AgenticSeek: an AI assistant for task automation that runs completely locally

General Introduction

AgenticSeek is an open source AI assistant that runs completely natively, based on the Deepseek R1 model development. It does not require a cloud API or incur hundreds of dollars in monthly fees. Its goal is to become an intelligent assistant similar to the movie "Jarvis", which can perform tasks such as code writing, file management, web browsing, etc. on the user's computer, while ensuring data privacy. Developers Martin. Legrand positioned it as Manus A local alternative to AI, emphasizing 100% local operation. As of March 2025, there are 181 stars and 16 forks on GitHub, and the community continues to be active and welcomes contributors.

Function List

• Fully local operation: No internet connection is required, all calculations are done on the user's device, and there is no data leakage.
• voice interaction: Supports voice input commands and voice responses to enhance operational convenience.
• file system operation: Browse and work with local files through Bash.
• Code writing and debugging: Support for code generation and bug fixing in Python, C, Golang, and more.
• autonomous error correction: When a command fails or a code error is encountered, the AI automatically retries and fixes it.
• Intelligent Tasking: Automatically selects the appropriate function module based on the type of task.
• Multi-agent collaboration: Complex tasks can be split and planned and executed by multiple agents.
• Tool Support: Built-in search, file exploration, and other utilities.
• memory function: Preserve conversation history and user preferences and support long contextual exchanges.
• web browser: Can navigate web pages and extract information on its own.

Using Help

The installation and use of AgenticSeek requires some technical grounding, but the steps are clear. Below is a detailed guide to ensure you get started quickly.

Installation process

1. environmental preparation
   • Make sure your computer supports Python 3, at least 8GB of RAM is recommended, and the GPU needs more than 8GB of video memory (12GB is recommended for 14B models).
   • Install Git for downloading code.
   • Download and install Ollama (Local Model Runner), which is a core dependency. Visit Ollama Official website Get it.
2. Download AgenticSeek
   • Open a terminal (CMD or PowerShell for Windows, Terminal for Mac/Linux).
   • Enter the command Clone Project:

     git clone https://github.com/Fosowl/agenticSeek.git
     

   • Go to the project catalog:

     cd agenticSeek
     

3. Creating a Virtual Environment
   • Enter the command to create a standalone environment:

     python3 -m venv agentic_seek_env
     

   • Activate the environment:
     • Windows:

       agentic_seek_env\Scripts\activate
       

     • Mac/Linux:

       source agentic_seek_env/bin/activate
       

4. Installation of dependencies
   • After activating the environment, run it:

     pip install -r requirements.txt
     

   • Or use an automated installation script (Mac/Linux):

     . /install.sh
     

5. Configure and download models
   • Ensure that Ollama is installed and start the service:

     ollama serve
     

   • downloading Deepseek R1 model (14B recommended, 7B for low configuration devices):

     ollama pull deepseek-r1:14b
     

   • compiler config.ini, set up to run locally:

     [MAIN]
     is_local = True
     provider_name = ollama
     provider_model = deepseek-r1:14b
     

6. Start AgenticSeek
   • Input:

     python main.py
     

   • Seeing the terminal prompt means it ran successfully.

Functional operation flow

voice interaction

• Method of opening: in config.ini set up in speak = True cap (a poem) listen = TrueThe
• utilization: Say "Start Voice Mode" when you start it, and then ask questions with your voice, such as "Write code for me".
• particulars: The AI will answer by voice, making sure the microphone and speakers are working.
• take note of: Volume may need to be adjusted for the first time, and there may be a delay on lower-end devices.

file system operation

• Setting up the catalog: in config.ini specified in work_dire.g. /Users/Your Username/DocumentsThe
• utilization: Enter "List all text files" and the AI will return the results.
• Advanced FeaturesIt says "analyze the contents of this file" and it extracts the key information.
• finesse: Support for Bash commands such as "create new folder".

Code writing and debugging

• Operating Methods: Type "Write a Python sorting algorithm" and the AI generates code.
• adjust components during testing: If there's an error in the code, say "fix this error" and the AI will analyze it and modify it.
• Supported Languages: Currently includes Python, C, Golang, will be expanded later.

web browser

• utilization: Type "Find the latest AI news" and the AI will navigate the page and organize the information.
• limitation: Currently only text is extracted, image or video parsing is not supported.
• finesse: You can specify keywords, such as "Search for Deepseek R1 updates".

autonomous error correction

• Operating Methods: No special settings are required. If the command fails, the AI automatically tries to repair it.
• typical example: If you enter the wrong command "Open File X", AI will indicate that the file does not exist and suggest an alternative.

Remote operation (optional)

• take: If the local device is underperforming, a server can be used to run the model.
• move::
  1. Start Ollama on the server and get the IP (e.g. 192.168.1.100).
  2. Run the server script:

     python server_ollama.py
     

  3. local modification config.ini::

     [MAIN]
     is_local = False
     provider_name = server
     provider_model = deepseek-r1:14b
     provider_server_address = 192.168.1.100:5000
     

  4. Running locally:

     python main.py
     

caveat

• hardware requirement: 7B models require 8GB of video memory, 14B requires 12GB (e.g. RTX 3060), and 32B requires 24GB+.
• adjust components during testing: If startup fails, check that Ollama is running, or check the terminal error log.
• update: Regular operation git pull Get the latest features.
• license: The project is open source and free under the GPL-3.0 license.

application scenario

1. Programming Aids
   You get stuck writing code, such as debugging complex logic. agenticSeek generates code and fixes errors, and explains code details.
2. Document management
   Computer files are disorganized. After setting up a directory, AI can list files, extract contents, and even perform batch operations.
3. Learning Support
   Students can use it to analyze documents, search for information, and take notes through voice interaction to improve efficiency.
4. Privacy mandate
   AgenticSeek runs locally without an Internet connection to ensure information security when handling sensitive data.

QA

1. Is it really 100% running locally?
   Yes, when using Ollama or Server mode, all processing is local, and non-local options (such as OpenAI) are only optional.
2. What if I don't have enough hardware?
   Use the 7B model, or run a higher model through a remote server.
3. What if something goes wrong?
   Make sure Ollama is running, check config.ini configuration. If it still fails, you can file an issue on GitHub.
4. What's better than Manus AI?
   It focuses on local operation and privacy protection for users who don't want to use cloud services.