AutoBrowser/agent.py at main · exponentialXP/AutoBrowser · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
import os
import json
import asyncio
import datetime
from playwright.async_api import async_playwright
import google.generativeai as genai
from utils import resize_image, add_grid, mark_click, SYSTEM_PROMPT
from dotenv import load_dotenv

# Silence GRPC and ABSL logs to prevent confusing error messages
os.environ["GRPC_VERBOSITY"] = "ERROR"
os.environ["GLOG_minloglevel"] = "2"

class WebAgent:
    def __init__(self, api_key=None, model_name="gemini-3-flash-preview", logger=None):
        self.api_key = api_key
        self.model_name = model_name
        self.model = None
        self.history = []
        self.user_data_dir = os.path.join(os.getcwd(), "browser_profile")
        self.session_file = os.path.join(self.user_data_dir, "last_url.txt")
        self.logger = logger # Callback for status updates
        self.paused = False
        self.stopped = False

        # Debug setup
        self.run_id = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
        self.debug_dir = os.path.join(os.getcwd(), "debug", f"run_{self.run_id}")
        os.makedirs(self.debug_dir, exist_ok=True)
        self._init_html_report()

        # Browser state
        self.playwright = None
        self.context = None
        self.page = None

    def log(self, message, type="info"):
        if self.logger:
            asyncio.create_task(self.logger(message, type))
        print(f"[{type.upper()}] {message}")

    def _init_html_report(self):
        report_path = os.path.join(self.debug_dir, "report.html")
        html = """
        <html>
        <head>
            <title>Web Agent Debug Report</title>
            <style>
                body { font-family: sans-serif; background: #1a1a1a; color: #eee; padding: 20px; }
                .step { border: 1px solid #444; margin-bottom: 30px; padding: 15px; border-radius: 8px; background: #2a2a2a; }
                .thought { font-style: italic; color: #aaa; margin-bottom: 15px; }
                .views { display: flex; gap: 20px; }
                .view { flex: 1; }
                img { width: 100%; border-radius: 4px; border: 1px solid #555; }
                h3 { margin-top: 0; color: #00d4ff; }
                .label { font-weight: bold; margin-bottom: 5px; }
            </style>
        </head>
        <body>
            <h1>Agent Run: """ + self.run_id + """</h1>
            <div id="steps"></div>
        </body>
        </html>
        """
        with open(report_path, "w") as f:
            f.write(html)

    def _add_to_report(self, step, thought, ai_view_path, action_results=None):
        report_path = os.path.join(self.debug_dir, "report.html")
        with open(report_path, "r") as f:
            content = f.read()

        actions_html = ""
        if action_results:
            for i, res_path in enumerate(action_results):
                actions_html += f'<div class="view"><div class="label">Action {i} Verification</div><img src="{os.path.basename(res_path)}"></div>'

        new_step = f"""
        <div class="step">
            <h3>Step {step}</h3>
            <div class="thought"><b>Thought:</b> {thought}</div>
            <div class="views">
                <div class="view"><div class="label">AI View (Grid Version)</div><img src="{os.path.basename(ai_view_path)}"></div>
                {actions_html}
            </div>
        </div>
        """
        content = content.replace('<div id="steps"></div>', new_step + '<div id="steps"></div>')
        with open(report_path, "w") as f:
            f.write(content)

    async def start_browser(self, url=None):
        """Initializes the browser and persists the context."""
        if self.page:
            return

        if not url:
            if os.path.exists(self.session_file):
                with open(self.session_file, "r") as f:
                    url = f.read().strip()

            if not url or not url.startswith("http"):
                url = "https://www.google.com"

        self.log(f"Starting browser at {url}...")

        self.playwright = await async_playwright().start()
        self.context = await self.playwright.chromium.launch_persistent_context(
            self.user_data_dir,
            channel="msedge",
            headless=False,
            viewport={'width': 1280, 'height': 720}
        )
        self.page = self.context.pages[0] if self.context.pages else await self.context.new_page()
        await self.page.goto(url)
        self.log("Browser ready. Agent taking over.")

    async def stop_browser(self):
        """Stops the browser and playwright."""
        if self.context:
            await self.context.close()
        if self.playwright:
            await self.playwright.stop()
        self.page = None
        self.context = None
        self.playwright = None

    async def run(self, task, api_key=None):
        # Configure API key and model if provided or if not already set
        current_key = api_key or self.api_key
        if not current_key:
            self.log("No API key provided. Please enter your Gemini API key.", "error")
            return

        if current_key != self.api_key or not self.model:
            self.api_key = current_key
            genai.configure(api_key=self.api_key)
            self.model = genai.GenerativeModel(self.model_name)

        # Ensure browser is started
        if not self.page:
            await self.start_browser()

        self.log(f"Starting task: {task}")

        try:
            for step in range(30): # Increased steps for longer tasks if needed
                # Wait loop if paused
                while self.paused and not self.stopped:
                    await asyncio.sleep(0.5)

                if self.stopped:
                    self.log("Agent stopped by user.", "warning")
                    self.stopped = False # Reset for next time
                    break

                if self.page.is_closed():
                    # Try to find another page if this one was closed
                    if self.context.pages:
                        self.page = self.context.pages[0]
                        if self.page.is_closed(): break
                    else:
                        self.log("All browser pages were closed. Exiting.", "error")
                        break

                try:
                    # 1. Take a screenshot
                    self.log(f"Step {step}: Analyzing screen...")
                    raw_screenshot = await self.page.screenshot()

                    # Process for AI (Resize + Grid)
                    resized_screenshot = resize_image(raw_screenshot)
                    grid_screenshot = add_grid(resized_screenshot)

                    # Save original resized for marking
                    debug_img = resized_screenshot

                    prompt = [
                        SYSTEM_PROMPT,
                        f"User Task: {task}",
                        f"History: {json.dumps(self.history[-3:])}", # Only send last 3 steps to save tokens
                        {
                            "mime_type": "image/jpeg",
                            "data": grid_screenshot
                        },
                        "Respond in JSON. Be precise with coordinates using the grid."
                    ]

                    # 3. Get response from Gemini with Retry logic
                    response = None
                    max_retries = 10
                    empty_retries = 0
                    for attempt in range(max_retries):
                        try:
                            response = self.model.generate_content(prompt)

                            # Check for empty response
                            if not response or not response.candidates or not response.candidates[0].content.parts:
                                if empty_retries < 2:
                                    empty_retries += 1
                                    self.log(f"Empty response received. Retrying ({empty_retries}/2)...", "warning")
                                    await asyncio.sleep(2)
                                    continue
                                else:
                                    self.log("AI returned empty content after retries.", "error")
                                    break

                            break
                        except Exception as e:
                            if "429" in str(e) or "ResourceExhausted" in str(e):
                                wait_time = (2 ** attempt) + 3 # 4, 5, 7, 11, 19...
                                self.log(f"Rate limit reached. Trying again in {wait_time}s", "warning")
                                await asyncio.sleep(wait_time)
                            else:
                                raise e

                    if not response or not response.candidates or not response.candidates[0].content.parts:
                        if not self.stopped:
                            self.log("Fatal Error: Could not get valid AI response. Stopping agent.", "error")
                        break

                    text = response.text.strip()
                    if "```json" in text:
                        text = text.split("```json")[-1].split("```")[0].strip()
                    elif "```" in text:
                        text = text.split("```")[-1].split("```")[0].strip()

                    res_json = json.loads(text)
                    self.history.append(res_json)
                    self.history = self.history[-10:] # Keep only last 10 steps
                    thought = res_json.get("thought", "None")
                    self.log(f"AI Thought: {thought}")

                    # Save session state (last URL)
                    try:
                        current_url = self.page.url
                        with open(self.session_file, "w") as f:
                            f.write(current_url)
                    except:
                        pass

                    # Save what the AI SAW
                    ai_view_path = os.path.join(self.debug_dir, f"step_{step}_ai.jpg")
                    with open(ai_view_path, "wb") as f:
                        f.write(grid_screenshot)

                    # Handle both single 'action' and multiple 'actions'
                    actions = res_json.get("actions", [])
                    if not actions and res_json.get("action"):
                        actions = [{"action": res_json.get("action"), "params": res_json.get("params", [])}]

                    action_results = []
                    for i, act_obj in enumerate(actions):
                        if self.paused or self.stopped:
                            break

                        if self.page.is_closed(): break

                        action = act_obj.get("action")
                        params = act_obj.get("params", [])

                        if action == "click" or action == "type" or action == "paste":
                            if action == "click":
                                x_pct, y_pct = params[0], params[1]
                            else:
                                x_pct, y_pct = params[1], params[2]

                            # Save debug image
                            marked_img = mark_click(debug_img, x_pct, y_pct)
                            click_view_path = os.path.join(self.debug_dir, f"step_{step}_click_{i}.jpg")
                            with open(click_view_path, "wb") as f:
                                f.write(marked_img)
                            action_results.append(click_view_path)
                            self.log(f"Performing {action} at ({x_pct}, {y_pct}).")

                            # Perform action
                            await self.page.mouse.click(x_pct * 1280 / 1000, y_pct * 720 / 1000)
                            if action == "type":
                                await self.page.keyboard.type(params[0])
                                await self.page.keyboard.press("Enter")
                            elif action == "paste":
                                await self.page.keyboard.insert_text(params[0])
                                await self.page.keyboard.press("Enter")

                        elif action == "scroll":
                            direction = params[0]
                            self.log(f"Scrolling {direction}...")
                            await self.page.focus("body")
                            await self.page.keyboard.press("PageDown" if direction == "down" else "PageUp")
                        elif action == "ask_user":
                            self.log(f"ASKING USER: {params[0]}", "warning")
                            # In the new UI mode, we should ideally wait for a message back.
                            # For now, we'll just log it and pause the loop in a non-blocking way if possible.
                            # But standard async agentic flow often skips standard input.
                            break
                        elif action == "wait":
                            self.log("Waiting for page to load...")
                            await asyncio.sleep(2)
                        elif action == "finish":
                            self.log("Task finished successfully!", "success")
                            self._add_to_report(step, thought, ai_view_path, action_results)
                            return

                        await asyncio.sleep(1.5)

                    # Add step to HTML report
                    self._add_to_report(step, thought, ai_view_path, action_results)
                except json.JSONDecodeError as je:
                    self.log(f"AI response format error: {je}", "error")
                    break
                except Exception as e:
                    if "Target page, context or browser has been closed" in str(e):
                        self.log("Browser window closed unexpectedly.", "error")
                    else:
                        self.log(f"Error: {e}", "error")
                        import traceback
                        traceback.print_exc()
                    break
        except Exception as e:
            self.log(f"Critical loop error: {e}", "error")
        # Removed context.close() from finally to keep browser open

if __name__ == "__main__":
    import sys
    task = "Find the best place to live in the world"
    if len(sys.argv) > 1:
        task = " ".join(sys.argv[1:])

    agent = WebAgent()
    asyncio.run(agent.run(task))