Build your local AI assistant with LobeChat

LobeChat is an open‑source framework for building secure, local AI chat experiences. It supports file handling, knowledge bases, and multimodal inputs, and it supports Ollama to run and switch local LLMs.

Olares streamlines and simplifies the deployment of both, allowing you to skip complex manual environment configurations.

This guide covers the installation, configuration, and practical usage of these tools to create your personalized AI assistants.

Learning objectives

By the end of this guide, you are able to:

• Configure LobeChat to communicate with your local Ollama instance.
• Use LobeChat for specific scenarios such as content writing and coding.

Prerequisites

Before you begin, make sure:

• Ollama is installed and running in your Olares environment.
• At least one model is installed using Ollama. For more information, see Ollama.

Install LobeChat

1. From the Olares Market, search for "LobeChat".

   

2. Click Get, and then click Install. Wait for the installation to finish.

Configure the connection

After the installation is completed, you must connect LobeChat to Ollama to make the chat interface work.

1. Open LobeChat from Launchpad.

2. Click the LobeChat icon in the upper-left corner, and then go to Settings > Language Model.

   

3. (Optional) To obtain your local Ollama host address, go to Olares Settings > Application > Ollama > Ollama API > Set up endpoint, and then copy the endpoint address.

   

4. In the Ollama section, find the Interface proxy address field, and then enter your local Ollama address.

   

5. Click Check under the Ollama model list to verify the connection. A Check Passed message indicates that the proxy address is correct.

   

Install language models

By default, LobeChat lists supported models, including models that are not installed yet.

You can click Get Model List to fetch available models from Ollama.

To download additional models, use one of the following options:

• Download via LobeChat UI

  When you are chatting with a selected language model, if the language model is not installed, you are prompted right in the chat to download and install it.

  

• Download using the Ollama CLI. For more information, see Download a model.

Create an assistant

LobeChat allows you to create specialized assistants to handle specific tasks by leveraging various language models and combining them with functional plug-ins.

• Flexible model switching: You can switch language models instantly within the same chat to achieve the best results. For example, if you are not satisfied with a response, you can select a different model from the list to leverage their unique strengths.

• Plug-in extensions: You can also install plug-ins to extend and enhance the capabilities of your assistant.

  INFO

  To install plug-ins, ensure that you select a model compatible with Function Calling. Look for brick next to the model name, which indicates the model supports function calls.

The following steps outline the standard workflow for creating and configuring any assistant in LobeChat. You can apply this procedure using specific settings provided in the use scenarios section.

1. Create a new assistant:

   • From the left navigation pane, click New Assistant.
   • If you already have active chats, click add_comment to create a new one.

2. Configure the assistant such as identity and role:

   a. Click Open Chat Settings.

   

   b. On the Assistant Info tab, set the avatar, name, and description, and then click Update Assistant Information.

   

   c. On the Role Configuration tab, enter your prompt for this specific role to define its behavior, and then click OK.

   d. Close the Session Settings page to return to the chat window.

3. Select the language model from the basic interaction area.

   

4. (Optional) Install LobeChat plug-ins to enhance the assistant's capabilities:

   a. In the basic interaction area, hover over the plug-in icon and click Plugin Store.

   

   b. On the LobeHub Plugins tab, search for the target plug-in, and then click Install.

5. Interact with the assistant.

6. (Optional) Pin for quick access:

   If you are satisfied with the assistant's performance, hover over the assistant in the sidebar, click more_vert, and then click Pin to keep it accessible at the top of your list.

Use scenarios

The following scenarios provide some practical examples for your daily tasks. Apply these specific settings during the general creation procedure to build specialized assistants tailored to your workflow.

Polish content and visualize ideas

Create a specialized assistant to help you refine text and generate images based on descriptions.

Configurations

• Name: Writing Bot

• Role prompt:

  You are a creative editor. When I provide text, review it for clarity 
  and tone. When I describe a scene, use the drawing plug-in to generate 
  an image based on my description.

• Language model: qwen2.5:7b

  INFO

  qwen2.5:7b excels at various NLP tasks such as contextual understanding and content writing. It is also compatible with functional calling, so you can install LobeChat plug-ins for enhanced capabilities.

• LobeChat plug-in: "Pollinate drawing", which is used to create images based on description

  

Interaction

1. Enter and send your draft content to get a refined version.
2. Hover over the plug-in icon to ensure that Pollinate drawing is enabled, and then ask the assistant to create a cover image for the content.
3. Brainstorm and iterate with the language model to get your ideal content textually and visually.

Coding assistant

Create a specialized assistant to help you write efficient code and act as a dedicated pair programmer.

Configurations

• Name: Dev Bot

• Role prompt:

  You are an expert developer. When I describe a task or requirement, 
  generate clean, efficient, and well-commented code to solve it.

• Language model: deepseek-coder-v2

  INFO

  deepseek-coder-v2 is good at coding use cases such as code generation and long text understanding.

Interaction

1. Describe a data generation task and send to the chat.

   Write a Python script to generate a CSV file named employees.csv with
   20 rows of mock data. Columns should include: ID, Name, Department, 
   and Salary. Use the random library to generate varied data.

2. The assistant processes your request and generates a standalone Python script with explanation.

   import csv
   import random
   
   def generate_mock_data():
      departments = ['HR', 'Engineering', 'Marketing', 'Sales', 'Finance']
      filename = "employees.csv"
   
      print(f"Generating {filename}...")
   
      with open(filename, 'w', newline='', encoding='utf-8') as csvfile:
         fieldnames = ['ID', 'Name', 'Department', 'Salary']
         writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
         writer.writeheader()
   
         for i in range(1, 21):
               writer.writerow({
                  'ID': f'EMP{i:03d}',
                  'Name': f'Employee {i}',
                  'Department': random.choice(departments),
                  'Salary': random.randint(50000, 120000)
               })
   
      print(f"Successfully created {filename} with 20 records.")
   
   if __name__ == "__main__":
      generate_mock_data()

3. Run the generated code to verify.

   a. Copy the generated Python code block and save it as generate_data.py.

   b. Open the Terminal, navigate to the folder, and run the following command:

   python3 generate_data.py

   c. Check your current folder. You should see a new file named employees.csv. Open it to verify the generated mock data.

   

Real-time news analyst

Build an assistant that keeps you updated with the latest technology trends. By using the Website Crawler plug-in, this assistant can read live news sites and provide instant summaries of what's happening right now.

Configurations

• Name: Daily Tech Digest

• Role prompt:

  You are a tech news reporter. When I send you a news site URL, 
  read the headlines and summarize the latest top five stories for me.
  Limit the list to five.

• Language model: qwen2.5:7b

  INFO

  qwen2.5:7b excels at various NLP tasks such as contextual understanding and content writing. It is also compatible with functional calling, so you can install LobeChat plug-ins for enhanced capabilities.

• LobeChat plug-in: "Website Crawler", which is used to access live web pages and analyze real-time content from provided URLs

  How Website Crawler works (Real-time vs. Offline)

  Standard local AI models are offline and rely on pre-trained data from the past. The Website Crawler plug-in, specifically the getWebsiteContent function, acts as a bridge to the live internet.

  When you provide a URL, the plug-in instantly accesses the web page in real time via an API, fetches the current content, and feeds it to the AI. This ensures that the AI model is accessing the latest live web content rather than using the old memory.

  

Interaction

1. In the basic interaction area, hover over the plug-in icon to ensure that the Website Crawler plug-in is enabled.
2. Send the URL address to the chat. For example, https://github.com/trending.
3. Paste and send the URL to the chat. The assistant lists specific news stories with summaries.

FAQ

Why did the connection check fail when I connected to Ollama?

This usually happens due to network or authentication settings. Follow these steps to resolve it:

1. Open Settings, and ensure that the Authentication level for Ollama is set to Internal.
2. Turn on the VPN in the LarePass desktop client, and confirm that the status changes to Intranet or P2P.
3. Run the connection check again.

TIP

If you are on the same local network as your device and the Authentication level for Ollama is set to None, you do not need to enable the LarePass VPN.

For more information, see Ensure network connectivity.