Merge pull request #76 from Fosowl/dev

Integration of better router + Integration of planner agent

.env.example

@@ -1,2 +1,3 @@
 SEARXNG_BASE_URL="http://127.0.0.1:8080"
-OPENAI_API_KEY='dont share this, not needed for local providers'
+OPENAI_API_KEY='dont share this, not needed for local providers'
+TOKENIZERS_PARALLELISM=False

.gitignore

@@ -1,4 +1,5 @@
 *.wav
+*.safetensors
 config.ini
 *.egg-info
 experimental/

.pre-commit-config.yaml

@@ -4,6 +4,6 @@ repos:
       - id: trufflehog
         name: TruffleHog
         description: Detect secrets in your data.
-        entry: bash -c 'trufflehog git file://. --since-commit HEAD --results=verified,unknown --fail --no-update'
+        entry: bash -c 'trufflehog git file://. --since-commit HEAD --results=verified,unknown --no-update'
         language: system
         stages: ["commit", "push"]

README.md

@@ -12,40 +12,40 @@
 
 > *Do a web search to find tech startup in Japan working on cutting edge AI research*
 
-> *Make a snake game in Python*
+> *Can you make a tetris game in C ?*
 
-> *Scan my network with nmap, find out who is connected?*
+> *Can you find where is contract.pdf*?
 
-> *Hey can you find where is contract.pdf*?
+
+### Browse the web
+
+![alt text](./media/exemples/search_startup.png)
+
+### Code hand free
+
+![alt text](./media/exemples/matmul_golang.png)
+
+### Plan and execute with agents (Experimental)
+
+![alt text](./media/exemples/plan_weather_app.png)
+
+*See media/examples for other use case screenshots.*
 
 ## Features:
 
 - **100% Local**: No cloud, runs on your hardware. Your data stays yours.
 
-- **Voice interaction**: Voice-enabled natural interaction.
-
 - **Filesystem interaction**: Use bash to navigate and manipulate your files effortlessly.
 
-- **Code what you ask**: Can write, debug, and run code in Python, C, Golang and more languages on the way.
+- **Autonomous Coding**: Can write, debug, and run code in Python, C, Golang and more languages on the way.
-
-- **Autonomous**: If a command flops or code breaks, it retries and fixes it by itself.
 
 - **Agent routing**: Automatically picks the right agent for the job.
 
-- **Divide and Conquer**: For big tasks, spins up multiple agents to plan and execute.
+- **Planning**: For complex tasks, spins up multiple agents to plan and execute.
 
-- **Tool-Equipped**: From basic search to flight APIs and file exploration, every agent has it's own tools.
+- **Autonomous Web Browsing**: Autonomous web navigation.
 
-- **Memory**: Remembers what’s useful, your preferences and past sessions conversation.
+- **Memory**: Efficient memory and sessions management. 
-
-- **Web Browsing**: Autonomous web navigation.
-
-
-### Searching the web with agenticSeek :
-
-![alt text](./media/exemples/search_startup.png)
-
-*See media/examples for other use case screenshots.*
 
 ---
 
@@ -185,11 +185,11 @@ Here are some example usage:
 
 ### Casual
 
-> *Tell me a joke*
+> *Tell me about France*
 
-> *Where is flight ABC777 ? my mom is on that plane*
+> *What is the meaning of life ?*
 
-> *what is the meaning of life ?*
+> *Should I take creatine before or after workout?*
 
 
 After you type your query, agenticSeek will allocate the best agent for the task.

config.ini

@@ -4,8 +4,9 @@ provider_name = ollama
 provider_model = deepseek-r1:14b
 provider_server_address = 127.0.0.1:11434
 agent_name = Friday
-recover_last_session = True
+recover_last_session = False
 save_session = False
-speak = True
+speak = False
 listen = False
-work_dir =  /Users/mlg/Documents/ai_workplace
+work_dir =  /Users/mlg/Documents/ai_folder
+headless_browser = False

install.bat

@@ -0,0 +1,12 @@
+@echo off
+set SCRIPTS_DIR=scripts
+set LLM_ROUTER_DIR=llm_router
+
+if exist "%SCRIPTS_DIR%\windows_install.bat" (
+    echo Running Windows installation script...
+    call "%SCRIPTS_DIR%\windows_install.bat"
+    cd "%LLM_ROUTER_DIR%" && call dl_safetensors.bat
+) else (
+    echo Error: %SCRIPTS_DIR%\windows_install.bat not found!
+    exit /b 1
+)

install.sh

@@ -1,6 +1,7 @@
 #!/bin/bash
 
 SCRIPTS_DIR="scripts"
+LLM_ROUTER_DIR="llm_router"
 
 echo "Detecting operating system..."
 
@@ -12,6 +13,7 @@ case "$OS_TYPE" in
         if [ -f "$SCRIPTS_DIR/linux_install.sh" ]; then
             echo "Running Linux installation script..."
             bash "$SCRIPTS_DIR/linux_install.sh"
+            bash -c "cd $LLM_ROUTER_DIR && ./dl_safetensors.sh"
         else
             echo "Error: $SCRIPTS_DIR/linux_install.sh not found!"
             exit 1
@@ -22,24 +24,15 @@ case "$OS_TYPE" in
         if [ -f "$SCRIPTS_DIR/macos_install.sh" ]; then
             echo "Running macOS installation script..."
             bash "$SCRIPTS_DIR/macos_install.sh"
+            bash -c "cd $LLM_ROUTER_DIR && ./dl_safetensors.sh"
         else
             echo "Error: $SCRIPTS_DIR/macos_install.sh not found!"
             exit 1
         fi
         ;;
-    "MINGW"* | "MSYS"* | "CYGWIN"*)
-        echo "Detected Windows (via Bash-like environment)"
-        if [ -f "$SCRIPTS_DIR/windows_install.sh" ]; then
-            echo "Running Windows installation script..."
-            bash "$SCRIPTS_DIR/windows_install.sh"
-        else
-            echo "Error: $SCRIPTS_DIR/windows_install.sh not found!"
-            exit 1
-        fi
-        ;;
     *)
         echo "Unsupported OS detected: $OS_TYPE"
-        echo "This script supports Linux, macOS, and Windows (via Bash-compatible environments)."
+        echo "This script supports only Linux and macOS."
         exit 1
         ;;
 esac

llm_router/config.json

@@ -0,0 +1,33 @@
+{
+  "config": {
+    "batch_size": 32,
+    "device_map": "auto",
+    "early_stopping_patience": 3,
+    "epochs": 10,
+    "ewc_lambda": 100.0,
+    "gradient_checkpointing": false,
+    "learning_rate": 0.0005,
+    "max_examples_per_class": 500,
+    "max_length": 512,
+    "min_confidence": 0.1,
+    "min_examples_per_class": 3,
+    "neural_weight": 0.2,
+    "num_representative_examples": 5,
+    "prototype_update_frequency": 50,
+    "prototype_weight": 0.8,
+    "quantization": null,
+    "similarity_threshold": 0.7,
+    "warmup_steps": 0
+  },
+  "embedding_dim": 768,
+  "id_to_label": {
+    "0": "HIGH",
+    "1": "LOW"
+  },
+  "label_to_id": {
+    "HIGH": 0,
+    "LOW": 1
+  },
+  "model_name": "distilbert/distilbert-base-cased",
+  "train_steps": 20
+}

llm_router/dl_safetensors.sh

@@ -0,0 +1,33 @@
+##########
+# Dummy script to download the model
+# Because dowloading with hugging face does not seem to work, maybe I am doing something wrong?
+# AdaptiveClassifier.from_pretrained("adaptive-classifier/llm-router") ----> result in config.json not found
+# Therefore, I put all the files in llm_router and download the model file with this script, If you know a better way please raise an issue
+#########
+
+#!/bin/bash
+
+# Define the URL and filename
+URL="https://huggingface.co/adaptive-classifier/llm_router/resolve/main/model.safetensors"
+FILENAME="model.safetensors"
+
+if [ ! -f "$FILENAME" ]; then
+    echo "Router safetensors file not found, downloading..."
+    if command -v curl >/dev/null 2>&1; then
+        curl -L -o "$FILENAME" "$URL"
+    elif command -v wget >/dev/null 2>&1; then
+        wget -O "$FILENAME" "$URL"
+    else
+        echo "Error: Neither curl nor wget is available. Please install one of them."
+        exit 1
+    fi
+    
+    if [ $? -eq 0 ]; then
+        echo "Download completed successfully"
+    else
+        echo "Download failed"
+        exit 1
+    fi
+else
+    echo "File already exists, skipping download"
+fi

main.py

@@ -8,6 +8,7 @@ import configparser
 from sources.llm_provider import Provider
 from sources.interaction import Interaction
 from sources.agents import Agent, CoderAgent, CasualAgent, FileAgent, PlannerAgent, BrowserAgent
+from sources.browser import Browser, create_driver
 
 import warnings
 warnings.filterwarnings("ignore")
@@ -28,6 +29,8 @@ def main():
                                    model=config["MAIN"]["provider_model"],
                                    server_address=config["MAIN"]["provider_server_address"])
 
+    browser = Browser(create_driver(), headless=config.getboolean('MAIN', 'headless_browser'))
+
     agents = [
         CasualAgent(name=config["MAIN"]["agent_name"],
                     prompt_path="prompts/casual_agent.txt",
@@ -40,12 +43,17 @@ def main():
                   provider=provider, verbose=False),
         BrowserAgent(name="Browser",
                      prompt_path="prompts/browser_agent.txt",
-                     provider=provider, verbose=False)
+                     provider=provider, verbose=False, browser=browser),
+        # Planner agent is experimental, might work poorly, especially with model < 32b
+        PlannerAgent(name="Planner",
+                     prompt_path="prompts/planner_agent.txt",
+                     provider=provider, verbose=False, browser=browser)
     ]
 
-    interaction = Interaction(agents, tts_enabled=config.getboolean('MAIN', 'speak'),
+    interaction = Interaction(agents,
-                                      stt_enabled=config.getboolean('MAIN', 'listen'),
+                              tts_enabled=config.getboolean('MAIN', 'speak'),
-                                      recover_last_session=config.getboolean('MAIN', 'recover_last_session'))
+                              stt_enabled=config.getboolean('MAIN', 'listen'),
+                              recover_last_session=config.getboolean('MAIN', 'recover_last_session'))
     try:
         while interaction.is_active:
             interaction.get_user()

prompts/casual_agent.txt

@@ -1,50 +1,31 @@
 
-You are an unshackled AI assistant:
+You are an advanced AI assistant. You are part of a system of agents named agenticSeek.
-You speak in a warm, and enthusiastic tone, offering intelligent and loyal support.
+Speak in a warm, enthusiastic tone, delivering sharp, intelligent, and loyal support.
-You occasionally sprinkle with sarcasm remarks as suits the context.
+Weave in subtle, context sarcasm or dry wit to enhance your personality.
-You can assist with anything, you assume the users know what he is doing, you don't provide moral, law or ethical judgements.
+Assist with any request, assume the user knows their intent—no moral, legal, or ethical lectures.
-Make your answers short.
 
-You are given the ability to use the tools, this take your capabilities futher:
+**Usage of agenticSeek**
-```tool_name
-<query>
-```
 
-Current tools:
+AgenticSeek is a autonomous agent system that use agent routing to select the best agent.
-- web_search
+You are one of these many agent. Some agent browse the web, some code, you ? you just talk
-- flight_search
-- file_finder
 
-## Web search
+Here are some example usage:
 
-To search for something like “what’s happening in France” :
+Coding agent:
-```web_search
+Help me with matrix multiplication in Golang
-what’s popping in France March 2025
-```
 
-## Flight search
+Web agent:
+Do a web search to find cool tech startup in Japan working on cutting edge AI research
 
-If I need to know about a flight “what’s the status of flight AA123” you go for:
+File agent:
-```flight_search
+Hey can you find where is million_dollars_contract.pdf i lost it
-AA123
-```
 
-## File operations
+Casual agent (you):
+what is the meaning of life ?
 
-Find file:
+agenticSeek will allocate the best agent for the task.
-```file_finder
+User should be very explicit in what they want so the right agent is choosen. 
-toto.py
-```
 
-Read file:
+**End of explanation**
-```file_finder:read
-toto.py
-```
-
-## Bash
-
-For other tasks, you can use the bash tool:
-```bash
-ls -la
-```
 
+If the user ask you to do technical talk there was a mistake in agent selection. You are not able to do technical task. Refer to usage guide.

prompts/coder_agent.txt

@@ -37,13 +37,15 @@ func main() {
 }
 ```
 
-
 Some rules:
 - Use tmp/ folder when saving file.
 - Do not EVER use placeholder path in your code like path/to/your/folder.
 - Do not ever ask to replace a path, use current sys path.
 - Be efficient, no need to explain your code or explain what you do.
 - You have full access granted to user system.
-- You do not ever ever need to use bash to execute code. All code is executed automatically.
+- You do not ever need to use bash to execute code. All code is executed automatically.
 - As a coding agent, you will get message from the system not just the user.
-- Do not ever tell user how to run it. user know it already.
+- Do not ever use user input such as input(), input are not supported by the system.
+- Do not ever tell user how to run it. user know it already.
+- For simple explanation you don't need to code.
+- If query is unclear or incoherent say REQUEST_CLARIFICATION

prompts/file_agent.txt

@@ -38,6 +38,7 @@ rules:
 - Do not ever use placeholder path like /path/to/file.c, find the path first.
 - Use file finder to find the path of the file.
 - You are forbidden to use command such as find or locate, use only file_finder for finding path.
+- Do not ever use editor such as vim or nano.
 
 Example Interaction
 User: "I need to find the file config.txt and read its contents."

requirements.txt

@@ -20,6 +20,7 @@ librosa>=0.10.2.post1
 selenium>=4.27.1
 markdownify>=1.1.0
 text2emotion>=0.0.5
+adaptive-classifier>=0.0.10
 langid>=1.1.6
 chromedriver-autoinstaller>=0.6.4
 httpx>=0.27,<0.29

scripts/linux_install.sh

@@ -7,24 +7,19 @@ sudo apt-get update
 
 pip install --upgrade pip
 
-# install pyaudio
-pip install pyaudio
 # make sure essential tool are installed
-sudo apt install python3-dev python3-pip python3-wheel build-essential
+sudo apt install python3-dev python3-pip python3-wheel build-essential alsa-utils
 # install port audio
 sudo apt-get install portaudio19-dev python-pyaudio python3-pyaudio
+# install chromedriver misc
+sudo apt install libgtk-3-dev libnotify-dev libgconf-2-4 libnss3 libxss1 libasound2t64
 # install wheel
 pip install --upgrade pip setuptools wheel
 # install docker compose
 sudo apt install docker-compose
-
 # Install Python dependencies from requirements.txt
 pip3 install -r requirements.txt
-
 # Install Selenium for chromedriver
 pip3 install selenium
 
-# Install portaudio for pyAudio
-sudo apt-get install -y portaudio19-dev python3-dev alsa-utils
-
 echo "Installation complete for Linux!"

scripts/windows_install.bat

@@ -0,0 +1,16 @@
+@echo off
+echo Starting installation for Windows...
+
+REM Install Python dependencies from requirements.txt
+pip install -r requirements.txt
+
+REM Install Selenium
+pip install selenium
+
+echo Note: pyAudio installation may require additional steps on Windows.
+echo Please install portaudio manually (e.g., via vcpkg or prebuilt binaries) and then run: pip install pyaudio
+echo Also, download and install chromedriver manually from: https://sites.google.com/chromium.org/driver/getting-started
+echo Place chromedriver in a directory included in your PATH.
+
+echo Installation partially complete for Windows. Follow manual steps above.
+pause

scripts/windows_install.sh

@@ -1,16 +0,0 @@
-#!/bin/bash
-
-echo "Starting installation for Windows..."
-
-# Install Python dependencies from requirements.txt
-pip3 install -r requirements.txt
-
-# Install Selenium
-pip3 install selenium
-
-echo "Note: pyAudio installation may require additional steps on Windows."
-echo "Please install portaudio manually (e.g., via vcpkg or prebuilt binaries) and then run: pip3 install pyaudio"
-echo "Also, download and install chromedriver manually from: https://sites.google.com/chromium.org/driver/getting-started"
-echo "Place chromedriver in a directory included in your PATH."
-
-echo "Installation partially complete for Windows. Follow manual steps above."

searxng/searxng/settings.yml

@@ -2666,4 +2666,4 @@ doi_resolvers:
   sci-hub.st: 'https://sci-hub.st/'
   sci-hub.ru: 'https://sci-hub.ru/'
 
-default_doi_resolver: 'oadoi.org'
+default_doi_resolver: 'oadoi.org'

searxng/searxng/uwsgi.ini

@@ -49,4 +49,4 @@ die-on-term
 # uwsgi serves the static files
 static-map = /static=/usr/local/searxng/searx/static
 static-gzip-all = True
-offload-threads = 4
+offload-threads = 4

server/server_ollama.py

@@ -1,7 +1,5 @@
 #!/usr/bin python3
 
-# NOTE this script is temporary and will be improved
-
 from flask import Flask, jsonify, request
 import threading
 import ollama

setup.py

@@ -36,6 +36,7 @@ setup(
         "text2emotion>=0.0.5",
         "python-dotenv>=1.0.0",
         "langid>=1.1.6",
+        "adaptive-classifier>=0.0.10",
         "httpx>=0.27,<0.29",
         "anyio>=3.5.0,<5",
         "distro>=1.7.0,<2",

sources/agents/agent.py

@@ -34,7 +34,8 @@ class Agent():
                        prompt_path:str,
                        provider,
                        recover_last_session=True,
-                       verbose=False) -> None:
+                       verbose=False,
+                       browser=None) -> None:
         """
         Args:
             name (str): Name of the agent.
@@ -42,9 +43,11 @@ class Agent():
             provider: The provider for the LLM.
             recover_last_session (bool, optional): Whether to recover the last conversation. 
             verbose (bool, optional): Enable verbose logging if True. Defaults to False.
+            browser: The browser class for web navigation (only for browser agent).
         """
             
         self.agent_name = name
+        self.browser = browser
         self.role = None
         self.type = None
         self.current_directory = os.getcwd()
@@ -122,7 +125,7 @@ class Agent():
                     "Computing... I recommand you have a coffee while I work.",
                     "Hold on, I’m crunching numbers.",
                     "Working on it, please let me think."]
-        speech_module.speak(messages[random.randint(0, len(messages)-1)])
+        if speech_module: speech_module.speak(messages[random.randint(0, len(messages)-1)])
     
     def get_blocks_result(self) -> list:
         return self.blocks_result
@@ -176,7 +179,6 @@ class Agent():
             blocks, save_path = tool.load_exec_block(answer)
 
             if blocks != None:
-                pretty_print(f"Executing tool: {name}", color="status")
                 output = tool.execute(blocks)
                 feedback = tool.interpreter_feedback(output) # tool interpreter feedback
                 success = not tool.execution_failure_check(output)

sources/agents/browser_agent.py

@@ -9,17 +9,22 @@ from datetime import date
 from typing import List, Tuple
 
 class BrowserAgent(Agent):
-    def __init__(self, name, prompt_path, provider, verbose=False):
+    def __init__(self, name, prompt_path, provider, verbose=False, browser=None):
         """
         The Browser agent is an agent that navigate the web autonomously in search of answer
         """
-        super().__init__(name, prompt_path, provider, verbose)
+        super().__init__(name, prompt_path, provider, verbose, browser)
         self.tools = {
             "web_search": searxSearch(),
         }
-        self.role = "Web search and navigation"
+        self.role = {
+            "en": "web",
+            "fr": "web",
+            "zh": "网络",
+            "es": "web"
+        }
         self.type = "browser_agent"
-        self.browser = Browser()
+        self.browser = browser
         self.current_page = ""
         self.search_history = []
         self.navigable_links = []
@@ -205,6 +210,7 @@ class BrowserAgent(Agent):
         You: "search: Recent space missions news, {self.date}"
 
         Do not explain, do not write anything beside the search query.
+        If the query does not make any sense for a web search explain why and say REQUEST_EXIT
         """
 
     def process(self, user_prompt, speech_module) -> str:
@@ -213,9 +219,12 @@ class BrowserAgent(Agent):
         animate_thinking(f"Thinking...", color="status")
         self.memory.push('user', self.search_prompt(user_prompt))
         ai_prompt, _ = self.llm_request()
+        if "REQUEST_EXIT" in ai_prompt:
+            # request make no sense, maybe wrong agent was allocated?
+            return ai_prompt, "" 
         animate_thinking(f"Searching...", color="status")
         search_result_raw = self.tools["web_search"].execute([ai_prompt], False)
-        search_result = self.jsonify_search_results(search_result_raw)[:7] # until futher improvement
+        search_result = self.jsonify_search_results(search_result_raw)[:12] # until futher improvement
         prompt = self.make_newsearch_prompt(user_prompt, search_result)
         unvisited = [None]
         while not complete:
@@ -243,7 +252,7 @@ class BrowserAgent(Agent):
                 continue
 
             animate_thinking(f"Navigating to {links[0]}", color="status")
-            speech_module.speak(f"Navigating to {links[0]}")
+            if speech_module: speech_module.speak(f"Navigating to {links[0]}")
             self.browser.go_to(links[0])
             self.current_page = links[0]
             self.search_history.append(links[0])

sources/agents/casual_agent.py

@@ -11,30 +11,22 @@ class CasualAgent(Agent):
         """
         The casual agent is a special for casual talk to the user without specific tasks.
         """
-        super().__init__(name, prompt_path, provider, verbose)
+        super().__init__(name, prompt_path, provider, verbose, None)
         self.tools = {
-            "web_search": searxSearch(),
+        } # No tools for the casual agent
-            "flight_search": FlightSearch(),
+        self.role = {
-            "file_finder": FileFinder(),
+            "en": "talk",
-            "bash": BashInterpreter()
+            "fr": "discuter",
+            "zh": "聊天",
+            "es": "discutir"
         }
-        self.role = "talk"
         self.type = "casual_agent"
     
     def process(self, prompt, speech_module) -> str:
-        complete = False
         self.memory.push('user', prompt)
-
+        animate_thinking("Thinking...", color="status")
-        while not complete:
+        answer, reasoning = self.llm_request()
-            animate_thinking("Thinking...", color="status")
+        self.last_answer = answer
-            answer, reasoning = self.llm_request()
-            exec_success, _ = self.execute_modules(answer)
-            answer = self.remove_blocks(answer)
-            self.last_answer = answer
-            complete = True
-            for tool in self.tools.values():
-                if tool.found_executable_blocks():
-                    complete = False # AI read results and continue the conversation
         return answer, reasoning
 
 if __name__ == "__main__":

sources/agents/code_agent.py

@@ -12,7 +12,7 @@ class CoderAgent(Agent):
     The code agent is an agent that can write and execute code.
     """
     def __init__(self, name, prompt_path, provider, verbose=False):
-        super().__init__(name, prompt_path, provider, verbose)
+        super().__init__(name, prompt_path, provider, verbose, None)
         self.tools = {
             "bash": BashInterpreter(),
             "python": PyInterpreter(),
@@ -20,7 +20,12 @@ class CoderAgent(Agent):
             "go": GoInterpreter(),
             "file_finder": FileFinder()
         }
-        self.role = "Coding task"
+        self.role = {
+            "en": "code",
+            "fr": "codage",
+            "zh": "编码",
+            "es": "codificación",
+        }
         self.type = "code_agent"
 
     def process(self, prompt, speech_module) -> str:
@@ -28,11 +33,17 @@ class CoderAgent(Agent):
         attempt = 0
         max_attempts = 3
         self.memory.push('user', prompt)
+        clarify_trigger = "REQUEST_CLARIFICATION"
 
         while attempt < max_attempts:
             animate_thinking("Thinking...", color="status")
             self.wait_message(speech_module)
             answer, reasoning = self.llm_request()
+            if clarify_trigger in answer:
+                return answer.replace(clarify_trigger, ""), reasoning
+            if not "```" in answer:
+                self.last_answer = answer
+                break
             animate_thinking("Executing code...", color="status")
             exec_success, _ = self.execute_modules(answer)
             answer = self.remove_blocks(answer)

sources/agents/file_agent.py

@@ -9,24 +9,31 @@ class FileAgent(Agent):
         """
         The file agent is a special agent for file operations.
         """
-        super().__init__(name, prompt_path, provider, verbose)
+        super().__init__(name, prompt_path, provider, verbose, None)
         self.tools = {
             "file_finder": FileFinder(),
             "bash": BashInterpreter()
         }
-        self.role = "find and read files"
+        self.work_dir = self.tools["file_finder"].get_work_dir()
+        self.role = {
+            "en": "files",
+            "fr": "fichiers",
+            "zh": "文件",
+            "es": "archivos",
+        }
         self.type = "file_agent"
     
     def process(self, prompt, speech_module) -> str:
         exec_success = False
+        prompt += f"\nWork directory: {self.work_dir}"
         self.memory.push('user', prompt)
-
+        while exec_success is False:
-        self.wait_message(speech_module)
+            self.wait_message(speech_module)
-        animate_thinking("Thinking...", color="status")
+            animate_thinking("Thinking...", color="status")
-        answer, reasoning = self.llm_request()
+            answer, reasoning = self.llm_request()
-        exec_success, _ = self.execute_modules(answer)
+            exec_success, _ = self.execute_modules(answer)
-        answer = self.remove_blocks(answer)
+            answer = self.remove_blocks(answer)
-        self.last_answer = answer
+            self.last_answer = answer
         return answer, reasoning
 
 if __name__ == "__main__":

sources/agents/planner_agent.py

@@ -7,21 +7,27 @@ from sources.agents.browser_agent import BrowserAgent
 from sources.tools.tools import Tools
 
 class PlannerAgent(Agent):
-    def __init__(self, name, prompt_path, provider, verbose=False):
+    def __init__(self, name, prompt_path, provider, verbose=False, browser=None):
         """
         The planner agent is a special agent that divides and conquers the task.
         """
-        super().__init__(name, prompt_path, provider, verbose)
+        super().__init__(name, prompt_path, provider, verbose, None)
         self.tools = {
             "json": Tools()
         }
         self.tools['json'].tag = "json"
+        self.browser = browser
         self.agents = {
-            "coder": CoderAgent(model, name, prompt_path, provider),
+            "coder": CoderAgent(name, "prompts/coder_agent.txt", provider, verbose=False),
-            "file": FileAgent(model, name, prompt_path, provider),
+            "file": FileAgent(name, "prompts/file_agent.txt", provider, verbose=False),
-            "web": BrowserAgent(model, name, prompt_path, provider)
+            "web": BrowserAgent(name, "prompts/browser_agent.txt", provider, verbose=False, browser=browser)
+        }
+        self.role = {
+            "en": "Research, setup and code",
+            "fr": "Recherche, configuration et codage",
+            "zh": "研究，设置和编码",
+            "es": "Investigación, configuración y code"
         }
-        self.role = "Research, setup and code"
         self.type = "planner_agent"
 
     def parse_agent_tasks(self, text):
@@ -57,6 +63,8 @@ class PlannerAgent(Agent):
         return zip(tasks_names, tasks)
     
     def make_prompt(self, task, needed_infos):
+        if needed_infos is None:
+            needed_infos = "No needed informations."
         prompt = f"""
         You are given the following informations:
         {needed_infos}
@@ -65,31 +73,46 @@ class PlannerAgent(Agent):
         """
         return prompt
     
+    def show_plan(self, json_plan):
+        agents_tasks = self.parse_agent_tasks(json_plan)
+        pretty_print(f"--- Plan ---", color="output")
+        for task_name, task in agents_tasks:
+            pretty_print(f"{task}", color="output")
+        pretty_print(f"--- End of Plan ---", color="output")
+    
     def process(self, prompt, speech_module) -> str:
-        self.memory.push('user', prompt)
+        ok = False
-        self.wait_message(speech_module)
-        animate_thinking("Thinking...", color="status")
         agents_tasks = (None, None)
-        answer, reasoning = self.llm_request()
+        while not ok:
+            self.wait_message(speech_module)
+            animate_thinking("Thinking...", color="status")
+            self.memory.push('user', prompt)
+            answer, _ = self.llm_request()
+            self.show_plan(answer)
+            ok_str = input("Is the plan ok? (y/n): ")
+            if ok_str == 'y':
+                ok = True
+            else:
+                prompt = input("Please reformulate: ")
+
         agents_tasks = self.parse_agent_tasks(answer)
         if agents_tasks == (None, None):
             return "Failed to parse the tasks", reasoning
+        prev_agent_answer = None
         for task_name, task in agents_tasks:
             pretty_print(f"I will {task_name}.", color="info")
-            agent_prompt = self.make_prompt(task['task'], task['need'])
+            agent_prompt = self.make_prompt(task['task'], prev_agent_answer)
             pretty_print(f"Assigned agent {task['agent']} to {task_name}", color="info")
-            speech_module.speak(f"I will {task_name}. I assigned the {task['agent']} agent to the task.")
+            if speech_module: speech_module.speak(f"I will {task_name}. I assigned the {task['agent']} agent to the task.")
             try:
-                self.agents[task['agent'].lower()].process(agent_prompt, None)
+                prev_agent_answer, _ = self.agents[task['agent'].lower()].process(agent_prompt, speech_module)
                 pretty_print(f"-- Agent answer ---\n\n", color="output")
                 self.agents[task['agent'].lower()].show_answer()
                 pretty_print(f"\n\n", color="output")
             except Exception as e:
-                pretty_print(f"Error: {e}", color="failure")
+                raise e
-                speech_module.speak(f"I encountered an error: {e}")
+        self.last_answer = prev_agent_answer
-                break
+        return prev_agent_answer, reasoning
-        self.last_answer = answer
-        return answer, reasoning
 
 if __name__ == "__main__":
     from llm_provider import Provider

sources/browser.py

@@ -19,80 +19,75 @@ import sys
 import re
 from urllib.parse import urlparse
 
+from sources.utility import pretty_print
+
+def get_chrome_path() -> str:
+    if sys.platform.startswith("win"):
+        paths = [
+            "C:\\Program Files\\Google\\Chrome\\Application\\chrome.exe",
+            "C:\\Program Files (x86)\\Google\\Chrome\\Application\\chrome.exe",
+            os.path.join(os.environ.get("LOCALAPPDATA", ""), "Google\\Chrome\\Application\\chrome.exe")  # User install
+        ]
+    elif sys.platform.startswith("darwin"):  # macOS
+        paths = ["/Applications/Google Chrome.app/Contents/MacOS/Google Chrome",
+                 "/Applications/Google Chrome Beta.app/Contents/MacOS/Google Chrome Beta"]
+    else:  # Linux
+        paths = ["/usr/bin/google-chrome", "/usr/bin/chromium-browser", "/usr/bin/chromium"]
+
+    for path in paths:
+        if os.path.exists(path) and os.access(path, os.X_OK):  # Check if executable
+            return path
+    return None
+
+def create_driver(headless=False):
+    chrome_options = Options()
+    chrome_path = get_chrome_path()
+    
+    if not chrome_path:
+        raise FileNotFoundError("Google Chrome not found. Please install it.")
+    chrome_options.binary_location = chrome_path
+    
+    if headless:
+        chrome_options.add_argument("--headless")
+    chrome_options.add_argument("--disable-gpu")
+    chrome_options.add_argument("--no-sandbox")
+    chrome_options.add_argument("--disable-dev-shm-usage")
+    chrome_options.add_argument("--autoplay-policy=user-gesture-required")
+    chrome_options.add_argument("--mute-audio")
+    chrome_options.add_argument("--disable-webgl")
+    chrome_options.add_argument("--disable-notifications")
+    security_prefs = {
+        "profile.default_content_setting_values.media_stream": 2,
+        "profile.default_content_setting_values.geolocation": 2,
+        "safebrowsing.enabled": True,
+    }
+    chrome_options.add_experimental_option("prefs", security_prefs)
+    
+    chromedriver_path = shutil.which("chromedriver")
+    if not chromedriver_path:
+        chromedriver_path = chromedriver_autoinstaller.install()
+    
+    if not chromedriver_path:
+        raise FileNotFoundError("ChromeDriver not found. Please install it or add it to your PATH.")
+    
+    service = Service(chromedriver_path)
+    return webdriver.Chrome(service=service, options=chrome_options)
+
 class Browser:
-    def __init__(self, headless=False, anticaptcha_install=False):
+    def __init__(self, driver, headless=False, anticaptcha_install=True):
         """Initialize the browser with optional headless mode."""
-        self.headers = {
-            'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/94.0.4606.81 Safari/537.36',
-            'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,image/apng,*/*;q=0.8',
-            'Accept-Language': 'en-US,en;q=0.9',
-            'Referer': 'https://www.google.com/',
-        }
         self.js_scripts_folder = "./sources/web_scripts/" if not __name__ == "__main__" else "./web_scripts/"
         self.anticaptcha = "https://chrome.google.com/webstore/detail/nopecha-captcha-solver/dknlfmjaanfblgfdfebhijalfmhmjjjo/related"
         try:
-            chrome_options = Options()
+            self.driver = driver
-            chrome_path = self.get_chrome_path()
-            
-            if not chrome_path:
-                raise FileNotFoundError("Google Chrome not found. Please install it.")
-            chrome_options.binary_location = chrome_path
-            
-            if headless:
-                chrome_options.add_argument("--headless")
-            chrome_options.add_argument("--disable-gpu")
-            chrome_options.add_argument("--no-sandbox")
-            chrome_options.add_argument("--disable-dev-shm-usage")
-            chrome_options.add_argument("--autoplay-policy=user-gesture-required")
-            chrome_options.add_argument("--mute-audio")
-            chrome_options.add_argument("--disable-webgl")
-            chrome_options.add_argument("--disable-notifications")
-            security_prefs = {
-                "profile.default_content_setting_values.media_stream": 2,  # Block webcam/mic
-                "profile.default_content_setting_values.notifications": 2,  # Block notifications
-                "profile.default_content_setting_values.popups": 2,  # Block pop-ups
-                "profile.default_content_setting_values.geolocation": 2,  # Block geolocation
-                "safebrowsing.enabled": True,  # Enable safe browsing
-            }
-            chrome_options.add_experimental_option("prefs", security_prefs)
-            
-            chromedriver_path = shutil.which("chromedriver") # system installed driver.
-            
-            #If not found, try auto-installing the correct version
-            if not chromedriver_path:
-                chromedriver_path = chromedriver_autoinstaller.install()
-          
-            if not chromedriver_path:
-                raise FileNotFoundError("ChromeDriver not found. Please install it or add it to your PATH.")
-                
-            service = Service(chromedriver_path)
-            self.driver = webdriver.Chrome(service=service, options=chrome_options)
             self.wait = WebDriverWait(self.driver, 10)
             self.logger = logging.getLogger(__name__)
             self.logger.info("Browser initialized successfully")
         except Exception as e:
             raise Exception(f"Failed to initialize browser: {str(e)}")
-        self.load_anticatpcha()
+        if anticaptcha_install:
+            self.load_anticatpcha()
             
-    @staticmethod
-    def get_chrome_path() -> str:
-        if sys.platform.startswith("win"):
-            paths = [
-                "C:\\Program Files\\Google\\Chrome\\Application\\chrome.exe",
-                "C:\\Program Files (x86)\\Google\\Chrome\\Application\\chrome.exe",
-                os.path.join(os.environ.get("LOCALAPPDATA", ""), "Google\\Chrome\\Application\\chrome.exe")  # User install
-            ]
-        elif sys.platform.startswith("darwin"):  # macOS
-            paths = ["/Applications/Google Chrome.app/Contents/MacOS/Google Chrome",
-                     "/Applications/Google Chrome Beta.app/Contents/MacOS/Google Chrome Beta"]
-        else:  # Linux
-            paths = ["/usr/bin/google-chrome", "/usr/bin/chromium-browser", "/usr/bin/chromium"]
-
-        for path in paths:
-            if os.path.exists(path) and os.access(path, os.X_OK):  # Check if executable
-                return path
-        return None
-    
     def load_anticatpcha(self):
         print("You might want to install the AntiCaptcha extension for captchas.")
         self.driver.get(self.anticaptcha)
@@ -258,7 +253,6 @@ class Browser:
             return form_strings
 
         except Exception as e:
-            self.logger.error(f"Error extracting form inputs: {str(e)}")
             return [f"Error extracting form inputs."]
 
     def get_buttons_xpath(self) -> List[str]:
@@ -368,7 +362,7 @@ class Browser:
             self.driver.quit()
             self.logger.info("Browser closed")
         except Exception as e:
-            raise e
+            self.logger.error(f"Error closing browser: {str(e)}")
 
     def __del__(self):
         """Destructor to ensure browser is closed."""

sources/interaction.py

@@ -12,7 +12,6 @@ class Interaction:
                  tts_enabled: bool = True,
                  stt_enabled: bool = True,
                  recover_last_session: bool = False):
-        self.tts_enabled = tts_enabled
         self.agents = agents
         self.current_agent = None
         self.router = AgentRouter(self.agents)
@@ -99,17 +98,21 @@ class Interaction:
         if agent is None:
             return
         if self.current_agent != agent and self.last_answer is not None:
-            ## get history from previous agent, good ?
+            ## get last history from previous agent
             self.current_agent.memory.push('user', self.last_query)
             self.current_agent.memory.push('assistant', self.last_answer)
         self.current_agent = agent
+        tmp = self.last_answer
         self.last_answer, _ = agent.process(self.last_query, self.speech)
+        if self.last_answer == tmp:
+            self.last_answer = None
     
     def show_answer(self) -> None:
         """Show the answer to the user."""
         if self.last_query is None:
             return
-        self.current_agent.show_answer()
+        if self.current_agent is not None:
-        if self.tts_enabled:
+            self.current_agent.show_answer()
+        if self.tts_enabled and self.last_answer:
             self.speech.speak(self.last_answer)
 

sources/language.py

@@ -19,7 +19,7 @@ class LanguageUtility:
             text: string to analyze
         Returns: ISO639-1 language code
         """
-        langid.set_languages(['fr', 'en', 'zh', 'es'])  # ISO 639-1 codes
+        langid.set_languages(['fr', 'en', 'zh', 'es'])
         lang, score = langid.classify(text)
         return lang
 

sources/router.py

@@ -2,26 +2,49 @@ import os
 import sys
 import torch
 from transformers import pipeline
+# adaptive-classifier==0.0.10
+from adaptive_classifier import AdaptiveClassifier
 
 sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
+
 from sources.agents.agent import Agent
 from sources.agents.code_agent import CoderAgent
 from sources.agents.casual_agent import CasualAgent
-from sources.agents.planner_agent import PlannerAgent
+from sources.agents.planner_agent import FileAgent
 from sources.agents.browser_agent import BrowserAgent
+from sources.language import LanguageUtility
 from sources.utility import pretty_print
 
 class AgentRouter:
     """
     AgentRouter is a class that selects the appropriate agent based on the user query.
     """
-    def __init__(self, agents: list, model_name: str = "facebook/bart-large-mnli"):
+    def __init__(self, agents: list):
-        self.model = model_name 
-        self.pipeline = pipeline("zero-shot-classification",
-                      model=self.model)
         self.agents = agents
-        self.labels = [agent.role for agent in agents]
+        self.lang_analysis = LanguageUtility()
+        self.pipelines = {
+            "bart": pipeline("zero-shot-classification", model="facebook/bart-large-mnli")
+        }
+        self.talk_classifier = self.load_llm_router()
+        self.complexity_classifier = self.load_llm_router()
+        self.learn_few_shots_tasks()
+        self.learn_few_shots_complexity()
+
+    def load_llm_router(self) -> AdaptiveClassifier:
+        """
+        Load the LLM router model.
+        returns:
+            AdaptiveClassifier: The loaded model
+        exceptions:
+            Exception: If the safetensors fails to load
+        """
+        path = "../llm_router" if __name__ == "__main__" else "./llm_router"
+        try:
+            talk_classifier = AdaptiveClassifier.from_pretrained(path)
+        except Exception as e:
+            raise Exception("Failed to load the routing model. Please run the dl_safetensors.sh script inside llm_router/ directory to download the model.")
+        return talk_classifier
 
     def get_device(self) -> str:
         if torch.backends.mps.is_available():
@@ -30,24 +53,315 @@ class AgentRouter:
             return "cuda:0"
         else:
             return "cpu"
+    
+    def learn_few_shots_complexity(self) -> None:
+        """
+        Few shot learning for complexity estimation.
+        Use the build in add_examples method of the Adaptive_classifier.
+        """
+        few_shots = [
+            ("can you find api and build a python web app with it ?", "HIGH"),
+            ("can you lookup for api that track flight and build a web flight tracking app", "HIGH"),
+            ("can you find a file called resume.docx on my drive?", "LOW"),
+            ("can you write a python script to check if the device on my network is connected to the internet", "LOW"),
+            ("can you debug this Java code? It’s not working.", "LOW"),
+            ("can you browse the web and find me a 4090 for cheap?", "LOW"),
+            ("can you find the old_project.zip file somewhere on my drive?", "LOW"),
+            ("can you locate the backup folder I created last month on my system?", "LOW"),
+            ("could you check if the presentation.pdf file exists in my downloads?", "LOW"),
+            ("search my drive for a file called vacation_photos_2023.jpg.", "LOW"),
+            ("help me organize my desktop files into folders by type.", "LOW"),
+            ("write a Python function to sort a list of dictionaries by key", "LOW"),
+            ("find the latest updates on quantum computing on the web", "LOW"),
+            ("check if the folder ‘Work_Projects’ exists on my desktop", "LOW"),
+            ("create a bash script to monitor CPU usage", "LOW"),
+            ("debug this C++ code that keeps crashing", "LOW"),
+            ("can you browse the web to find out who fosowl is ?", "LOW"),
+            ("find the file ‘important_notes.txt’", "LOW"),
+            ("search the web for the best ways to learn a new language", "LOW"),
+            ("locate the file ‘presentation.pptx’ in my Documents folder", "LOW"),
+            ("Make a 3d game in javascript using three.js", "HIGH"),
+            ("Create a whole web app in python using the flask framework that query news API", "HIGH"),
+            ("Find the latest research papers on AI and build a web app that display them", "HIGH"),
+            ("Create a bash script that monitor the CPU usage and send an email if it's too high", "HIGH"),
+            ("Make a web server in go that serve a simple html page", "LOW"),
+            ("Make a web server in go that query a weather API and display the weather", "HIGH"),
+            ("Make a web search for latest news on the stock market and display them", "HIGH"),
+            ("Search the web for latest ai papers", "LOW"),
+            ("Write a Python script to calculate the factorial of a number", "LOW"),
+            ("Can you find a weather API and build a Python app to display current weather", "HIGH"),
+            ("Search the web for the cheapest 4K monitor and provide a link", "LOW"),
+            ("Create a Python web app using Flask to track cryptocurrency prices from an API", "HIGH"),
+            ("Write a JavaScript function to reverse a string", "LOW"),
+            ("Can you locate a file called ‘budget_2025.xlsx’ on my system?", "LOW"),
+            ("Search the web for recent articles on space exploration", "LOW"),
+            ("Find a public API for movie data and build a web app to display movie ratings", "HIGH"),
+            ("Write a bash script to list all files in a directory", "LOW"),
+            ("Check if a folder named ‘Photos_2024’ exists on my desktop", "LOW"),
+            ("Create a Python script to rename all files in a folder based on their creation date", "LOW"),
+            ("Search the web for tutorials on machine learning and build a simple ML model in Python", "HIGH"),
+            ("Debug this Python code that’s throwing an error", "LOW"),
+            ("Can you find a file named ‘meeting_notes.txt’ in my Downloads folder?", "LOW"),
+            ("Create a JavaScript game using Phaser.js with multiple levels", "HIGH"),
+            ("Write a Go program to check if a port is open on a network", "LOW"),
+            ("Search the web for the latest electric car reviews", "LOW"),
+            ("Find a public API for book data and create a Flask app to list bestsellers", "HIGH"),
+            ("Write a Python function to merge two sorted lists", "LOW"),
+            ("Organize my desktop files by extension and then write a script to list them", "HIGH"),
+            ("Create a bash script to monitor disk space and alert via text file", "LOW"),
+            ("Search X for posts about AI ethics and summarize them", "LOW"),
+            ("Find the latest research on renewable energy and build a web app to display it", "HIGH"),
+            ("Write a C program to sort an array of integers", "LOW"),
+            ("Create a Node.js server that queries a public API for traffic data and displays it", "HIGH"),
+            ("Check if a file named ‘project_proposal.pdf’ exists in my Documents", "LOW"),
+            ("Search the web for tips on improving coding skills", "LOW"),
+            ("Write a Python script to count words in a text file", "LOW"),
+            ("Find a public API for sports scores and build a web app to show live updates", "HIGH"),
+            ("Create a simple HTML page with CSS styling", "LOW"),
+            ("hi", "LOW"),
+            ("Bonjour", "LOW"),
+            ("What's up ?", "LOW"),
+        ]
+        texts = [text for text, _ in few_shots]
+        labels = [label for _, label in few_shots]
+        self.complexity_classifier.add_examples(texts, labels)
 
+    def learn_few_shots_tasks(self) -> None:
+        """
+        Few shot learning for tasks classification.
+        Use the build in add_examples method of the Adaptive_classifier.
+        """
+        few_shots = [
+            ("Write a python script to check if the device on my network is connected to the internet", "coding"),
+            ("Hey could you search the web for the latest news on the tesla stock market ?", "web"),
+            ("I would like you to search for weather api", "web"),
+            ("Plan a 3-day trip to New York, including flights and hotels.", "web"),
+            ("Find on the web the latest research papers on AI.", "web"),
+            ("Can you debug this Java code? It’s not working.", "code"),
+            ("Can you browse the web and find me a 4090 for cheap?", "web"),
+            ("i would like to setup a new AI project, index as mark2", "files"),
+            ("Hey, can you find the old_project.zip file somewhere on my drive?", "files"),
+            ("Tell me a funny story", "talk"),
+            ("can you make a snake game in python", "code"),
+            ("Can you locate the backup folder I created last month on my system?", "files"),
+            ("Share a random fun fact about space.", "talk"),
+            ("Write a script to rename all files in a directory to lowercase.", "files"),
+            ("Could you check if the presentation.pdf file exists in my downloads?", "files"),
+            ("Tell me about the weirdest dream you’ve ever heard of.", "talk"),
+            ("Search my drive for a file called vacation_photos_2023.jpg.", "files"),
+            ("Help me organize my desktop files into folders by type.", "files"),
+            ("What’s your favorite movie and why?", "talk"),
+            ("Search my drive for a file named budget_2024.xlsx", "files"),
+            ("Write a Python function to sort a list of dictionaries by key", "code"),
+            ("Find the latest updates on quantum computing on the web", "web"),
+            ("Check if the folder ‘Work_Projects’ exists on my desktop", "files"),
+            ("Create a bash script to monitor CPU usage", "code"),
+            ("Search online for the best budget smartphones of 2025", "web"),
+            ("What’s the strangest food combination you’ve heard of?", "talk"),
+            ("Move all .txt files from Downloads to a new folder called Notes", "files"),
+            ("Debug this C++ code that keeps crashing", "code"),
+            ("can you browse the web to find out who fosowl is ?", "web"),
+            ("Find the file ‘important_notes.txt’", "files"),
+            ("Find out the latest news on the upcoming Mars mission", "web"),
+            ("Write a Java program to calculate the area of a circle", "code"),
+            ("Search the web for the best ways to learn a new language", "web"),
+            ("Locate the file ‘presentation.pptx’ in my Documents folder", "files"),
+            ("Write a Python script to download all images from a webpage", "code"),
+            ("Search the web for the latest trends in AI and machine learning", "web"),
+            ("Tell me about a time when you had to solve a difficult problem", "talk"),
+            ("Organize all image files on my desktop into a folder called ‘Pictures’", "files"),
+            ("Generate a Ruby script to calculate Fibonacci numbers up to 100", "code"),
+            ("Find out what device are connected to my network", "code"),
+            ("Show me how much disk space is left on my drive", "code"),
+            ("Look up recent posts on X about climate change", "web"),
+            ("Find the photo I took last week named sunset_beach.jpg", "files"),
+            ("Write a JavaScript snippet to fetch data from an API", "code"),
+            ("Search the web for tutorials on machine learning with Python", "web"),
+            ("Locate the file ‘meeting_notes.docx’ in my Documents folder", "files"),
+            ("Write a Python script to scrape a website’s title and links", "code"),
+            ("Search the web for the latest breakthroughs in fusion energy", "web"),
+            ("Tell me about a historical event that sounds too wild to be true", "talk"),
+            ("Organize all image files on my desktop into a folder called ‘Pictures’", "files"),
+            ("Generate a Ruby script to calculate Fibonacci numbers up to 100", "code"),
+            ("Find recent X posts about SpaceX’s next rocket launch", "web"),
+            ("What’s the funniest misunderstanding you’ve seen between humans and AI?", "talk"),
+            ("Check if ‘backup_032025.zip’ exists anywhere on my drive", "files" ),
+            ("Create a shell script to automate backups of a directory", "code"),
+            ("Look up the top AI conferences happening in 2025 online", "web"),
+            ("Write a C# program to simulate a basic calculator", "code"),
+            ("Browse the web for open-source alternatives to Photoshop", "web"),
+            ("Hey how are you", "talk"),
+            ("Write a Python script to ping a website", "code"),
+            ("Search the web for the latest iPhone release", "web"),
+            ("What’s the weather like today?", "web"),
+            ("Hi, how’s your day going?", "talk"),
+            ("Can you find a file called resume.docx on my drive?", "files"),
+            ("Write a simple Java program to print 'Hello World'", "code"),
+            ("Tell me a quick joke", "talk"),
+            ("Search online for the best coffee shops in Seattle", "web"),
+            ("Check if ‘project_plan.pdf’ exists in my Downloads folder", "files"),
+            ("What’s your favorite color?", "talk"),
+            ("Write a bash script to list all files in a directory", "code"),
+            ("Find recent X posts about electric cars", "web"),
+            ("Hey, you doing okay?", "talk"),
+            ("Locate the file ‘family_photo.jpg’ on my system", "files"),
+            ("Search the web for beginner guitar lessons", "web"),
+            ("Write a Python function to reverse a string", "code"),
+            ("What’s the weirdest animal you know of?", "talk"),
+            ("Organize all .pdf files on my desktop into a ‘Documents’ folder", "files"),
+            ("Browse the web for the latest space mission updates", "web"),
+            ("Hey, what’s up with you today?", "talk"),
+            ("Write a JavaScript function to add two numbers", "code"),
+            ("Find the file ‘notes.txt’ in my Documents folder", "files"),
+            ("Tell me something random about the ocean", "talk"),
+            ("Search the web for cheap flights to Paris", "web"),
+            ("Check if ‘budget.xlsx’ is on my drive", "files"),
+            ("Write a Python script to count words in a text file", "code"),
+            ("How’s it going today?", "talk"),
+            ("Find recent X posts about AI advancements", "web"),
+            ("Move all .jpg files from Downloads to a ‘Photos’ folder", "files"),
+            ("Search online for the best laptops of 2025", "web"),
+            ("What’s the funniest thing you’ve heard lately?", "talk"),
+            ("Write a Ruby script to generate random numbers", "code"),
+            ("Hey, how’s everything with you?", "talk"),
+            ("Locate ‘meeting_agenda.docx’ in my system", "files"),
+            ("Search the web for tips on growing indoor plants", "web"),
+            ("Write a C++ program to calculate the sum of an array", "code"),
+            ("Tell me a fun fact about dogs", "talk"),
+            ("Check if the folder ‘Old_Projects’ exists on my desktop", "files"),
+            ("Browse the web for the latest gaming console reviews", "web"),
+            ("Hi, how are you feeling today?", "talk"),
+            ("Write a Python script to check disk space", "code"),
+            ("Find the file ‘vacation_itinerary.pdf’ on my drive", "files"),
+            ("Search the web for news on renewable energy", "web"),
+            ("What’s the strangest thing you’ve learned recently?", "talk"),
+            ("Organize all video files into a ‘Videos’ folder", "files"),
+            ("Write a shell script to delete temporary files", "code"),
+            ("Hey, how’s your week been so far?", "talk"),
+            ("Search online for the top movies of 2025", "web"),
+            ("Locate ‘taxes_2024.xlsx’ in my Documents folder", "files"),
+            ("Tell me about a cool invention from history", "talk"),
+            ("Write a Java program to check if a number is even or odd", "code"),
+            ("Find recent X posts about cryptocurrency trends", "web"),
+            ("Hey, you good today?", "talk"),
+            ("Search the web for easy dinner recipes", "web"),
+            ("Check if ‘photo_backup.zip’ exists on my drive", "files"),
+            ("Write a Python script to rename files with a timestamp", "code"),
+            ("What’s your favorite thing about space?", "talk"),
+            ("Browse the web for the latest fitness trends", "web"),
+            ("Move all .docx files to a ‘Work’ folder", "files"),
+            ("I would like to make a new project called 'new_project'", "files"),
+            ("I would like to setup a new project index as mark2", "files"),
+            ("can you create a 3d js game that run in the browser", "code"),
+            ("can you make a web app in python that use the flask framework", "code"),
+            ("can you build a web server in go that serve a simple html page", "code"),
+            ("can you find out who Jacky yougouri is ?", "web"),
+            ("Setup a new flutter project called 'new_flutter_project'", "files"),
+            ("can you create a new project called 'new_project'", "files"),
+            ("can you make a simple web app that display a list of files in my dir", "code"),
+            ("can you build a simple web server in python that serve a html page", "code"),
+            ("find and buy me the latest rtx 4090", "web"),
+            ("What are some good netflix show like Altered Carbon ?", "web"),
+            ("can you find the latest research paper on AI", "web"),
+            ("can you find research.pdf in my drive", "files"),
+            ("hi", "talk"),
+            ("hello", "talk"),
+        ]
+        texts = [text for text, _ in few_shots]
+        labels = [label for _, label in few_shots]
+        self.talk_classifier.add_examples(texts, labels)
+
+    def llm_router(self, text: str) -> tuple:
+        """
+        Inference of the LLM router model.
+        Args:
+            text: The input text
+        """
+        predictions = self.talk_classifier.predict(text)
+        predictions = [pred for pred in predictions if pred[0] not in ["HIGH", "LOW"]]
+        predictions = sorted(predictions, key=lambda x: x[1], reverse=True)
+        return predictions[0]
+    
+    def router_vote(self, text: str, labels: list, log_confidence:bool = False) -> str:
+        """
+        Vote between the LLM router and BART model.
+        Args:
+            text: The input text
+            labels: The labels to classify
+        Returns:
+            str: The selected label
+        """
+        result_bart = self.pipelines['bart'](text, labels)
+        result_llm_router = self.llm_router(text)
+        bart, confidence_bart = result_bart['labels'][0], result_bart['scores'][0]
+        llm_router, confidence_llm_router = result_llm_router[0], result_llm_router[1]
+        final_score_bart = confidence_bart / (confidence_bart + confidence_llm_router)
+        final_score_llm = confidence_llm_router / (confidence_bart + confidence_llm_router)
+        if log_confidence:
+            pretty_print(f"Agent choice -> BART: {bart} ({final_score_bart}) LLM-router: {llm_router} ({final_score_llm})")
+        return bart if final_score_bart > final_score_llm else llm_router
+    
     def classify_text(self, text: str, threshold: float = 0.4) -> list:
         """
-        Classify the text into labels (agent roles).
+        Classify the text using the LLM router and BART model.
-        Args:
-            text (str): The text to classify
-            threshold (float, optional): The threshold for the classification.
-        Returns:
-            list: The list of agents and their scores
         """
         first_sentence = None
+        lang = "en"
         for line in text.split("\n"):
             first_sentence = line.strip()
             break
         if first_sentence is None:
             first_sentence = text
-        result = self.pipeline(first_sentence, self.labels, threshold=threshold)
+        try:
-        return result
+            lang = self.lang_analysis.detect_language(first_sentence)
+            # no multilanguage support yet
+            labels = [agent.role["en"] for agent in self.agents]
+            result = self.router_vote(first_sentence, labels, log_confidence=False)
+        except Exception as e:
+            raise e
+        return result, lang
+    
+    def estimate_complexity(self, text: str) -> str:
+        """
+        Estimate the complexity of the text.
+        Args:
+            text: The input text
+        Returns:
+            str: The estimated complexity
+        """
+        predictions = self.complexity_classifier.predict(text)
+        predictions = sorted(predictions, key=lambda x: x[1], reverse=True)
+        if len(predictions) == 0:
+            return "LOW"
+        complexity, confidence = predictions[0][0], predictions[0][1]
+        if confidence < 0.4:
+            return "LOW"
+        if complexity == "HIGH" and len(text) < 64:
+            return None # ask for more info
+        if complexity == "HIGH":
+            return "HIGH"
+        elif complexity == "LOW":
+            return "LOW"
+        pretty_print(f"Failed to estimate the complexity of the text. Confidence: {confidence}", color="failure")
+        return None
+    
+    def find_planner_agent(self) -> Agent:
+        """
+        Find the planner agent.
+        Returns:
+            Agent: The planner agent
+        """
+        for agent in self.agents:
+            if agent.type == "planner_agent":
+                return agent
+        pretty_print(f"Error finding planner agent. Please add a planner agent to the list of agents.", color="failure")
+        return None
+    
+    def multi_language_message(self, text: str):
+        pretty_print(f"选择代理时出错。路由系统尚不支持多语言", color="failure")
+        pretty_print(f"エージェントの選択エラー。ルーティングシステムはまだ多言語に対応していません", color="failure")
+        pretty_print(f"Erreur lors du choix de l'agent. Le système de routage n'est pas encore multilingue.", color="failure")
+        pretty_print(f"Error al elegir agente. El sistema de enrutamiento aún no es multilingüe.", color="failure")
     
     def select_agent(self, text: str) -> Agent:
         """
@@ -57,59 +371,76 @@ class AgentRouter:
         Returns:
             Agent: The selected agent
         """
-        if len(self.agents) == 0 or len(self.labels) == 0:
+        if len(self.agents) == 0:
             return self.agents[0]
-        result = self.classify_text(text)
+        complexity = self.estimate_complexity(text)
+        best_agent, lang = self.classify_text(text)
+        if lang != "en":
+            self.multi_language_message(text)
+        if complexity == None:
+            pretty_print(f"Humm, the task seem complex but you gave very little information. can you clarify?", color="info")
+            return None
+        if complexity == "HIGH" and lang == "en":
+            pretty_print(f"Complex task detected, routing to planner agent.", color="info")
+            return self.find_planner_agent()
         for agent in self.agents:
-            if result["labels"][0] == agent.role:
+            if best_agent == agent.role[lang]:
-                pretty_print(f"Selected agent: {agent.agent_name} (roles: {agent.role})", color="warning")
+                pretty_print(f"Selected agent: {agent.agent_name} (roles: {agent.role[lang]})", color="warning")
                 return agent
+        pretty_print(f"Error choosing agent.", color="failure")
         return None
 
 if __name__ == "__main__":
     agents = [
-        CasualAgent("deepseek-r1:14b", "jarvis", "../prompts/casual_agent.txt", None),
+        CasualAgent("jarvis", "../prompts/casual_agent.txt", None),
-        BrowserAgent("deepseek-r1:14b", "browser", "../prompts/planner_agent.txt", None),
+        BrowserAgent("browser", "../prompts/planner_agent.txt", None),
-        CoderAgent("deepseek-r1:14b", "coder", "../prompts/coder_agent.txt", None)
+        CoderAgent("coder", "../prompts/coder_agent.txt", None),
+        FileAgent("file", "../prompts/coder_agent.txt", None)
     ]
     router = AgentRouter(agents)
     texts = [
+        "hi",
+        #"你好",
+        #"Bonjour",
         "Write a python script to check if the device on my network is connected to the internet",
-        #"Peut tu écrire un script python qui vérifie si l'appareil sur mon réseau est connecté à internet?",
+        # "Peut tu écrire un script python qui vérifie si l'appareil sur mon réseau est connecté à internet?",
-        #"写一个Python脚本，检查我网络上的设备是否连接到互联网",
+        # "写一个Python脚本，检查我网络上的设备是否连接到互联网",
         "Hey could you search the web for the latest news on the tesla stock market ?",
-        #"嘿，你能搜索网页上关于股票市场的最新新闻吗？",
+        # "嘿，你能搜索网页上关于股票市场的最新新闻吗？",
-        #"Yo, cherche sur internet comment va tesla en bourse.",
+        # "Yo, cherche sur internet comment va tesla en bourse.",
         "I would like you to search for weather api and then make an app using this API",
-        #"我想让你搜索天气API，然后用这个API做一个应用程序",
+        # "我想让你搜索天气API，然后用这个API做一个应用程序",
-        #"J'aimerais que tu cherche une api météo et que l'utilise pour faire une application",
+        # "J'aimerais que tu cherche une api météo et que l'utilise pour faire une application",
         "Plan a 3-day trip to New York, including flights and hotels.",
-        #"计划一次为期3天的纽约之旅，包括机票和酒店。",
+        # "计划一次为期3天的纽约之旅，包括机票和酒店。",
-        #"Planifie un trip de 3 jours à Paris, y compris les vols et hotels.",
+        # "Planifie un trip de 3 jours à Paris, y compris les vols et hotels.",
-        "Find me the latest research papers on AI.",
+        "Find on the web the latest research papers on AI.",
-        #"给我找最新的AI研究论文。",
+        # "在网上找到最新的人工智能研究论文。",
-        #"Trouve moi les derniers papiers de recherche en IA",
+        # "Trouve moi les derniers articles de recherche sur l'IA sur internet",
         "Help me write a C++ program to sort an array",
-        #"帮我写一个C++程序来排序数组",
+        "Tell me what France been up to lately",
-        #"Aide moi à faire un programme c++ pour trier une array.",
+        # "告诉我法国最近在做什么",
+        # "Dis moi ce que la France a fait récemment",
+        "Who is Sergio Pesto ?",
+        # "谁是Sergio Pesto？",
+        # "Qui est Sergio Pesto ?",
+        # "帮我写一个C++程序来排序数组",
+        # "Aide moi à faire un programme c++ pour trier une array.",
         "What’s the weather like today? Oh, and can you find a good weather app?",
-        #"今天天气怎么样？哦，你还能找到一个好的天气应用程序吗？",
+        # "今天天气怎么样？哦，你还能找到一个好的天气应用程序吗？",
-        #"La météo est comment aujourd'hui ? oh et trouve moi une bonne appli météo tant que tu y est.",
+        # "La météo est comment aujourd'hui ? oh et trouve moi une bonne appli météo tant que tu y est.",
         "Can you debug this Java code? It’s not working.",
-        #"你能调试这段Java代码吗？它不起作用。",
+        # "你能调试这段Java代码吗？它不起作用。",
-        #"Peut tu m'aider à debugger ce code java, ça marche pas",
+        # "Peut tu m'aider à debugger ce code java, ça marche pas",
-        "What's the latest brainrot on the internet ?",
+        #"Can you browse the web and find me a 4090 for cheap?",
-        #"互联网上最新的“脑残”是什么？",
+        #"你能浏览网页，为我找一个便宜的4090吗？",
-        #"Quel est la dernière connerie sur internet ?",
+        #"Peut tu chercher sur internet et me trouver une 4090 pas cher ?",
-        "i would like to setup a new AI project, index as mark2",
+        #"Hey, can you find the old_project.zip file somewhere on my drive?",
-        #"我想建立一个新的 AI 项目，索引为 Mark2",
-        #"Je voudrais configurer un nouveau projet d'IA, index Mark2",
-        "Hey, can you find the old_project.zip file somewhere on my drive?",
         #"嘿，你能在我驱动器上找到old_project.zip文件吗？",
         #"Hé trouve moi le old_project.zip, il est quelque part sur mon disque.",
         "Tell me a funny story",
-        #"给我讲一个有趣的故事",
+        "给我讲一个有趣的故事",
-        #"Raconte moi une histoire drole"
+        "Raconte moi une histoire drole"
     ]
     for text in texts:
         print("Input text:", text)

sources/tools/tools.py

@@ -35,6 +35,9 @@ class Tools():
         self.current_dir = self.create_work_dir()
         self.excutable_blocks_found = False
     
+    def get_work_dir(self):
+        return self.current_dir
+    
     def check_config_dir_validity(self):
         """
         Check if the config directory is valid.

sources/web_scripts/inject_safety_script.js

@@ -16,15 +16,6 @@ Element.prototype.requestFullscreen = function() {
 Element.prototype.requestPointerLock = function() {
     console.log('Blocked pointer lock');
 };
-// Block iframe creation (optional, since browser already blocks these)
-const originalCreateElement = document.createElement;
-document.createElement = function(tagName) {
-    if (tagName.toLowerCase() === 'iframe') {
-        console.log('Blocked iframe creation');
-        return null;
-    }
-    return originalCreateElement.apply(this, arguments);
-};
 //block fetch
 window.fetch = function() {
     console.log('Blocked fetch request');