Introducing Theia Coder - the open AI coding agent with full control

March 6, 2025 | 9 min Read

Meet Theia Coder, an open, flexible, and transparent AI coding assistant designed to seamlessly integrate into your development workflow. It can generate and apply code and arbitrary file changes based on natural language requests. Whether you’re generating new projects from scratch or refining an existing complex codebase, Coder streamlines the process while keeping you in full control.

Unlike proprietary AI-powered coding tools, Coder operates with full transparency, allowing you to see exactly what it’s doing, review changes before applying them, and customize its behavior to fit your needs. Built into the AI-powered Theia IDE and based on Theia AI, Coder provides an adaptive and fully open approach to AI-assisted coding – you can tailor its functionality to match your preferred workflows, LLMs, and project requirements.

Let’s dive in and explore what Coder can do!

If you’re unfamiliar with Theia AI or the AI-powered Theia IDE, check out the Theia AI introduction and the AI Theia IDE overview, download the AI-powered Theia IDE and check out the Theia Coder documentation.

Let’s start generating Code with Theia Coder!

Instead of diving into technical details, let’s just explore a few examples to demonstrate how Theia Coder works and what it can do for you. You can try Theia Coder yourself by downloading the AI-powered Theia IDE! Check out  the Theia Coder documentation to get yourself set up to start using Theia Coder today.

As a first example, let’s do something simple from scratch. In the following video, we show how to create a quiz game as a node-based web application with Coder. The prompt is very simple:

@Coder Create a quiz game as a node-based web application. The quiz should be about AI and LLMs.

As you can see in the following video, Coder will quickly summarize the plan and then directly start to create the necessary files. Depending on the speed of the underlying LLMs, this might take a while. All generated files are added to a so-called change set first. As a user, you can choose to review them in a diff editor, apply them one by one, or apply all of them at once when the generation is complete.

But let’s look at the actual result that was generated, which looks pretty nice! Note, that this was fully generated, so the LLM even created the content of the questions for us!

Especially frontier LLMs are surprisingly good at generating entire applications up until a certain size from scratch. As Coder can be connected to any LLM of your choice, you can benefit from the power of those frontier models in Coder with the huge advantage that Coder can create those files directly in the IDE at the right place in a transparent process for you.

However, most development tasks are actually not greenfield engineering, they are about changing existing code. And this is where the advantage of an IDE-integrated AI assistant such as Coder becomes even more apparent. So let’s do this with Theia Coder!

In the following video, we show how we add Coder to change something in the created code base using the following prompt:

@Coder Add the option to skip a question

Again, you can choose to review the proposed file changes (as done in the video) or just wait until it’s done and accept them all. As you can see, this time, Coder looks at the existing files it intends to change to get their latest state before applying the necessary changes. Coder makes these operations fully transparent to the user, you can always see what Coder looks at, and which files it wants to modify.

The result of the modified quiz game now includes a fully functional “Skip” button, as shown below.

With this example, we can easily see that Coder is able to incrementally evolve an existing (small) code base. This also works for larger change requests. We could for example ask it:

I wanted to use React

And with the proper LLM under the hood, it would very likely refactor the entire existing application correctly for us.

However, most development tasks are not about greenfield engineering but rather about modifying existing code in larger codebases! So let’s do this with Coder! 🙂

In the next example, we apply a functional change to the Theia IDE (the IDE hosting the AI assistant we are talking about).

@Coder Allow the user to start already started MCP servers again via the command in the UI

It is not important to understand the purpose of this change. What’s more interesting in the context of this article, is how Coder performs on a very large code base with very little information. As you can see in the following video, it starts to browse the code base, identifies the correct file and applies the change correctly.

The great thing about Coder here is the transparency. By optionally expanding the function calls in the chat (as shown in the video), the user can exactly see where Coder is looking for the relevant files. This is not just great for the sake  of transparency. It helps developers to understand the basis on which the LLM created its answer, and optimize or rephrase the prompt if needed to improve efficiency of the AI assistant.

If you are familiar with the code base and or if you experience that the LLM is not capable of finding the right files, you can very easily provide “context” explicitly to the query. In the following video, we do the exact same task, but this time, we provide Coder with the file to modify right away. As you can see, the process is much faster now and in practice, this approach is also significantly more reliable. The Theia IDE provides various options to add context using variables with auto completion, referring to opened files or using drag and drop, see the AI chat documentation for more details and this article about the context framework in Theia AI to learn how it works internally, and how you can easily adapt it.

So the key takeaway is: Coder is capable of gathering context on its own, but the more relevant context you provide to it, the faster and better the results will be. So you should develop a habit of providing everything you know, just as you would do with a human co-worker.

Learn more about how to use Theia Coder in our documentation!

Theia Coder in the AI-powered Theia IDE, based on Theia AI

Theia Coder is not just an extension. It is embedded into the flexible and fully open AI-powered Theia IDE. This enriches Theia Coder with many additional features available in this AI-powered tool ecosystem, including:

  • Prompt editing and Custom agents: The prompt of Coder is not only open source, you can see and modify it as a user! Thereby, you can optimize it to your liking, to your project or to the LLM of your choice. And as the Theia IDE supports creating custom agents on the fly, you can even create custom versions of coder for specialized purposes (e.g. coding, documentation, writing tests, etc.) or other agents, such as for automating Git workflows.

  • MCP Integration: Using the Model Context Protocol, you can connect agents such as Coder to external tools and services, such as Git, Github, Playwright, and many more. Read also our article on what you can do with MCP in the Theia IDE.

  • SCANOSS Integration: As you can also see in the video above, all generated content can be automatically scanned via SCANOSS for matches in open-source databases and providing detailed insights if the code appears to be copied. We will provide more detail about this feature in a dedicated article soon, see also our documentation.

  • More agents: the AI-powered Theia IDE hosts multiple agents, e.g. an “architect”, a terminal assistant and of course AI-driven code completion

  • Freedom of LLM selection: You can use any LLM including well-known candidates such as Anthropic and OpenAI, but also self-hosted and local LLMs, see “Why Theia supports any LLM!

Theia Coder and the AI-powered Theia IDE are both built on Theia AI, a powerful open framework that enables tool builders to seamlessly integrate AI into their own tools, whether it is a specialized IDE or a domain-specific tool. Theia AI provides full control over every aspect of AI-powered interactions, this flexibility allows you to adapt literally anything about Theia Coder and the Theia IDE. Whether you want to fine-tune how AI generates code, modify the user experience, or introduce entirely new AI-driven workflows, Theia AI provides the foundation to do so. Moreover, Theia AI isn’t limited to just coding—its open architecture allows tool builders to develop AI-native IDEs without the need to fork existing solutions, as well as AI-powered tools for non-code-centric domains, such as engineering, automation, or research.

If you’re a tool builder looking to create the next generation of AI-driven development environments, Theia AI provides the ideal foundation. With its open framework and modular design, it empowers you to push the boundaries of AI-assisted tooling while keeping full control over your solution.

Want to learn more? Discover Theia AI—the open framework for AI integration on your terms!

Conclusion

Theia Coder demonstrates that a well-designed AI coding assistant can be really powerful, but also fully transparent, flexible, and adaptable to different workflows and project requirements at the same time. Whether generating code from scratch, modifying existing codebases, or optimizing large-scale projects, Coder ensures that developers remain in full control of the process. Unlike proprietary alternatives, it operates with complete transparency for users, even allowing you to refine prompts, review file changes, and choose your preferred LLMs.

By integrating seamlessly into the AI-powered Theia IDE and leveraging Theia AI’s open architecture, Theia Coder not only enhances productivity but also sets a new tone for AI-assisted development. Whether you’re a developer looking for a smarter way to automate repetitive tasks or a tool builder seeking to create advanced AI-powered IDEs, Theia AI provides the foundation to make it happen—on your terms.

🚀 Ready to try it yourself? Download the AI-powered Theia IDE and start coding with Theia Coder today!

If you are interested in building custom AI-powered Tools, EclipseSource provides consulting and implementation services backed by our extensive experience with successful AI tool projects. We also specialize in web- and cloud-based tools and support for popular platforms like Eclipse Theia and VS Code.

👉 Get in touch with us to learn more about how we can help you build custom AI-powered tools.

Jonas, Maximilian & Philip

Jonas Helming, Maximilian Koegel and Philip Langer co-lead EclipseSource, specializing in consulting and engineering innovative, customized tools and IDEs, with a strong …