NocoBase

Posted on Jun 3

Top 6 Open Source AI Tools by GitHub Stars for Stronger AI Agents

#ai #opensource #github #nocode

Originally published at https://www.nocobase.com/en/blog/6-github-top-open-source-ai-tools-for-ai-agents

Introduction

Over the past year, the number of open source AI tools has grown rapidly. On GitHub, the ai-tools topic has already gathered a large number of projects.

We found that many popular open source AI tools are designed to enhance the capabilities of existing AI systems and AI Agents.

They work like capability boosters for AI Agents: some help Agents operate webpages, some make Agents better at writing code, some allow Agents to call external tools, some make AI output more stable, and some help AI work inside real business systems.

So, in this article, we start from the most popular open source projects under GitHub’s ai-tools topic. Based on GitHub Stars, we will look at 6 of the most popular open source AI tools today and explain which AI capabilities they enhance.

💬 Hey, you're reading the NocoBase blog. NocoBase is the most extensible AI-powered no-code/low-code development platform for building enterprise applications, internal tools, and all kinds of systems. It’s fully self-hosted, plugin-based, and developer-friendly. → Explore NocoBase on GitHub

How did we select these tools?

To keep this article as objective as possible, we selected the tools based on publicly available GitHub data.

This selection mainly follows these rules:

Data source: GitHub Topics ai-tools
Sorting method: Based on GitHub Stars, shown from rank 6 to rank 1

Rank	Tool	GitHub Stars	Main direction	Best suited for
6	NocoBase	22.6k	AI-powered no-code business systems	Building enterprise internal systems such as CRM, approval, ticketing, and operations backends
5	OpenCLI	23.4k	AI tool interface / CLI	Packaging websites, local tools, or browser sessions into commands that AI Agents can call
4	OpenClaude	28.3k	AI Coding Agent	Using a multi-model AI Coding Agent in the terminal
3	prompt-optimizer	30.3k	Prompt optimization	Optimizing, testing, and reusing prompts
2	cc-switch	90k	AI Coding tool management	Managing configurations for AI Coding tools such as Claude Code, Codex, and OpenCode
1	browser-use	96.8k	Browser automation / AI Agent	Letting AI operate webpages and complete browser tasks

Data collected on June 3, 2026. Since GitHub Stars change over time, the actual numbers may be slightly different when you read this article.

NocoBase

Website: https://www.nocobase.com/

Online Demo: demo.nocobase.com/new

Documentation: docs.nocobase.com

GitHub: https://github.com/nocobase/nocobase

GitHub Stars: 22.6k

First open-sourced: 2021

License: Apache-2.0

Project introduction

NocoBase is an open source AI-powered no-code platform for quickly building enterprise internal business systems.

In NocoBase, AI can directly participate in application building. You can describe your business requirements in natural language and let AI help create data models, configure pages, orchestrate workflows, and even assist with plugin development.

This means AI is not simply generating a pile of code from scratch. Instead, it builds systems inside a platform that already has data models, pages, permissions, workflows, and a plugin system. After generation, users can continue checking, adjusting, and maintaining the system through NocoBase’s no-code interface.

What problem does it solve?

“How can AI build business systems inside a structured platform, instead of temporarily generating code in an empty project?”

In NocoBase, AI can work with the platform’s existing business system capabilities, including data models, page building, permission control, workflows, plugin extensions, and AI Employees. This lets AI speed up system building, while humans confirm business logic, adjust details, and control boundaries.

For enterprise internal systems, this approach is more reliable. It preserves AI efficiency while avoiding the maintenance problems that often come with generating code completely from scratch.

Best way to use it

NocoBase is best suited for building enterprise internal systems with clear structures and long-term iteration needs.

A better way to use it is:

Start with a specific business scenario

Do not begin by asking AI to “build me a complete ERP.” A better approach is to start with a specific scenario, such as customer management, purchase requests, equipment ledgers, after-sales tickets, contract approval, or project dashboards.

Use natural language to let AI build the system foundation

You can first describe the business objects, field relationships, page requirements, and workflow rules, then let AI create data models, configure pages, and orchestrate workflows in NocoBase. For example:

Help me build a customer management system, including customers, contacts, opportunities, and follow-up records. A customer can be linked to multiple contacts and opportunities. Each opportunity should include stage, amount, expected closing date, and owner.

Check and adjust in the no-code interface

After AI generates the system, do not treat it as the final version right away. Go into NocoBase’s visual interface and check whether the data tables, fields, relationships, page layouts, action buttons, and permission configurations match the real business needs.

Configure permissions and workflow boundaries

For internal systems, permissions and workflows are not optional. Different roles should see different data and perform different actions. Workflows involving approvals, notifications, status changes, and data validation should also have clear rules.

Let AI participate in system operations

After the system is built, AI can continue participating in business operations. For example, AI Employees can help with data analysis, report generation, translation, decision support, form filling, workflow node processing, and more.

Keep iterating instead of generating once

The strength of NocoBase is not “generate everything once.” After AI builds the system, teams can still adjust fields, pages, workflows, and permissions in a no-code way. This makes it more suitable for the long-term changes of real business systems.

Suitable scenarios

NocoBase is suitable for these scenarios:

CRM customer management systems
Approval systems
Ticketing systems
Project management systems
Lightweight ERP modules such as procurement, inventory, and orders
Operations backends and admin backends
Equipment management, asset management, and archive management
Customer follow-up and service management
Enterprise internal systems that require private deployment
Business applications that require permissions, workflows, data models, and continuous iteration
Teams that want to build systems with AI and no-code together

Usage boundaries

NocoBase’s core strength is business system building, but it is not suitable for every AI tool scenario.

Important boundaries include:

Not ideal for one-off web pages: If you only want to quickly generate a simple display page, campaign page, or one-off demo, NocoBase is not the most direct choice. It is better suited for business systems with data models, permissions, workflows, and long-term maintenance needs.
Not a full replacement for professional development frameworks: If you need highly customized frontend interactions, complex consumer-facing products, or internet applications for massive user bases, traditional development frameworks are still more suitable.
Business modeling is still needed upfront: NocoBase can speed up system building, but users still need to understand their business objects, data relationships, and workflow boundaries. If the data model is poorly designed, the system will become more complex later.
Complex customization still requires development skills: Although many functions can be configured without code, deep plugin development, complex integrations, and special business logic still require developers.

More precisely, NocoBase is suitable for building maintainable, extensible, and governable business systems. It is not meant to replace all AI tools. Instead, it provides a stable system foundation for AI to enter real business workflows.

Installation Prompt

You can copy the following Prompt to Claude Code, Cursor, Codex, OpenCode, or other AI Coding Agents:

Please help me install and initialize NocoBase locally.

Requirements:
1. Open the official GitHub repository:
   https://github.com/nocobase/nocobase

2. Read the official README and follow the AI Agent Access quickstart.

3. Check whether Node.js and npm are installed.

4. Install the NocoBase CLI beta version:
   npm install -g @nocobase/cli@beta

5. Create a new project directory:
   mkdir my-nocobase && cd my-nocobase

6. Initialize NocoBase with UI:
   nb init --ui

7. After initialization, explain what files and directories were created.

8. Start or restart the AI agent session inside this directory.
   For example:
   cd my-nocobase && codex

OpenCLI

Website: https://opencli.info/

GitHub: https://github.com/jackwener/OpenCLI

GitHub Stars: 23.4k

First open-sourced: March 2026

License: Apache-2.0

Project introduction

OpenCLI is an open source project that converts websites, browser sessions, Electron apps, and local tools into CLI interfaces. It can turn GUI-based tools into more predictable command-line interfaces, so both humans and AI Agents can complete actions through commands.

For example, it can package website capabilities into commands:

opencli hackernews top --limit 5
opencli bilibili hot --limit 5

It can also let AI Agents use opencli browser to operate Chrome pages where the user is already logged in, completing actions such as opening webpages, reading pages, clicking buttons, filling out forms, and extracting data.

Suitable scenarios

OpenCLI is suitable for these scenarios:

Turning common website capabilities into CLI commands
Letting AI Agents operate logged-in Chrome pages
Providing more stable external tool interfaces for Agents
Packaging operations for content platforms, community platforms, job sites, and social platforms
Extracting structured data from webpages
Registering local CLI tools for Agents to use
Creating adapters for repetitive web tasks
Letting tools such as Claude Code and Cursor call external website capabilities

Installation Prompt

You can copy the following Prompt to Claude Code, Cursor, Codex, or other AI Coding Agents:

Please help me install and configure OpenCLI locally.

Requirements:
1. Check whether Node.js >= 20 is installed.
2. Install OpenCLI using the official command:
   npm install -g @jackwener/opencli

3. Verify the installation:
   opencli --version

4. Install the OpenCLI Browser Bridge extension for Chrome.
   Prefer the Chrome Web Store installation if available.
   If not, guide me to install it manually from GitHub Releases.

5. Run the environment check:
   opencli doctor

6. If I have multiple Chrome profiles, help me list and rename them:
   opencli profile list
   opencli profile rename  work
   opencli profile use work

7. Run a simple test command:
   opencli list
   opencli hackernews top --limit 5

8. Explain what each step does.

9. Do not access private pages, send messages, publish content, delete data, or submit forms during the initial test.

10. After setup, explain:
   - How OpenCLI connects to Chrome
   - Where OpenCLI stores local configuration
   - How to safely use it with AI agents
   - What actions should require manual confirmation

OpenClaude

Website: https://openclaude.gitlawb.com/

GitHub: https://github.com/Gitlawb/openclaude

GitHub Stars: 28.3k

First open-sourced: March 2026

License: Custom license

Project introduction

OpenClaude is an open source AI Coding Agent CLI. It lets developers use different models and service providers in the terminal to complete code-related tasks, instead of being locked into a single model or a closed tool.

OpenClaude supports multiple model backends, including OpenAI-compatible APIs, Gemini, GitHub Models, Codex, Ollama, Atomic Chat, and more. It also supports common coding agent capabilities, such as reading files, modifying files, running Bash commands, searching code, calling tools, using MCP, executing slash commands, and streaming output.

Best way to use it

OpenClaude is best suited for developers who are already used to terminal-based development and want more flexibility when using different models.

Suitable scenarios

Using an AI Coding Agent in the terminal
Using different models to handle coding tasks
Letting AI understand and modify local codebases
Using AI to generate tests, documentation, and refactoring suggestions
Testing different backends such as OpenAI-compatible APIs, Gemini, Codex, and Ollama
Combining MCP, Bash, file tools, and code search
Building a more open AI coding workflow
Comparing how different models perform on coding tasks

More precisely, OpenClaude is suitable as an AI coding workspace for developers. It helps you use different models more flexibly for coding tasks, but it still requires developers to provide clear goals, review results, and control execution risks.

Installation Prompt

You can copy the following Prompt to Claude Code, Cursor, Codex, or other AI Coding Agents:

Please help me install and configure OpenClaude locally.

Requirements:
1. Check whether Node.js and npm are installed.
2. Install OpenClaude globally using the official command:
   npm install -g @gitlawb/openclaude@latest

3. Verify the installation:
   openclaude --version

4. Start OpenClaude:
   openclaude

5. Help me configure one model provider.
   Prefer OpenAI-compatible setup first.

6. Use environment variables for API keys.
   Do not hardcode or print any API key.

7. If I want to use a local model, help me check whether Ollama is installed.
   If Ollama is available, configure OpenClaude with:
   OPENAI_BASE_URL=http://localhost:11434/v1
   OPENAI_MODEL=qwen2.5-coder:7b

8. Create a small test task:
   - Open a simple local project
   - Ask OpenClaude to explain the project structure
   - Ask it to suggest one small improvement
   - Do not allow it to modify files until I confirm

9. After the test, explain:
   - Which provider is currently active
   - Where the configuration is stored
   - How to switch providers later
   - How to safely review file changes

Prompt Optimizer

Website: always200.com

GitHub: https://github.com/linshenkx/prompt-optimizer

GitHub Stars: 30.3k

First open-sourced: February 2025

License: Custom license

Project introduction

Prompt Optimizer is an open source prompt optimization tool that helps users write prompts that are clearer, more stable, and easier to reuse.

Best way to use it

Prompt Optimizer is best suited for AI use cases that are high-frequency, reusable, and have requirements for output quality. It is especially useful for content creation, AI application development, customer service replies, coding assistance, image generation, data analysis explanations, knowledge base Q&A, and similar scenarios.

Installation Prompt

If you just want to try it quickly, you can use the online version directly.

If you want to deploy it locally or use it privately, you can copy the following Prompt to Claude Code, Cursor, Codex, or other AI Coding Agents:

Please help me deploy Prompt Optimizer locally.

Requirements:
1. Open the official GitHub repository:
   https://github.com/linshenkx/prompt-optimizer

2. Read the official README and choose the simplest local deployment method.

3. Prefer Docker deployment if Docker is installed:
   docker run -d -p 8081:80 --restart unless-stopped --name prompt-optimizer linshen/prompt-optimizer

4. If Docker is not installed, help me check whether Node.js and pnpm are available, then guide me through local development setup.

5. After deployment, open the web interface at:
   http://localhost:8081

6. Help me configure at least one AI model provider, such as OpenAI, Gemini, DeepSeek, or a custom OpenAI-compatible API.

7. Do not print or hardcode any API key in the terminal or source code.

8. Explain where the configuration is stored and how to update or remove API keys later.

9. Create a simple test:
   - Input a short prompt: "Help me write a product update email"
   - Optimize it
   - Compare the original and optimized prompt
   - Explain why the optimized version is better or worse

10. Make sure the final setup can be reused later.

CC Switch

Website: https://ccswitch.io

GitHub: https://github.com/farion1231/cc-switch

GitHub Stars: 90k

First open-sourced: August 2025

License: MIT

Project introduction

CC Switch is a cross-platform desktop tool for centrally managing model providers, API Keys, MCP, prompts, and Skills configurations across multiple AI Coding / AI CLI tools.

It supports tools including Claude Code, Claude Desktop, Codex, Gemini CLI, OpenCode, OpenClaw, and Hermes.

CC Switch centralizes scattered configurations into one desktop app, allowing developers to manage and switch them through a graphical interface instead of repeatedly editing JSON, TOML, or .env files by hand.

Best way to use it

CC Switch is best suited for heavy users of AI Coding tools, especially developers who already use multiple CLI tools, multiple model providers, or multiple API Keys at the same time.

CC Switch is suitable for these scenarios:

Using multiple AI Coding tools at the same time
Frequently switching between tools such as Claude Code, Codex, Gemini CLI, and OpenCode
Managing multiple model providers and API Keys
Reducing the risk of manually editing configuration files
Centrally managing MCP, prompts, and Skills
Syncing AI tool configurations across multiple devices
Managing AI Coding workflows through a desktop interface

Installation Prompt

You can copy the following Prompt directly to Claude Code, Cursor, Codex, or other AI Coding Agents:

Please help me install and configure CC Switch on my local machine.

Requirements:
1. Detect my operating system: macOS, Windows, or Linux.
2. Follow the official installation method from the CC Switch GitHub repository.
3. If I am on macOS, prefer installing via Homebrew:
   brew install --cask cc-switch
4. If I am on Windows or Linux, guide me to download the correct installer from GitHub Releases.
5. After installation, help me launch CC Switch.
6. Check whether my existing AI CLI tools are installed, such as Claude Code, Codex, Gemini CLI, OpenCode, or OpenClaw.
7. Help me import the existing configuration if available.
8. Create one test Provider configuration.
9. Explain where CC Switch stores its local data and backups.
10. Do not expose or print any API key in the terminal output.

Official GitHub repository:
farion1231/cc-switch

Official website:
ccswitch.io

browser-use

Official website: https://browser-use.com/
GitHub: https://github.com/browser-use/browser-use
GitHub Stars: 96.8k
First open-sourced: November 2024
License: MIT

Project introduction

browser-use is an open source tool that lets AI Agents operate browsers.

Most websites were originally designed for humans, not for AI Agents. If AI wants to complete web tasks, it usually needs to understand page content, identify buttons and input fields, click elements, fill out forms, navigate between pages, and even handle multi-step workflows.

browser-use can turn the browser into an execution environment for AI Agents, allowing AI to complete tasks on webpages like a human.

Best way to use it

browser-use works best as part of an AI Agent toolchain, rather than as a complete standalone product.

A better way to use it is:

Define a very specific web task first

For example: log in to a backend system, find orders from the past 7 days, and export the results.

Let the AI Agent use browser-use to operate the web

browser-use handles opening pages, identifying elements, clicking buttons, filling in content, and reading webpage states.

Add restrictions and checks for key workflows

For example, do not let AI submit payments, delete data, or modify production environment configurations without control. For high-risk actions, it is best to add human confirmation.

Turn stable scenarios into automated workflows

For highly repetitive web tasks, successful workflows can be saved and then gradually optimized.

Prioritize browser stability in production environments

If a task involves high concurrency, proxies, CAPTCHAs, browser fingerprints, or login state management, you should consider not only local execution but also cloud browsers or hosted environments.

browser-use is suitable for these scenarios:

AI Agent browser automation
Web form filling
Automated operation of backend systems
Data collection and information queries
Simulating human web operations
Testing AI Agents’ ability to perform tasks on the web
Adding browser action capabilities to existing Agents

Installation Prompt

You can copy the following Prompt directly to Claude Code, Cursor, Codex, or other AI Coding Agents:

Please help me install and run browser-use locally.

Requirements:
1. Check whether Python 3.11 or above is installed.
2. Use uv to create a clean Python environment.
3. Install browser-use.
4. Install Chromium if it is not available.
5. Create a simple example script that uses browser-use to open a webpage and complete a basic task.
6. Use an environment variable for the LLM API key. Do not hardcode the API key in the script.
7. Add clear comments so I can understand each step.
8. After installation, run the example and help me verify whether browser-use works correctly.

Official GitHub repository:
browser-use/browser-use

Please follow the official quickstart as much as possible.

If you want to test a simple task directly, you can change the task to:

Create a browser-use example that opens the browser-use GitHub repository, finds the current number of GitHub stars, and prints the result.

FAQ

Are these open source AI tools suitable for enterprise use?

It depends on the specific tool and use case.

Tools like browser-use and OpenCLI are better suited for Agent automation and external tool calling. NocoBase is better suited for building enterprise internal business systems.

What is the difference between browser-use and OpenCLI?

Both can enhance an AI Agent’s ability to operate webpages and external tools, but they do it in different ways.

browser-use is more like letting AI directly operate the browser. AI opens webpages, reads pages, clicks buttons, fills out forms, and decides the next step based on the page state.

OpenCLI focuses more on packaging websites, local tools, or browser sessions into CLI commands. This allows AI Agents to call tools through commands instead of starting from the webpage interface every time.

If I want to build enterprise internal systems, what type of AI tool should I choose?

If your goal is to build enterprise internal systems such as CRM, approval systems, ticketing systems, project management, and operations backends, AI Coding tools alone are usually not enough.

These systems often require data models, permission control, page configuration, workflows, private deployment, and ongoing maintenance. AI can help speed up system building, but the system itself still needs a stable business application foundation.

This is where platforms like NocoBase fit well. It does not simply let AI generate code. Instead, it lets AI build maintainable business systems based on data models, pages, permissions, and workflows.

Summary

In the past, we paid more attention to “what AI can generate.” But in real workflows, the more important questions are: Can AI operate real tools? Can it call external systems? Can it enter business workflows? Can the generated result continue to be maintained?

This is exactly the value of these open source AI tools.

Related reading

DEV Community

Top 6 Open Source AI Tools by GitHub Stars for Stronger AI Agents

Introduction

How did we select these tools?

NocoBase

Project introduction

What problem does it solve?

Best way to use it

Suitable scenarios

Usage boundaries

Installation Prompt

OpenCLI

Project introduction

Suitable scenarios

Installation Prompt

OpenClaude

Project introduction

Best way to use it

Installation Prompt

Prompt Optimizer

Project introduction

Best way to use it

Installation Prompt

CC Switch

Project introduction

Best way to use it

Installation Prompt

browser-use

Project introduction

Best way to use it

Installation Prompt

FAQ

Summary

Top comments (0)