AgenticSeek: 100% Local Alternative to Manus AI

A 100% local alternative to Manus AI, a voice-enabled AI assistant that can browse web pages, write code and plan tasks independently while keeping all data on your device. Tailored for on-premises inference models, run entirely on your own hardware, ensuring full privacy and zero cloud dependency.

Main functions

• Fully localized and privacy protection - All features run on your device - no cloud service, no data sharing. Your files, conversations, and searches are always private.

• Intelligent web browsing - AgenticSeek can browse the Internet independently - search, read, extract information, fill in web forms - no manual operation is required throughout the process.

• Autonomous encoding assistant - Need code? It can write, debug and run programs in multiple languages ​​such as Python, C, Go, Java, etc. - without supervision throughout the process.

• Smart proxy selection - You ask, it will automatically select the agent that is most suitable for the task. It's like having a team of experts on standby.

• Planning and executing complex tasks - From travel planning to complex projects - It breaks down large tasks into steps and uses multiple AI agents to complete the work.

• Voice function - Clear, fast, and futuristic voice and voice-to-text functions, allowing you to talk to your personal AI assistant like in a science fiction movie.

Installation

Make sure Chrome is installed driver, Docker and Python 3.10 (or newer).

We strongly recommend that you use Python 3.10 for settings, otherwise a dependency error may occur.

For questions about Chrome driver, see the Chromedriver section.

1️⃣ Copy repository and set environment variables

git clone https://github.com/Fosowl/agenticSeek.git
cd agentSeek
mv .env.example .env

2️ Create a virtual environment

python3 -m venv agentic_seek_env
source agentic_seek_env/bin/activate
# On Windows: agentic_seek_env\Scripts\activate

3️⃣ Install the required kit

Automatic installation:

./install.sh

** To make the text-to-speech (TTS) function support Chinese, you need to install jieba (Chinese dictionary) and cn2an (Chinese digital conversion library): **

pip3 install jieba cn2an

Manual installation:

Note: For any operating system, make sure that the ChromeDriver you installed matches the version of Chrome you have installed. Run google-chrome --version . If your Chrome version > 135, see known issues

• Linux :

Update package list: sudo apt update

Installation dependencies: sudo apt install -y alsa-utils portaudio19-dev python3-pyaudio libgtk-3-dev libnotify-dev libgconf-2-4 libnss3 libxss1

Install ChromeDriver that matches your Chrome browser version: sudo apt install -y chromium-chromiumdriver

Installation requirements: pip3 install -r requirements.txt

• Macos :

Update brew: brew update

Installation chromedriver: brew install --cask chromedriver

Install portaudio: brew install portaudio

Upgrade pip: python3 -m pip install --upgrade pip

Upgrade wheel: pip3 install --upgrade setuptools wheel

Installation requirements: pip3 install -r requirements.txt

• Windows :

Install pyreadline3: pip install pyreadline3

Install portaudio manually (for example, via vcpkg or precompiled binary) and run: pip install pyaudio

Download and install chromedriver manually from the following URL: https://sites.google.com/chromium.org/driver/getting-started

Place chromedriver in a directory contained in your PATH.

Installation requirements: pip3 install -r requirements.txt

Run on a local machine AgenticSeek

It is recommended to use at least the model with DeepSeek 14B parameters above. Smaller models are difficult to use the assistant function and will soon forget the relationship between the contexts.

Native Run Assistant

Start your local provider, for example, using Ollama:

ollama serve

See the list of supported local providers below.

Update config.ini

Modify the config.ini file to set provider_name as a supported provider and set provider_model to the LLM supported by that provider. We recommend using models with reasoning capabilities such as *Qwen* or *Deepseek*.

See the FAQ section at the end of README for required hardware.

[MAIN]
is_local = True # Whether running locally or using a remote provider. 
provider_name = ollama # or lm-studio, openai, etc..
provider_model = deepseek-r1:14b # Choose a model that suits your hardware
provider_server_address = 127.0.0.1:11434
agent_name = Jarvis # name of your AI assistant
recover_last_session = True # whether to restore the previous session
save_session = True # whether to remember the current session
speak = True # text to speech
listen = False # Voice to text, only available for command line interfaces
work_dir = /Users/mlg/Documents/workspace # AgenticSeek's workspace. 
jarvis_personality = False # Whether to use a more "Jarvis" style personality, it is not recommended to use on small models
languages ​​= en zh # Language list, text to speech will default to using the first language in the list
[BROWSER]
headless_browser = True # Whether to use a headless browser, it is only recommended when using a web interface. 
stealth_mode = True # Use undetectable selenium to reduce browser detection

Warning: When running LLM using LM-studio, please *do not* set provider_name to openai . Please set it to lm-studio .

Note: Some providers (such as lm-studio) need to prefix their IP with http:// . For example, http://127.0.0.1:1234

List of local providers

Practical exercises

Run with an API

Set config.ini .

[MAIN]
is_local = False
provider_name = openai
provider_model = gpt-4o
provider_server_address = 127.0.0.1:5000

Warning: Make sure config.ini No end of line spaces.

If using native-based openai-based api, put is_local Set to True .

Change your IP to the openai-based API at the same time.

Next: Start services and run AgenticSeek

Start services and Run

(Start the service and run)

Activate your Python environment if needed.

source agent_seek_env/bin/activate

Start the required services. This will start docker-compose.yml , including:

• searxng

• redis (powered by redis)

• frontend

sudo ./start_services.sh # MacOS
start ./start_services.cmd # Windows

Option 1: Run using the CLI interface.

python3 cli.py

Option 2: Run using the web interface.

Note: Currently we recommend you to use the CLI interface. The web interface is still actively developing.

Start the backend service.

python3 api.py

Access http://localhost:3000/ , you should see the web interface.

Please note that the current web interface does not support message streaming.