llm-workbench: A Local-First Multi-LLM Workbench for Developers

llm-workbench is a free tool built for developers who want to test and compare different artificial intelligence models without sending their data to the cloud. It runs entirely in your web browser, which means your API keys and private documents stay safe on your own computer. The tool lets you write prompts with special variables, run them against up to four different models at the same time, and see the results side by side with helpful charts and cost estimates.

Benefits

The main advantage of llm-workbench is its focus on privacy and control. Unlike many online tools that require you to upload your data to a server, this application keeps everything local. It uses strong encryption to store your API keys in a secure vault inside your browser. This ensures that sensitive information like passwords and keys never leave your device unless you choose to send them to a specific provider.

Another key benefit is the ability to compare models instantly. You can select up to four different providers, such as OpenAI, Google Gemini, or local models running on your machine, and run your prompt on all of them at once. The tool then displays the responses in a clear grid so you can see the differences in quality and style immediately. It also tracks important performance metrics like how fast the model responds and how much it would cost to use.

The tool also helps developers save time by generating code snippets automatically. After you design a prompt, the workbench can create ready-to-use code in languages like TypeScript or PHP. This code is designed to be safe because it does not include your API keys directly. Instead, it uses environment variables, which is a best practice for secure software development.

Use Cases

Developers can use llm-workbench to test new prompts before deploying them to a live application. For example, a developer might want to see how a customer service bot answers questions differently when powered by two different models. They can type their question into the tool, run it against both models, and choose the one that provides the best answer.

The tool is also useful for teams that need to evaluate multiple options quickly. By running parallel tests, a team can compare the speed and cost of different models without waiting for each request to finish one by one. The visual dashboards help them see which model is the fastest and which one might be too expensive for their budget.

Users who work with private documents can use the built-in RAG feature to upload files like PDFs or text documents. The tool can read these files and use the information to answer questions. This is helpful for building internal tools where the AI needs to know about specific company data that should not be shared with public models.

Pricing

llm-workbench is completely free to use. The software itself is source-available, meaning the code is open for anyone to inspect or modify. While the tool is free, you will need to provide your own API keys for cloud providers like OpenAI or Anthropic if you want to use their services. If you choose to run local models using Ollama, you do not need any paid API keys. The tool does not charge any fees for its own usage or for hosting.

Vibes

The community response to llm-workbench has been positive, especially among developers who value privacy and open-source tools. Users appreciate the clean interface and the ability to run complex comparisons without setting up complicated servers. The emphasis on security, such as the encrypted vault for API keys, has been highlighted as a major strength. Reviewers note that the tool makes it easy to experiment with different models without worrying about data leaks. The mobile-friendly design also receives praise for allowing developers to check their work on smaller screens.

Additional Information

The project is built using modern web technologies like Nuxt 4 and Vue 3. It is designed to be flexible and can be deployed in several ways. Developers can host it on GitHub Pages for a simple static site or use platforms like Vercel for full server support. The code is available under a source-available license, which encourages others to contribute to the project. The development team welcomes new contributors who want to help improve the tool or add new features. The project also includes a roadmap and documentation to guide users and developers who want to get involved.