1 Introduction

My usual habits for using AI have actually always been very “simple”—the priority has always been like this: 1. Official ChatGPT Client(Using the free quota, since I don't use it much anyway; if I occasionally exceed it, I just wait a few hours, treating it as a forced break); 2. LobeChat Frontend + OhMyGPT As an emergency backup (used when the free quota is insufficient, the official service is glitching, or when I need to connect to different models).

For friends who are unfamiliar with the deployment, setup, and use of LobeChat, you can refer to my previous articles:Docker Series: Detailed Deployment Tutorial Based on the Open Source Large Language Model UI Framework: LobeChat和Home Data Center Series: Unlocking the Full Potential of LobeChat: A Complete Guide from Setup to Practical Application。。

For friends who are unfamiliar with OhMyGPT, you can refer to my previous articles:Home Data Center Series: Starting Your AI Journey: A Detailed Introduction to Local Large Language Model UIs and Large Language Model API Providers。

Speaking of LobeChat, I actually appreciate its overall design:Multi-device data synchronization, support for extension plugins, and even the ability to customize some UI details. If you don't have the official ChatGPT client, LobeChat is an excellent alternative.

But for me, it has two “natural disadvantages”: my primary way of using AI is the official ChatGPT client (Mac version + iOS version), which natively supports data synchronization, so LobeChat's multi-device synchronization is not a must-have for me; I don't use it often, so the plugin system and complex configuration features provided by LobeChat feel a bit like “using a sledgehammer to crack a nut.” A more realistic issue is that there are two aspects of LobeChat in daily use that I'm not quite satisfied with:

Relatively slow response speed

Every time I click send and see that circle spinning on the assistant's avatar, I can't help but feel a bit anxious waiting. The official team also said they would optimize it, but after all this time, I feel the speed is still not fast—it might be that its underlying streaming mechanism naturally has latency, which isn't something that can be solved by simple adjustments.

Slightly poorer immersion on the browser side

Ever since I got used to the native experience of the official ChatGPT client, using a browser-based client (even a PWA) always makes me feel like the interaction lacks a bit of immersion—for example, window switching isn't smooth enough, and shortcut key support is limited.

For these reasons, this time after I switched my third-party API provider from OhMyGPT to OpenRouter (see article:Home Data Center Series: Highly Internationally Recognized ChatGPT API Relay Solution: Comprehensive Analysis of OpenRouter + Domestic Payment Guide), I had the idea of changing the frontend UI: I hoped to find a frontend tool with faster response speeds and smoother interactive experiences.

With this idea in mind, I put several common AI frontend tools together and made a panoramic comparison to see which one best meets my needs.

2 AI Frontend Tool Selection Map: From Browsers to Desktop Applications

1 Introduction to Three Representative Frontends on the Market

Let's first take a look at the three common AI frontend forms on the market and their representative products.

LobeChat: Full-featured, but heavy

LobeChat is an open-source, highly comprehensive frontend tool, positioned more like an “AI productivity workbench,” representing a feature-rich browser-side frontend with a beautiful interface and strong extensibility;

Its characteristics are very distinct. In terms of advantages, LobeChat supports multi-device sync, maintaining a consistent user experience whether on desktop browsers, mobile browsers, or via PWA; it also has a complete plugin extension capability, allowing access to external features such as search and drawing; the interface design is also quite beautiful, with a reasonable layout of functional modules.

However, it also has some shortcomings. The response speed is relatively slow, especially for users who are sensitive to streaming output latency, this speed might make them a bit impatient; although the features are rich, for people who just want to quickly complete a conversation, it may seem a bit bloated; moreover, it runs based on the browser, and even with PWA, it is still limited by the constraints of the browser interaction experience.

Target Audience: Users who need seamless data sync across devices and want to complete multiple AI tasks in a single tool.

BetterChatGPT: Lightweight but Single-purpose

BetterChatGPT is also an open-source browser-side frontend, belonging to the lightweight browser-side category, focusing on simplicity and response speed. Compared to LobeChat, it is characterized by being more lightweight, with features more focused on “chat” itself:

Its advantages lie in its simple interface and very fast onboarding, while being compatible with multiple APIs (including OpenAI, OpenRouter, etc.), flexible deployment, and low device performance requirements.

However, its features are relatively single-purpose, lacking a plugin ecosystem, and it does not have built-in multi-device sync capabilities—data mainly relies on local browser storage, which means session history cannot be directly shared between different devices. In addition, as a browser-side tool, it still falls short of native applications in terms of interactive immersion.

Target Audience: Users who want a lightweight, easy-to-deploy, simple chat tool with customizable APIs.

Chatbox: Immersive Experience of Native Apps

Chatbox takes another approach—it completely bypasses the browser, providing a desktop client version of native applications with cross-platform support (Windows / macOS / Linux), bringing an immersive experience closer to desktop software:

The interactive smoothness and immersion brought by native apps are significantly better than those of the browser side, and its response speed is very fast, with a smooth and natural streaming output experience. It also supports custom API Keys, allowing access to third-party APIs such as OpenRouter and OpenAI. However, Chatbox does not have a built-in automatic multi-device data sync feature (though it can be manually exported and imported), session history is saved locally, and there is no plugin ecosystem. Its features are relatively streamlined, making it more suitable for users who pursue a native experience and response speed.

Target Audience: Users who pursue a native application experience, want fast response, and do not rely on automatic multi-device sync.

2 Comparison of Three Frontends

Comparison Table of the Three

As can be seen from the comparison above, if your core need isMulti-device Sync + Feature Ecosystem, LobeChat is undoubtedly the most suitable; if you want aminimalist chat tool with customizable API, BetterChatGPT is a good choice.

And if you, like me, have an obsession withthe smooth interaction and faster response speed of native applications, then Chatbox is the only solution that fits.

In addition to being a frontend tool with locally configurable APIs, Chatbox also offers its own subscription-based AI service, divided into three tiers: Lite, Pro, and Pro+, with prices ranging from $3.99 to $33.3: Lite focuses on low cost and standard model access (2 million computing points per month, web search, image chat, etc.); Pro targets mid-to-high-end users (10 million points per month, access to advanced models, document chat, more drawing attempts); Pro+ is the flagship configuration (20 million points, full set of models and features, configuration-free knowledge base, and MCP integration):

The main advantage of the official service is “out-of-the-box” usability, eliminating the hassle of applying for API Keys, route optimization, and maintenance, with guaranteed speed and stability; it also comes with built-in web search, image recognition, document Q&A, and other features, saving the step of integrating external services. The disadvantage is that the price is relatively fixed, the choice of models depends on the list provided officially, and once the subscription is stopped, these features can no longer be used.

In contrast, bringing your own API Key (such as OpenRouter, OpenAI, Claude, Gemini, etc.) offers higher flexibility, allowing you to change model providers at any time, pay as you go, and even access some models not provided officially. However, you need to handle network availability, rate limits, account security, and other issues yourself, which presents a certain technical barrier for non-technical beginners.

3 Chatbox Installation and Settings

1 Chatbox Installation

Since it is a desktop version, you can just follow the conventional steps for installing applications on each platform, so I won't go into detail here. You can download the installation version suitable for your operating system from the official website (https://chatboxai.app/zh#download)：

Good news for iOS users is that Chatbox can currently be downloaded directly from the App Store in the China region:

2 Chatbox Adding OpenRouter as a Model Provider

Model Provideris one of the most core settings in Chatbox, used to manage the AI service platforms you connect to.

Through here, you can add, edit, or delete different API providers (such as OpenRouter, OpenAI, Anthropic, DeepSeek, etc.), and configure exclusive API host addresses, paths, keys, and other information for them. In other words, this area determines which AI platform Chatbox calls “behind the scenes” and the specific connection method used to interact with it.

Properly configuring model providers not only allows you to freely switch between models from different platforms in Chatbox, but also optimizes speed, stability, and cost based on your needs. Since I just switched from OhMyGPT to OpenRouter, I will use OpenRouter as an example for demonstration.

Since Chatbox does not include OpenRouter as a built-in model provider by default, we need to create a new one. We will use the macOS version of Chatbox as an example to demonstrate the steps for adding it.

From ”Settings” - “Model Provider” - “Add”:

Add a new model provider OpenRouter:

https://openrouter.ai

API Path is filled with:

/api/v1

As shown in the figure below:

3 Default Model

Default ModelThe settings area is used to specify the AI models that Chatbox prioritizes calling in different scenarios.

The configuration here will directly affect the experience of functions such as new sessions, topic naming, search construction, and image recognition:

• Default Chat Model: The model automatically selected for new chats, avoiding manual switching every time (for friends who don't know how to choose, you can set openai/gpt-5-mini, which is highly cost-effective). If set to “Auto”, it will continue to use the last used model.
• Default Topic Naming Model: The model used to automatically generate titles for sessions, which can be the same as the chat model or set separately.
• Search Term Construction Model: The model used to optimize search keywords when performing web searches.
• OCR Model: Used to extract text from images and then pass it to models that do not support image input for processing.

By properly setting these default values, you can let Chatbox automatically call the most appropriate model in different tasks, thereby reducing repetitive operations and improving efficiency.

Below is a simple comparison of several models currently provided by OpenAI that I personally find practical, for your reference:

Daily Use Recommendation: For most daily use scenarios, such as text generation, Q&A assistance, writing, and coding assistance, the advantage of GPT-5 Mini compared to GPT-4o Mini is not very obvious. GPT-5 Mini may be slightly more robust in complex conversations, multi-turn reasoning, or specific professional tasks, but for general users, this improvement is not highly perceptible.

From a cost-effectiveness perspective, GPT-4o Mini is significantly cheaper than GPT-5 Mini, while its performance is already sufficient to handle the vast majority of daily needs, making it a more cost-effective choice. The advantage of GPT-5 Mini mainly lies in potentially faster future feature updates, making it suitable for users who want to pursue the latest technology or occasionally perform complex tasks.

If you have a limited budget or mainly use it for daily tasks, choosing GPT-4o Mini will give you a stable, fast, and cost-controlled experience; if you want to use the latest GPT-5 series features in the long term and are willing to pay for slightly higher performance, you can consider GPT-5 Mini.

4 Web Search

web searchThe settings area determines which search service Chatbox will use to retrieve results when it needs to find real-time information.

You can choose freely based on speed, accuracy, and data source:

• Chatbox: Results are directly provided by Chatbox's built-in search function, which is fast and highly integrated, suitable for looking up simple, instant questions.
• Bing: Microsoft's search engine, good at synthesizing news, images, and web page information, suitable for scenarios requiring authority and broad information coverage.
• Tavily: A search service optimized specifically for AI applications, focusing more on extracting structured, useful key information from web pages, reducing interference from irrelevant content, and particularly suitable for tasks that require AI to quickly grasp key points.

After setting up the web search source, when the model needs to supplement external information, Chatbox will “go online to gather materials” in the way you selected, making the answers more timely and accurate.

Note: Actually, the web search function did not work when I used it. I don't know if it's because my posture was incorrect, or if there are other usage tips?

5 MCP

MCP (Model Context Protocol) It is an extension mechanism supported by Chatbox, which allows large language models to access additional tools and data sources, giving the models richer capabilities in conversations. Chatbox also has a built-in MCP server, but unfortunately, it is only open to users who have subscribed to Chatbox AI:

• Fetch Can crawl web page content and automatically convert it into easier-to-read Markdown;
• Sequential Thinking Allows the model to reason step-by-step before answering, reducing hasty mistakes;
• EdgeOne Pages Can quickly deploy HTML content and generate public access links;
• arXiv Allows AI to directly retrieve the latest papers and extract summaries;
• Context7 Can obtain the latest documentation and examples for any programming library.

For friends who have not subscribed to Chatbox AI and just use Chatbox as a frontend, you can also add Custom MCP Serverhere to extend more exclusive capabilities for the model, such as accessing corporate intranet documents, private databases, etc., to achieve a true “exclusive AI assistant”.

Chatbox comes with a huge list of MCP services, covering almost everything from web scraping and search engines to interfaces for various cloud services, databases, office platforms, and design tools, looking dazzling. However, please note that these services are like mobile app stores—although there are many, only a few categories are truly stable, easy to use, and readily accessible to ordinary users (such as search or information retrieval types like Fetch, Context7, and Bing/Tavily).

Some of these services can be used with a single click, while others require you to register a developer account, apply for an API Key, configure authorization, etc. The barrier to entry is not low, making them more suitable for technical personnel or teams with clear business needs. Therefore, there is no need to be intimidated or misled by the long list; just picking a few MCP services that match your needs and are easy to get started with is enough.

I have organized the MCP services provided by Chatbox into the following two tables based on ”commonly used” and ”high barrier to entry” for your reference.

Commonly Used & Easy-to-Use MCP Services (Suitable for Ordinary Users):

High Barrier & Professional MCP Services (Suitable for Dev/Ops/Specific Business Scenarios):

6 Knowledge Base

Knowledge Base This is the dedicated place in Chatbox to “feed” your AI. You can import documents, images, notes, and other materials here, allowing the AI to not only rely on its general knowledge during conversations but also reference your own exclusive information—equivalent to giving it a private reference manual:

When creating a knowledge base, you can choose the model provider (either Chatbox AI's built-in provider or a custom provider you have connected) and set different model types:

• embedding model: Responsible for converting your text into feature vectors, making it easy for the AI to retrieve content efficiently.
• Rerank Model (Optional): Further optimizes the ranking based on search results to improve answer accuracy.
• Vision Model (Optional): Can preprocess image files, allowing the AI to recognize and understand the content within them.

To build your own ”knowledge base,” you need to feed your own materials (documents, notes, images, etc.) to the AI, allowing it to “remember” this content during conversations and answer questions based on it. For people who need to handle corporate materials, technical documents, or personal study notes, this can significantly reduce instances of “the AI giving irrelevant answers.” Once configured, your AI can call upon the knowledge base for real-time references during chats—whether it's technical documentation, company policies, or a travel plan you made, it can be pulled up and used on the spot at any time.

Building your own ”knowledge base” is not highly difficult to get started with—just import files and select an embedding model. However, if you want it to perform at its best, you still need to spend some effort organizing the content (removing irrelevant information, chunking reasonably, and choosing a better embedding model). Ordinary users can use it as a small-scale “personal knowledge assistant,” but don't expect it to effortlessly handle thousands of documents; the scale and retrieval accuracy are limited by local processing capabilities.

In contrast,LobeChat “s knowledge base is architecturally more ”extensible,“ supporting connections to external vector databases, long-text storage, etc., making it suitable for technical users to build large-scale RAG solutions; whereas Chatbox takes an ”out-of-the-box" approach, running in just a few minutes, making it suitable for users who want to quickly try it out or use it on a small scale.

Note: Chatbox's knowledge base feature relies on embedding model to vectorize documents for retrieval. Platforms like OpenRouter、OhMyGPT which are third-party multi-model access platforms, only provide generative LLM calls themselves and do not provide embedding model APIs. Therefore, if you want to use the knowledge base in Chatbox, you must either use the officially provided embedding model API (such as OpenAI's text-embedding-3-small/large) or build your own embedding model API (such as using Ollama's nomic-embed-text, which I will introduce in a dedicated article later).

For most individual users or small teams, whether to build a custom knowledge base requires careful evaluation. When the volume of documents is small and the structure is simple, the AI can meet most needs through regular conversations or web searches, even without relying on a knowledge base. Building a knowledge base involves tasks like document organization, vectorization, and index management; even lightweight tools require a certain amount of configuration and maintenance effort. If documents are updated frequently, continuous investment is required, otherwise the effectiveness will decline. Overall, for individual users or small teams, a knowledge base is more of an auxiliary tool that brings convenience but is not a necessity; the investment and return need to be weighed.

7 Other Settings

In addition to the core settings features introduced earlier, Chatbox also providesChat Settings, Keyboard Shortcuts, and General Settingsthree categories of options to optimize the user experience. For example, you can adjust how conversations are displayed, set shortcuts for common actions, or modify the interface and startup behavior. These options lean toward personal preferences, and you can adjust them according to your own habits.

Chat Settings:

Keyboard Shortcuts:

General Settings:

At this point, all the main settings areas of Chatbox have been introduced. The previous sections (Model Providers, Web Search, MCP, Knowledge Base) determine its “capability ceiling,” while these daily settings are more like the icing on the cake, helping you shape Chatbox into a more convenient tool.

4 Chatbox Daily Use

1 Using the Chat Area

Entering Chatbox's core area—the chat area—you will find it is almost no different from everyday chat tools: type your question, press Enter, and wait for the reply to complete a natural AI conversation. There is almost no learning curve to get started.

What makes Chatbox special is that it provides many ready-made “assistants.” These assistants can be understood as preset AIs for different scenarios: writing assistants, translation assistants, programming assistants, Xiaohongshu copywriting assistants... You only need to choose the appropriate assistant according to your needs to directly enter the corresponding conversation scenario, without having to figure out prompts or parameters yourself:

This approach makes the user experience much more lightweight: whether it's writing an email, looking up a concept, debugging code, or chatting with AI about life topics, you can switch to the right assistant in seconds and start communicating immediately. Compared to those advanced settings mentioned earlier, this is the simplest and most direct part—opening Chatbox is like opening a “chat tool filled with assistants.”

Note: The default Image Creator (image creation assistant) is only available to users subscribed to Chatbox AI. Unsubscribed users can simply create an image assistant of their own.

In addition, the functional options provided below the Chatbox assistant chat box are very intuitive, making it easy for users to expand their interaction methods during the chat:

You can click “New Topic” to start an independent conversation thread and keep the context clear; through “Add Image” 或 “Select File” to upload various types of files to help the AI understand the content; use “Add Link” to provide web pages or document sources for the model to reference external information;“Select MCP Server” to allow switching between different custom model interfaces, adding specific functions or accessing private data for the conversation;“Knowledge Base” feature to let the AI query imported documents or notes for more precise content retrieval; and “Web Search” allows the AI to access the internet in real-time to get the latest data or information. Combined, these features make chatting not just about text input, but also integrate multimodal information, customization capabilities, and dynamic content, greatly enhancing the flexibility and practicality of interaction.

2 Customizing Assistants

If you have more personalized needs, you can also customize your own assistant. For example, you can set a specific tone or role for it, or bind certain exclusive functions to make it truly “your assistant.” Taking the translation assistant as an example:

3 Summary

In a side-by-side comparison, the number and variety of built-in assistants in Chatbox are indeed not as diverse as those in LobeChat. LobeChat's assistant ecosystem is more like an “app store,” where you can find preset roles of almost any style and for any purpose:

On the other hand, the assistants provided by Chatbox are relatively streamlined, and the coverage is not as fancy. However, looking at it from another perspective, for most daily users, this level of simplicity is actually enough. Common writing, translation, programming, and learning assistants are all available, easily covering over 80% of scenarios (and you can also create your own). Not to mention, Chatbox has a very obvious advantage in response speed, allowing you to get results faster and use it more smoothly.

This is like a trade-off between an “all-in-one feature bundle” and an “efficient, lightweight version”: LobeChat gives you richer choices, suitable for those who like to tinker and pursue a personalized experience; Chatbox emphasizes speed and simplicity, suitable for users who want an “out-of-the-box, no-nonsense” experience. After all, you can't have your cake and eat it too; which one you choose depends on what you value more.

5 Advanced Tips: Syncing Chatbox Data Across Devices

1 Overview

Chatbox itself does not provide an official cross-device data synchronization feature, but through cloud drives and symbolic links (symlinks), we can indirectly achieve multi-device sharing of chat history, settings, knowledge bases, and other data. It should be noted that this method relies on the file system characteristics of the operating system, and the implementation methods differ slightly across different systems.

5.2 macOS

On macOS, you can use iCloud Drive to sync data. Suppose you move the Chatbox data to the iCloud Drive directory, for example:

~/iCloud Drive/Apps/Chatbox/xyz.chatboxapp.app

Then execute in the terminal:

# 先备份原始数据
mv ~/Library/Application\ Support/xyz.chatboxapp.app ~/Library/Application\ Support/xyz.chatboxapp.app.backup

# 创建符号链接
ln -s ~/iCloud\ Drive/Apps/Chatbox/xyz.chatboxapp.app ~/Library/Application\ Support/xyz.chatboxapp.app

This way, Chatbox will continue to read data from the original path, but the data is actually synced in iCloud, allowing multiple Mac devices to share the same data.

5.3 Windows

On Windows, you can use OneDrive to sync data. Suppose you move the Chatbox data to the OneDrive directory, for example:

C:\Users\<用户名>\OneDrive\Chatbox\xyz.chatboxapp.app

Then execute in the Command Prompt (Administrator mode):

REM 先备份原始数据
move "%APPDATA%\xyz.chatboxapp.app" "%APPDATA%\xyz.chatboxapp.app.backup"

REM 创建符号链接
mklink /D "%APPDATA%\xyz.chatboxapp.app" "C:\Users\<用户名>\OneDrive\Chatbox\xyz.chatboxapp.app"

Note: Replace with your own Windows username.

4 Summary

The essence of this method is to point Chatbox's data directory to the cloud drive sync directory, achieving “indirect sync”. Although it is not an official solution, in personal and small team scenarios, it can achieve an effect almost equivalent to cross-device synchronization, though there are some precautions.

• System Limitations: macOS's iCloud Drive can only sync between Mac devices, and Windows's OneDrive can only sync between Windows devices. To achieve cross-system (e.g., Mac ↔ Windows) sync, you must rely on third-party cloud drives (such as Dropbox, Google Drive).
• Sync Delay: Cloud drive synchronization will have some delay, especially when the data volume is large or contains a large number of small files, so you need to wait patiently.
• Risk of Simultaneous Multi-Device Use: Avoid opening Chatbox on multiple devices at the same time, otherwise it may cause database conflicts or data corruption.

The applicable scenarios for this method are as follows:

• Single System, Multiple Devices: If you mainly use Chatbox on multiple devices of the same system (such as multiple Macs or multiple Windows PCs), cloud drive sync is the most direct and efficient.
• Individuals or Small Teams: Limited data volume, moderate usage frequency, avoiding investment in complex enterprise-level sync solutions.
• Seeking Cross-Device Convenience: Hoping to keep chat history, settings, knowledge bases, etc., consistent, reducing repetitive configuration and improving efficiency.

Overall, this method is a clever workaround: It achieves personal cross-device data synchronization without relying on official features. It avoids complex deployment and meets daily usage needs, but there are still certain risks, such as sync conflicts or delay issues, making it unsuitable for large-scale enterprise scenarios.

This cross-device automatic sync solutionis only applicable between desktop clients (Mac/Windows), and due to the limitations of iOS's sandbox mechanism, currently seamless sync between desktop and iOS is not possible, however, ”non-seamless” sync is still possible, as Chatbox itself provides data backup and data recovery functions:

6 Afterword

In summary, Chatbox provides quite comprehensive and practical features in terms of installation convenience, interface operation, assistant selection, custom MCP extensions, knowledge base management, and cross-device synchronization. For average users, the built-in assistants and intuitive interface can already meet the vast majority of daily usage needs without additional configuration. For advanced users, the optional custom MCP server and knowledge base features offer high flexibility, allowing users to build their own exclusive AI assistants according to their needs, achieving more efficient information processing and knowledge management.

From the perspective of domestic users in China, Chatbox's desktop client supports multi-platform downloads, and the iOS client can be obtained directly from the China App Store without bypassing regional restrictions or performing tedious configurations. This is extremely convenient for the vast majority of domestic users (no need to mess around with frontends like LobeChat, BetterChatGPT, etc.). Combined with the multi-model access platform OpenRouter introduced in my previous article, users can both call ChatGPT or other mainstream AI models and flexibly extend exclusive functions in local or self-built environments. Whether for quick-response daily conversations or custom management of data and knowledge bases, this combination is lightweight and efficient.

Overall, the Chatbox + OpenRouter solution strikes a good balance between feature completeness, ease of operation, and controllability. For domestic users who want to use mainstream foreign AI services but do not want too much hassle, it is undoubtedly an ideal path. This not only lowers the barrier to entry but also provides a reliable and efficient AI assistant experience for daily work, study, and creation, while retaining space for personalized extensions and advanced operations.