semantio 0.0.3__py3-none-any.whl → 0.0.4__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
semantio/agent.py CHANGED
@@ -56,8 +56,11 @@ class Agent(BaseModel):
56
56
  super().__init__(**kwargs)
57
57
  # Initialize the model and tools here if needed
58
58
  self._initialize_model()
59
- # Automatically discover and register tools if not provided
59
+ # Initialize tools as an empty list if not provided
60
60
  if self.tools is None:
61
+ self.tools = []
62
+ # Automatically discover and register tools if auto tool is enabled
63
+ if self.auto_tool and not self.tools:
61
64
  self.tools = self._discover_tools()
62
65
  # Pass the LLM instance to each tool
63
66
  for tool in self.tools:
@@ -213,7 +216,6 @@ class Agent(BaseModel):
213
216
  message: Optional[Union[str, Image, List, Dict]] = None,
214
217
  stream: bool = False,
215
218
  markdown: bool = False,
216
- tools: Optional[List[BaseTool]] = None,
217
219
  team: Optional[List['Agent']] = None,
218
220
  **kwargs,
219
221
  ) -> Union[str, Dict]: # Add return type hint
@@ -228,7 +230,7 @@ class Agent(BaseModel):
228
230
  return response
229
231
  else:
230
232
  # Generate and return the response
231
- response = self._generate_response(message, markdown=markdown, tools=tools, team=team, **kwargs)
233
+ response = self._generate_response(message, markdown=markdown, team=team, **kwargs)
232
234
  print(response) # Print the response to the console
233
235
  return response
234
236
 
@@ -245,43 +247,6 @@ class Agent(BaseModel):
245
247
  if self.tools is None:
246
248
  self.tools = []
247
249
  self.tools.append(tool)
248
-
249
- def _detect_tool_call(self, message: str) -> Optional[Dict[str, Any]]:
250
- """
251
- Use the LLM to detect which tool should be called based on the user's query.
252
- """
253
- if not self.tools:
254
- logger.warning("No tools available to detect.")
255
- return None
256
-
257
- # Create a prompt for the LLM
258
- prompt = f"""
259
- You are an AI agent that helps users by selecting the most appropriate tool to answer their query. Below is a list of available tools and their functionalities:
260
-
261
- {self._get_tool_descriptions()}
262
-
263
- Based on the user's query, select the most appropriate tool. Respond with the name of the tool (e.g., "CryptoPriceChecker"). If no tool is suitable, respond with "None".
264
-
265
- User Query: "{message}"
266
- """
267
-
268
- try:
269
- # Call the LLM to generate the response
270
- response = self.llm_instance.generate(prompt=prompt)
271
- tool_name = response.strip().replace('"', '').replace("'", "")
272
-
273
- # Find the tool in the list of available tools
274
- tool = next((t for t in self.tools if t.name.lower() == tool_name.lower()), None)
275
- if tool:
276
- logger.info(f"Detected tool call: {tool.name}")
277
- return {
278
- "tool": tool.name,
279
- "input": {"query": message}
280
- }
281
- except Exception as e:
282
- logger.error(f"Failed to detect tool call: {e}")
283
-
284
- return None
285
250
 
286
251
  def _analyze_query_and_select_tools(self, query: str) -> List[Dict[str, Any]]:
287
252
  """
@@ -324,17 +289,16 @@ class Agent(BaseModel):
324
289
  return []
325
290
 
326
291
 
327
- def _generate_response(self, message: str, markdown: bool = False, tools: Optional[List[BaseTool]] = None, team: Optional[List['Agent']] = None, **kwargs) -> str:
292
+ def _generate_response(self, message: str, markdown: bool = False, team: Optional[List['Agent']] = None, **kwargs) -> str:
328
293
  """Generate the agent's response, including tool execution and context retrieval."""
329
- # Use the specified tools or team if provided
330
- if tools is not None:
331
- self.tools = tools
294
+ # Use the specified team if provided
332
295
  if team is not None:
333
296
  return self._generate_team_response(message, team, markdown=markdown, **kwargs)
334
297
 
335
298
  # Initialize tool_outputs as an empty dictionary
336
299
  tool_outputs = {}
337
300
  responses = []
301
+ tool_calls = []
338
302
 
339
303
  # Use the LLM to analyze the query and dynamically select tools when auto_tool is enabled
340
304
  if self.auto_tool:
@@ -344,7 +308,7 @@ class Agent(BaseModel):
344
308
  if self.tools:
345
309
  tool_calls = [
346
310
  {
347
- "tool": tool.__class__.__name__,
311
+ "tool": tool.name,
348
312
  "input": {
349
313
  "query": message, # Use the message as the query
350
314
  "context": None, # No context provided by default
@@ -352,10 +316,8 @@ class Agent(BaseModel):
352
316
  }
353
317
  for tool in self.tools
354
318
  ]
355
- else:
356
- tool_calls = kwargs.get("tool_calls", [])
357
319
 
358
- # Execute tools if any are detected
320
+ # Execute tools if any are detected
359
321
  if tool_calls:
360
322
  for tool_call in tool_calls:
361
323
  tool_name = tool_call["tool"]
@@ -396,7 +358,7 @@ class Agent(BaseModel):
396
358
  except Exception as e:
397
359
  logger.error(f"Failed to generate LLM response: {e}")
398
360
  responses.append(f"An error occurred while generating the analysis: {e}")
399
- if not tool_calls:
361
+ if not self.tools and not tool_calls:
400
362
  # If no tools were executed, proceed with the original logic
401
363
  # Retrieve relevant context using RAG
402
364
  rag_context = self.rag.retrieve(message) if self.rag else None
@@ -0,0 +1,271 @@
1
+ # web_browser.py
2
+ from typing import Dict, Any, List, Optional
3
+ from pydantic import Field, BaseModel
4
+ from selenium import webdriver
5
+ from selenium.webdriver.common.by import By
6
+ from selenium.webdriver.support.ui import WebDriverWait
7
+ from selenium.webdriver.support import expected_conditions as EC
8
+ from selenium.webdriver.chrome.options import Options
9
+ from selenium.webdriver.chrome.service import Service
10
+ from webdriver_manager.chrome import ChromeDriverManager
11
+ from bs4 import BeautifulSoup
12
+ import json
13
+ import time
14
+ import re
15
+ import logging
16
+ from .base_tool import BaseTool
17
+
18
+ logger = logging.getLogger(__name__)
19
+
20
+ class BrowserPlan(BaseModel):
21
+ tasks: List[Dict[str, Any]] = Field(
22
+ ...,
23
+ description="List of automation tasks to execute"
24
+ )
25
+
26
+ class WebBrowserTool(BaseTool):
27
+ name: str = Field("WebBrowser", description="Name of the tool")
28
+ description: str = Field(
29
+ "Universal web automation tool for dynamic website interactions",
30
+ description="Tool description"
31
+ )
32
+
33
+ def execute(self, input: Dict[str, Any]) -> Dict[str, Any]:
34
+ """Execute dynamic web automation workflow"""
35
+ driver = None
36
+ try:
37
+ driver = self._init_browser(input.get("headless", False))
38
+ results = []
39
+ current_url = ""
40
+
41
+ # Generate initial plan
42
+ plan = self._generate_plan(input['query'], current_url)
43
+
44
+ for task in plan.tasks:
45
+ result = self._execute_safe_task(driver, task)
46
+ results.append(result)
47
+
48
+ if not result['success']:
49
+ break
50
+
51
+ # Update context for next tasks
52
+ current_url = driver.current_url
53
+
54
+ return {"status": "success", "results": results}
55
+
56
+ except Exception as e:
57
+ return {"status": "error", "message": str(e)}
58
+ finally:
59
+ if driver:
60
+ driver.quit()
61
+
62
+ def _init_browser(self, headless: bool) -> webdriver.Chrome:
63
+ """Initialize browser with advanced options"""
64
+ options = Options()
65
+ options.add_argument("--start-maximized")
66
+ options.add_argument("--disable-blink-features=AutomationControlled")
67
+ options.add_experimental_option("excludeSwitches", ["enable-automation"])
68
+
69
+ if headless:
70
+ options.add_argument("--headless=new")
71
+
72
+ return webdriver.Chrome(
73
+ service=Service(ChromeDriverManager().install()),
74
+ options=options
75
+ )
76
+
77
+ def _generate_plan(self, query: str, current_url: str) -> BrowserPlan:
78
+ """Generate adaptive execution plan using LLM"""
79
+ prompt = f"""Generate browser automation plan for: {query}
80
+
81
+ Current URL: {current_url or 'No page loaded yet'}
82
+
83
+ Required JSON format:
84
+ {{
85
+ "tasks": [
86
+ {{
87
+ "action": "navigate|click|type|wait|scroll",
88
+ "selector": "CSS selector (optional)",
89
+ "value": "input text/URL/seconds",
90
+ "description": "action purpose"
91
+ }}
92
+ ]
93
+ }}
94
+
95
+ Guidelines:
96
+ 1. Prefer IDs in selectors (#element-id)
97
+ 2. Use semantic attributes (aria-label, name)
98
+ 3. Include wait steps after navigation
99
+ 4. Prioritize visible elements
100
+ 5. Add scroll steps for hidden elements
101
+ """
102
+
103
+ response = self.llm.generate(prompt=prompt)
104
+ return self._parse_plan(response)
105
+
106
+ def _parse_plan(self, response: str) -> BrowserPlan:
107
+ """Robust JSON parsing with multiple fallback strategies"""
108
+ try:
109
+ # Try extracting JSON from markdown code block
110
+ json_match = re.search(r'```json\n?(.+?)\n?```', response, re.DOTALL)
111
+ if json_match:
112
+ plan_data = json.loads(json_match.group(1).strip())
113
+ else:
114
+ # Fallback to extract first JSON object
115
+ json_str = re.search(r'\{.*\}', response, re.DOTALL).group()
116
+ plan_data = json.loads(json_str)
117
+
118
+ # Validate tasks structure
119
+ validated_tasks = []
120
+ for task in plan_data.get("tasks", []):
121
+ if not all(key in task for key in ["action", "description"]):
122
+ continue
123
+ validated_tasks.append({
124
+ "action": task["action"],
125
+ "selector": task.get("selector", ""),
126
+ "value": task.get("value", ""),
127
+ "description": task["description"]
128
+ })
129
+
130
+ return BrowserPlan(tasks=validated_tasks)
131
+
132
+ except (json.JSONDecodeError, AttributeError) as e:
133
+ logger.error(f"Plan parsing failed: {e}")
134
+ return BrowserPlan(tasks=[])
135
+
136
+ def _execute_safe_task(self, driver, task: Dict) -> Dict[str, Any]:
137
+ """Execute task with comprehensive error handling"""
138
+ try:
139
+ action = task["action"].lower()
140
+ selector = task.get("selector", "")
141
+ value = task.get("value", "")
142
+
143
+ if action == "navigate":
144
+ return self._handle_navigation(driver, value)
145
+
146
+ elif action == "click":
147
+ return self._handle_click(driver, selector)
148
+
149
+ elif action == "type":
150
+ return self._handle_typing(driver, selector, value)
151
+
152
+ elif action == "wait":
153
+ return self._handle_wait(value)
154
+
155
+ elif action == "scroll":
156
+ return self._handle_scroll(driver, selector)
157
+
158
+ return {
159
+ "action": action,
160
+ "success": False,
161
+ "message": f"Unsupported action: {action}"
162
+ }
163
+
164
+ except Exception as e:
165
+ return {
166
+ "action": action,
167
+ "success": False,
168
+ "message": f"Critical error: {str(e)}"
169
+ }
170
+
171
+ def _handle_navigation(self, driver, url: str) -> Dict[str, Any]:
172
+ """Smart navigation handler"""
173
+ if not url.startswith(("http://", "https://")):
174
+ url = f"https://{url}"
175
+
176
+ try:
177
+ driver.get(url)
178
+ WebDriverWait(driver, 15).until(
179
+ EC.presence_of_element_located((By.TAG_NAME, "body"))
180
+ )
181
+ return {
182
+ "action": "navigate",
183
+ "success": True,
184
+ "message": f"Navigated to {url}"
185
+ }
186
+ except Exception as e:
187
+ return {
188
+ "action": "navigate",
189
+ "success": False,
190
+ "message": f"Navigation failed: {str(e)}"
191
+ }
192
+
193
+ def _handle_click(self, driver, selector: str) -> Dict[str, Any]:
194
+ """Dynamic click handler"""
195
+ try:
196
+ element = WebDriverWait(driver, 15).until(
197
+ EC.element_to_be_clickable((By.CSS_SELECTOR, selector))
198
+ )
199
+ driver.execute_script("arguments[0].scrollIntoView({behavior: 'smooth'});", element)
200
+ element.click()
201
+ return {
202
+ "action": "click",
203
+ "success": True,
204
+ "message": f"Clicked element: {selector}"
205
+ }
206
+ except Exception as e:
207
+ return {
208
+ "action": "click",
209
+ "success": False,
210
+ "message": f"Click failed: {str(e)}"
211
+ }
212
+
213
+ def _handle_typing(self, driver, selector: str, text: str) -> Dict[str, Any]:
214
+ """Universal typing handler"""
215
+ try:
216
+ element = WebDriverWait(driver, 15).until(
217
+ EC.presence_of_element_located((By.CSS_SELECTOR, selector))
218
+ )
219
+ element.clear()
220
+ element.send_keys(text)
221
+ return {
222
+ "action": "type",
223
+ "success": True,
224
+ "message": f"Typed '{text}' into {selector}"
225
+ }
226
+ except Exception as e:
227
+ return {
228
+ "action": "type",
229
+ "success": False,
230
+ "message": f"Typing failed: {str(e)}"
231
+ }
232
+
233
+ def _handle_wait(self, seconds: str) -> Dict[str, Any]:
234
+ """Configurable wait handler"""
235
+ try:
236
+ wait_time = float(seconds)
237
+ time.sleep(wait_time)
238
+ return {
239
+ "action": "wait",
240
+ "success": True,
241
+ "message": f"Waited {wait_time} seconds"
242
+ }
243
+ except ValueError:
244
+ return {
245
+ "action": "wait",
246
+ "success": False,
247
+ "message": "Invalid wait time"
248
+ }
249
+
250
+ def _handle_scroll(self, driver, selector: str) -> Dict[str, Any]:
251
+ """Smart scroll handler"""
252
+ try:
253
+ if selector:
254
+ element = WebDriverWait(driver, 15).until(
255
+ EC.presence_of_element_located((By.CSS_SELECTOR, selector))
256
+ )
257
+ driver.execute_script("arguments[0].scrollIntoView({behavior: 'smooth'});", element)
258
+ else:
259
+ driver.execute_script("window.scrollTo(0, document.body.scrollHeight);")
260
+
261
+ return {
262
+ "action": "scroll",
263
+ "success": True,
264
+ "message": f"Scrolled to {selector or 'page bottom'}"
265
+ }
266
+ except Exception as e:
267
+ return {
268
+ "action": "scroll",
269
+ "success": False,
270
+ "message": f"Scroll failed: {str(e)}"
271
+ }
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.1
2
2
  Name: semantio
3
- Version: 0.0.3
3
+ Version: 0.0.4
4
4
  Summary: A powerful SDK for building AI agents
5
5
  Home-page: https://github.com/Syenah/semantio
6
6
  Author: Rakesh
@@ -33,6 +33,10 @@ Requires-Dist: sentence-transformers
33
33
  Requires-Dist: fuzzywuzzy
34
34
  Requires-Dist: duckduckgo-search
35
35
  Requires-Dist: yfinance
36
+ Requires-Dist: selenium
37
+ Requires-Dist: beautifulsoup4
38
+ Requires-Dist: webdriver-manager
39
+ Requires-Dist: validators
36
40
 
37
41
  # Semantio: The Mother of Your AI Agents
38
42
 
@@ -1,5 +1,5 @@
1
1
  semantio/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
2
- semantio/agent.py,sha256=plQ4D76cnJ1FaGlEuKDeA53aW_hMDvt5sbmUuTHqvFQ,30143
2
+ semantio/agent.py,sha256=hKytSI5LqNnxqVvwI2hOINqPgrdhUXY9MS_90_crZPs,28584
3
3
  semantio/memory.py,sha256=eNAwyAokppHzMcIyFgOw2hT2wnLQBd9GL4T5eallNV4,281
4
4
  semantio/rag.py,sha256=ROy3Pa1NURcDs6qQZ8IMoa5Xlzt6I-msEq0C1p8UgB0,472
5
5
  semantio/api/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -27,15 +27,16 @@ semantio/tools/base_tool.py,sha256=xBNSa_8a8WmA4BGRLG2dE7wj9GnBcZo7-P2SyD86GvY,5
27
27
  semantio/tools/crypto.py,sha256=mut1ztvpPcUUP3b563dh_FmKtP68KmNis3Qm8WENj8w,5559
28
28
  semantio/tools/duckduckgo.py,sha256=6mGn0js0cIsVxQlAgB8AYNLP05H8WmJKnSVosiO9iH0,5034
29
29
  semantio/tools/stocks.py,sha256=BVuK61O9OmWQjj0YdiCJY6TzpiFJ_An1UJB2RkDfX2k,5393
30
+ semantio/tools/web_browser.py,sha256=wqr5pj2GybkK9IHDb8C1BipS8ujV2l36WlwA8ZbKd88,9711
30
31
  semantio/utils/__init__.py,sha256=Lx4X4iJpRhZzRmpQb80XXh5Ve8ZMOkadWAxXSmHpO_8,244
31
32
  semantio/utils/config.py,sha256=ZTwUTqxjW3-w94zoU7GzivWyJe0JJGvBfuB4RUOuEs8,1198
32
33
  semantio/utils/date_utils.py,sha256=x3oqRGv6ee_KCJ0LvCqqZh_FSgS6YGOHBwZQS4TJetY,1471
33
34
  semantio/utils/file_utils.py,sha256=b_cMuJINEGk9ikNuNHSn9lsmICWwvtnCDZ03ndH_S2I,1779
34
35
  semantio/utils/logger.py,sha256=TmGbP8BRjLMWjXi2GWzZ0RIXt70x9qX3FuIqghCNlwM,510
35
36
  semantio/utils/validation_utils.py,sha256=iwoxEb4Q5ILqV6tbesMjPWPCCoL3AmPLejGUy6q8YvQ,1284
36
- semantio-0.0.3.dist-info/LICENSE,sha256=teQbWD2Zlcl1_Fo29o2tNbs6G26hbCQiUzds5fQGYlY,1063
37
- semantio-0.0.3.dist-info/METADATA,sha256=M5Q-waTknpyWrD_HV9G76jMKgPHPrBBwM5Hl8we4ulo,6800
38
- semantio-0.0.3.dist-info/WHEEL,sha256=ewwEueio1C2XeHTvT17n8dZUJgOvyCWCt0WVNLClP9o,92
39
- semantio-0.0.3.dist-info/entry_points.txt,sha256=zbPgevSLwcLpdRHqI_atE8EOt8lK2vRF1AoDflDTo18,53
40
- semantio-0.0.3.dist-info/top_level.txt,sha256=Yte_6mb-bh-I_lQwMjk1GijZkxPoX4Zmp3kBftC1ZlA,9
41
- semantio-0.0.3.dist-info/RECORD,,
37
+ semantio-0.0.4.dist-info/LICENSE,sha256=teQbWD2Zlcl1_Fo29o2tNbs6G26hbCQiUzds5fQGYlY,1063
38
+ semantio-0.0.4.dist-info/METADATA,sha256=youxODbkR3gNERG-mD7zbUbe5ix-0lUiWCHUI1_Y5IY,6913
39
+ semantio-0.0.4.dist-info/WHEEL,sha256=ewwEueio1C2XeHTvT17n8dZUJgOvyCWCt0WVNLClP9o,92
40
+ semantio-0.0.4.dist-info/entry_points.txt,sha256=zbPgevSLwcLpdRHqI_atE8EOt8lK2vRF1AoDflDTo18,53
41
+ semantio-0.0.4.dist-info/top_level.txt,sha256=Yte_6mb-bh-I_lQwMjk1GijZkxPoX4Zmp3kBftC1ZlA,9
42
+ semantio-0.0.4.dist-info/RECORD,,