Theia Coder in Action

With Theia IDE 1.70, the Theia Coder agent takes another significant step forward. This release introduces Capabilities — a new concept that lets you enable agent features like E2E testing, shell access, and GitHub integration with a single click. Beyond that, there are many improvements to the basic development flow — the agent mode has been improved significantly, and the UX for plan-driven development has been refined . In our latest demo, we walk through a complete development workflow — from code generation, to plan-driven development, to automated E2E testing.

Watch the Demo

Code Generation and Iterating on Existing Code

The video starts with the Coder agent generating a full app from a single prompt. The agent mode has been improved significantly — as shown in the video, it specifically handles complex, multi-file changes even more reliably and efficiently. More interesting is what comes next: adding a new feature to the existing codebase, where the agent analyzes the existing code and applies targeted changes.

Plan First, Code Second

For larger tasks, jumping straight into code generation is often not ideal. This is where the plan mode comes in — initially introduced almost a year ago and continuously improved since then. In this release, specifically the UX has been refined further to ease the interaction with created plans. Plan mode creates a structured implementation plan as a Markdown file — reviewable, editable, and persistent in the workspace. You can iterate on this plan using AI (which keeps everything consistent) or edit the file directly. Once you’re satisfied, a single click starts a fresh coding session with the plan as context. This session splitting is intentional: it keeps the planning context separate from implementation, avoiding context window bloat and auto-compaction.

One-Click Capabilities: E2E Testing, Shell, GitHub

The most notable addition in this release is Capabilities — high-level features you can toggle with a single click. Instead of manually configuring MCP servers, sub-agents, or prompt fragments, you simply enable what you need. In the demo, we activate E2E app testing. The Coder agent then delegates to a specialized testing sub-agent that controls the running application via an MCP server — verifying UI behavior like division by zero handling. Other available capabilities include shell execution and GitHub access, for example to report bugs directly from the agent chat.

Flexible LLM Configuration

Theia Coder works with any major LLM provider — Anthropic, OpenAI, Google Gemini, or self-hosted models via Ollama. We also recently added support for GitHub Copilot subscriptions. A model alias system lets you assign models to agents flexibly, or you can just use the defaults.

Try It Yourself

🧠 Download the AI-powered Theia IDE

🧰 Learn more about Theia AI

📺 Watch the full YouTube demo

Build Your Own AI-Native Tools

Everything shown in this video is open source. Theia Coder is built on Theia AI, a framework for building AI-native tools — not just for coding, but for diagrams, configurations, or any domain-specific workflow. We have many adopters building tools in domains far beyond coding.

At EclipseSource, we help teams design and implement AI-native tools and IDEs.

👉 Services for Building AI-Enhanced Tools and IDEs

👉 Contact us to discuss your project

💼 Follow us: EclipseSource on LinkedIn

🎥 Subscribe to our YouTube channel: EclipseSource on YouTube