644--- a/ui/src/lib/components/BrowserWidget.svelte+++ b/ui/src/lib/components/BrowserWidget.svelte@@ -1,6 +1,14 @@<script>import { agentState } from "$lib/store";- import { API_BASE_URL } from "$lib/api";+ import { API_BASE_URL, socket } from "$lib/api";++ socket.on('screenshot', function(msg) {+ console.log("screenshot: ", msg);+ const data = msg['data'];+ const img = document.querySelector('.browser-img');+ img.src = `data:image/png;base64,${data}`;+ });+</script><div class="flex flex-col border-2 rounded-lg h-1/2 overflow-y-auto">@@ -23,7 +31,6 @@{#if $agentState?.browser_session.screenshot}<imgclass="browser-img"- src={API_BASE_URL + "/api/get-browser-snapshot?snapshot_path=" + $agentState?.browser_session.screenshot}alt="Browser snapshot"/>{:else}