Eclipse Theia 1.68 Release: News and Noteworthy

We are happy to announce the Eclipse Theia 1.68 release! The release contains in total 75 merged pull requests. In this article, we will highlight some selected improvements and provide an overview of the latest news around Theia.

For those new to Eclipse Theia, it is the next-generation platform for building IDEs and tools for the web or desktop, based on modern state-of-the-art web technologies. With Theia AI, part of the Theia platform, you can also build AI-powered tools and IDEs with ease. For more details about Eclipse Theia, please refer to this article and visit the Theia website.

Important note for adopters: Theia 1.68 is the release candidate for the upcoming 2026-02 community release. We encourage all adopters to verify this release early and report any issues to ensure a smooth community release. Please test your Theia-based applications with 1.68 and provide feedback via GitHub issues.

The Theia project also releases a product, the Theia IDE. The Eclipse Theia IDE is a modern, AI-native, and open IDE for cloud and desktop environments, aimed at end users. The Theia IDE is based on the Theia platform and also includes advanced AI powered features. For more details, see the Theia IDE website.

If you are looking for a simple way to check out the new release, please download and install the Theia IDE, which is based on Theia 1.68.

Eclipse Theia 1.68: Selected features and improvements

In the following, we will highlight some selected improvements in the new release. As usual, we cannot mention all 75 improvements, however we will focus on the most notable changes as well as changes visible to end users. The corresponding pull requests are linked under the respective heading when applicable.

GitHub Copilot Language Model Integration

Users with an existing GitHub Copilot subscription can now leverage their investment directly within the Theia IDE and any tool built on Theia AI—no additional API keys or subscriptions required. Simply authenticate with GitHub, and Copilot models including GPT-4o and Claude 3.5 Sonnet become available for all AI features, from chat assistance to code generation and beyond.

This feature is available both for end users of the Theia IDE and for adopters building custom tools with Theia AI, making it easy to offer GitHub Copilot integration to your users out of the box.

Technically, Theia 1.68 introduces a new @theia/ai-copilot package that integrates GitHub Copilot as a language model provider in Theia’s AI framework. The integration features complete OAuth device flow authentication with GitHub, making the sign-in process straightforward. A status bar indicator shows the current authentication state, and dedicated commands for login/logout and model management are provided. Once authenticated, Copilot models become available in the AI Configuration view and can be assigned to any AI agent just like other language model providers.

Mode Selection for Agents

Agents now feature mode selection directly in the chat input UI, allowing users to switch between different operational modes without diving into settings. The Coder Agent offers Edit Mode, Agent Mode, and Agent Mode (Next), while the Architect Agent provides Default Mode, Simple Mode, and Plan Mode. See the following sections for details on the new modes of these two agents.

When a mode is selected, it overrides the settings-configured prompt variant for that session. The mode selector initializes to reflect the current default settings, and users can cycle through modes using the Shift+Tab keyboard shortcut. This makes it much more accessible to use agent behaviors without navigating through configuration screens.

Mode selection is available for all agents built with Theia AI if the agent designer chooses to expose this feature to the end user.

Enhanced Agent Mode (Next) for Theia Coder

The new Agent Mode (“Next”) of Theia Coder prompt template represents a significant evolution in how Theia Coder in agent mode operates autonomously. This refined prompt in combination with several other refinements, e.g. on tool functions, make the agent more effective, reliable, and autonomous in its approach to coding tasks.

To benefit from the new “Agent Mode (Next)”, users currently need to switch to it either globally in the Agent configuration or for a session using the new mode selector (see previous Mode Selection above). We recommend all users try the new “Next” mode and provide feedback to help us improve it further.

The new mode also integrates with the todo tool (see below), allowing the agent to create and update visual task lists as it works through complex multi-step implementations. This provides users with clear visibility into the agent’s progress and plan.

Improved Architect Planning Mode

The Architect agent’s planning capabilities have been significantly enhanced with new task context functions that allow direct creation and management of implementation plans. The agent can now create, read, edit, and list task contexts (implementation plans) without requiring separate LLM summarization calls.

Again, to benefit from the improved Architect Planning Mode, users currently need to switch to it either globally in the Agent configuration or for a session using the new mode selector. We recommend all users try the new planning mode and provide feedback to help us improve it further.

Similar to the Coder improvements, these enhancements are also interesting for Theia AI adopters who want to leverage or customize the planning capabilities in their own tools. New functions include createTaskContext for creating plans with auto-generated metadata, getTaskContext for reading the current plan, editTaskContext for modifying plans using string replacement, and listTaskContexts for listing all plans in a session. Multiple plans per session are supported, with each plan getting its own “Execute with Coder” action in the UI.

AppTester Improvements and Automated Testing Workflow

The AppTester agent receives significant enhancements in Theia 1.68, introducing a powerful new variant that uses the DevTools MCP server instead of Playwright for testing interactions. This architectural change provides more robust and flexible testing capabilities directly within the IDE environment.

To benefit from the new AppTester variant, users currently need to switch to the “next” variant globally in the Agent configuration. We recommend all users try the new AppTester mode and provide feedback to help us improve it further.

Furthermore, a new /with-apptester slash command has been added that enables an automated implement-and-test workflow. When using Coder in agent mode, you can now include this command in your request (e.g., @Coder <feature description> /with-apptester), and after implementing the requested changes, Coder will automatically delegate to AppTester to verify the implementation. If tests don’t pass, the agents can iterate to fix issues—creating a complete automated development loop.

Todo Tool for AI Task Tracking

A new todo_write tool enables AI agents to track multi-step task progress visually. Agents can create and update todo lists that render directly in the chat interface with status icons for pending, in-progress, and completed items. Previous todo lists are automatically hidden when updated.

This tool is used by the new Agent Code Next mode (see section above) and is available to all agents built with Theia AI, making it easy for adopters to integrate visual task tracking into their custom tools.

Shell Execution Tool for AI Agents (alpha)

A powerful new shellExecute tool has been added in Theia AI that enables AI agents to run shell commands on the host system with proper safety measures. This capability opens up significant possibilities for autonomous agent workflows while maintaining user control through a comprehensive confirmation system.

The tool includes configurable timeouts (default 2 minutes, maximum 10 minutes), working directory support, and intelligent output truncation to prevent context overflow. When an agent requests to execute a shell command, users see a detailed confirmation dialog showing the command and can choose to allow once, allow for the session, always allow, or deny with an optional reason that gets passed back to the agent.

For security, the tool defaults to requiring confirmation even if the global tool confirmation setting is “Always Allow”, with a warning dialog when users choose to auto-approve shell commands.

The shell execution tool is not yet added by default to any agent. We plan to optionally integrate it into Theia Coder. In the meantime, you can either add it to a user request (as shown in the demo above using ~shellExecute) or make it part of an agent by modifying its prompt and adding ~{shellExecute}.

Agent Skills Support (alpha)

Theia 1.68 introduces Agent Skills—a specification allowing AI agents to access reusable instructions and domain knowledge from SKILL.md files. Skills are directories containing a SKILL.md file with YAML frontmatter (defining name and description) and markdown content that provides specialized instructions or knowledge for agents.

The system automatically discovers skills from configured directories and the ~/.theia/skills/ location. You can use specific skills in a request using a slash command /skillName.

Alternatively you can allow your agent to load skills on demand. To enable this the {{skills}} variable can be injected into agent prompts, a lists of all available skills for the agent to leverage. The ~getSkillFileContent~ function allows an agent to load a selected skill on demand. As before, we are working on exposing this skill capability for agents by default in future release.

CreateSkill Agent (alpha)

The new CreateSkill agent enables users to create new project-specific skills directly through the AI chat interface. Simply describe the skill you want to create, and the agent will generate a properly structured SKILL.md file with the appropriate frontmatter and content.

Note that to use the CreateSkill agent, you currently need to add .prompts/skills to the list of skill directories in your configuration. This allows the agent to save newly created skills to your project’s local skills directory.

Debug Console Enhancements

Debugging workflows benefit from two significant improvements to the debug console. First, console output is now preserved when a debug session ends, allowing developers to review logs from terminated sessions and switch between active and stopped sessions using the session selector. This addresses a common pain point where valuable debugging information was lost when sessions terminated.

Second, a text filter has been added to the debug console toolbar, supporting both include and exclude patterns. Users can type terms like error, warning to show only matching lines, use !debug to hide lines containing specific terms, or combine them with patterns like info, !verbose. The filter state is preserved per session, and hover tooltips explain the syntax for easy discovery.

Additional improvements include hover tooltips explaining the severity selector, session selector, and filter syntax, plus accessibility enhancements with aria-labels on interactive elements. Sessions are sorted with active ones appearing first and terminated sessions appearing last in the dropdown.

Chat and AI Infrastructure Improvements

Beyond the headline features, Theia 1.68 delivers numerous improvements to the AI chat infrastructure. Chat session storage now defaults to workspace scope rather than global storage, keeping project-specific conversations with their associated projects. A new preference allows configuring whether sessions are stored per-workspace or globally, with configurable paths for both options.

Chat accessibility has been significantly enhanced with focus navigation commands (Ctrl/Cmd+Up/Down to move between input and responses), comprehensive ARIA attributes for screen readers, and keyboard accessibility for all chat buttons.

Tool function references (~functionName) in chat messages are now displayed with badge styling matching agents and variables, with hover tooltips showing tool descriptions. The tool handler context types have been refactored for better type safety, properly distinguishing between chat-dependent and chat-independent tools.

Notifications can now be copied via a context menu action, and error chat responses properly support the copy message feature. The MCP SDK has been updated to v1.25.1 with migration to Zod v4, and the HuggingFace SDK to v4.x for the new API endpoint.

Additional Improvements

As always, the 1.68 release contains much more than described in this article. The compatibility with VS Code Extensions has been upgraded to API version 1.108.0, including QuickPick improvements with new prompt and resourceUri properties, and ThemeIcon support for WebviewPanel icons.

The Source Control view now displays action buttons contributed by SCM providers like the built-in Git extension, showing a prominent “Commit” button with a dropdown for secondary commands that automatically updates based on repository state.

Several merge editor bugs have been fixed, including issues with manual resolution detection, conflict handling, “Mark as Handled” functionality, and “Go to Previous Unhandled Conflict” navigation.

The Terminal Manager has received fixes for preference handling and element deletion and active page handling on terminal page deletion. Memory leaks from disposable collections have been addressed, and Monaco editors are now properly connected to their models only when visible.

The settings UI now displays preference tags (like “experimental”) as styled badges, and properly indicates when values are explicitly set even if they match the default. Output widget state restoration has been fixed, and theme and color definitions have been updated to align with VS Code definitions.

Default LLM models have been updated to include GPT-5.2, GPT-5.2 Pro, and Gemini Flash 3, with GPT-5.2 becoming the default in model aliases.

For a complete overview of all changes in this release, please refer to the 1.68 milestone. All these features and improvements (in total 75) were the result of one month of intensive development. Eclipse Theia follows a monthly release schedule. We are looking forward to the next release due next month, stay tuned! To be notified about future releases, follow us on LinkedIn or follow Theia on Twitter and subscribe to our mailing list.

If you are interested in building custom tools or IDEs based on Eclipse Theia, EclipseSource provides consulting and implementation services for Eclipse Theia, for AI-powered tools, as well as for web-based tools in general.

Furthermore, if you want to extend Theia, Theia AI or the Theia IDE with new features or strategically invest into the project, EclipseSource provides sponsored development for Theia, too. Finally, we provide consulting and support for hosting web-based tools in the cloud.

👉 Get in contact with us, to discuss your use case!

👉 Subscribe to our YouTube channel