Eclipse Theia 1.70 Release: News and Noteworthy

We are happy to announce the Eclipse Theia 1.70 release! This is a landmark release: AI Coding features in the Theia IDE officially graduate from beta, reflecting over a year of usage, continuous evolution, and growing adoption. This follows the underyling framework, Theia AI, graduating from beta about a year ago. 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.

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 1.70.

Eclipse Theia 1.70: 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.

AI Coding Features in the Theia IDE Graduate from Beta

With Theia 1.70, all AI Coding features integrated in the Theia IDE officially leave beta. This includes IDE-specific AI Views, the Theia Coder agent and related agents. Some agents and features, especially new ones, remain marked as experimental or alpha while its UX is still being refined.

It is worth providing some context on what this means: The underlying framework for building AI-native tools, Theia AI, already graduated from beta about a year ago and has since been adopted by tool builders who reuse some of Theia’s built-in agents as-is while customizing others for their specific use cases. The AI features built on top of this framework for the Theia IDE, e.g. Theia Coder have been used by many users for well over a year.

Reflecting this milestone, the default workbench layout now includes the AI Chat (and Terminal) panels open by default for new workspaces, making AI features a first-class, always-visible part of the IDE.

At the same time, ever since AI entered the tools and IDE space, innovation has not stood still for a single week. New models, new capabilities, technologies such as MCP and Skills, new patterns like agent mode and plan mode, the landscape continously evolves. A tool like Theia Coder and AI coding support in general will never be “complete”, at least not for the foreseeable future. These features are actively used, and continuously improving based on community contributions.

Theia Coder Improvements

Theia 1.70 ships two meaningful updates to Theia Coder.

Agent Mode as Default with First-Use Confirmation: The Coder agent now defaults to Agent Mode for all new users (instead of Edit Mode), delivering the full agentic coding experience out of the box. This reflects the reality that the vast majority of Theia Coder users already work in Agent Mode. The very first time a request is made, a confirmation dialog appears in the chat explaining that Agent Mode can directly write to workspace files. Users can then decide whether to continue with Agent Mode or use Edit Mode instead. If they choose Edit Mode, that choice is persisted as the default mode. The default mode can also be changed at any time in the AI Configuration view, and users can still switch modes for the current session via the mode selector in the chat input area, as shown in the video below. For adopters building on Theia AI, the confirmation logic is extracted into a dedicated AgentModeConfirmationService interface, making it straightforward to customize or skip the flow entirely.

Improved Plan Execution: When the Architect agent creates a task context and delegates to the Coder agent, the Coder now follows the plan steps more closely and performs less redundant codebase explorations, resulting in faster and more focused implementations. Adopters who use the Architect/Coder agent pair in their own tools benefit from this directly.

Theia AI Improvements

Theia AI is the framework within the Theia platform for building AI-powered tools and IDEs. The Theia IDE and Theia Coder are built on top of Theia AI, but adopters can also use it to build their own domain-specific AI solutions. Theia 1.70 brings several improvements to Theia AI that benefit both thei IDE end users and adopters of Theia AI.

GitHub Copilot: Auto-Discovery and OAuth Migration: The GitHub Copilot integration now supports automatic model discovery and a migration to OAuth App-based authentication.Available models are now fetched dynamically from the Copilot API and deduplicated by model family. Models are re-discovered automatically when the authentication state or enterprise URL changes. For users who need specific control, model overrides via the ai-features.copilot.modelOverrides preference remain fully functional. The authentication has been migrated from the previous GitHub App to a new OAuth App (“Theia Copilot OAuth Access”). Tokens from the old app are detected and cleared automatically, with a notification prompting users to sign in again. A new ai-features.copilot.enabled preference gates all Copilot UI elements and model registration, and the ai-features.copilot.enterpriseUrl preference is now fully wired up for both API requests and model discovery.

AI Proxy Support: All AI providers (OpenAI, Anthropic, Ollama, HuggingFace) now respect proxy settings including http_proxy, HTTP_PROXY, no_proxy environment variables, and the http.proxy preference. This resolves connection errors in corporate environments with HTTP proxies.

Undo/Redo Preservation for AI Suggestions: Applying AI code suggestions now preserves the editor’s undo/redo history by using minimal edits instead of replacing the entire document content.

AI Agent Streaming Improvements: The getFileContent tool now supports streaming for paginated reads, allowing AI agents to efficiently analyze very large files (300 MB+) without loading the full content into memory.

Token usage for Gemini models has been corrected to avoid duplicate counting.

Persistent Capability Settings: Agent capability overrides and generic capability selections can now be saved to settings.json at user, folder, or workspace scope. The AI Configuration view shows editable capability toggles with a blue bar indicator for overridden values, a “Reset All to Defaults” button, and a “Save to settings” action in the chat capabilities panel. An unsaved-changes badge on the capabilities toggle keeps changes visible even when the panel is collapsed.

Improved Skill Slash Commands: Skill slash commands have been improved to load skills via tool calls instead of pasting content, keeping the user message clean and making it clear to the LLM what is user input versus skill content.

Base Technology Updates

Monaco Editor Uplift to 1.108: Theia 1.70 includes a uplift of the underlying Monaco editor to version 1.108, bringing upstream improvements to the core editing experience, including improved inline completions and various rendering and accessibility enhancements.

Node.js 24 Support and Electron 39.7: Theia 1.70 adds support for Node.js 24, drops Node.js 20, and updates Electron to version 39.7. CI workflows now run on Node.js 24 by default, ensuring the platform stays current with the latest LTS and active Node.js releases.

VS Code API and Built-in Extensions: VS Code API compatibility has been upgraded to version 1.110.1, and VS Code built-in extensions have been updated to 1.108.2, including eslint, jupyter and python.

Git & SCM Improvements

Multi-Root SCM Support with Repositories Widget: The Source Control view gains proper multi-root workspace support with a new SCM Repositories Widget. When working in a workspace that contains multiple git repositories — such as worktrees or nested repos — all repositories are now listed in a dedicated panel within the SCM view container, grouped by provider ID. The badge count in the SCM view aggregates changes across all repositories, and the view title has been updated to “Changes” for clarity.

Git blame annotations in the editor and status bar: Theia now supports git blame information as text decorations directly in the editor and in the status bar, making it easy to see who last changed the current line or file without leaving the current context.

Git blame information shown as text decorations in the editor and in the status bar.

As a next step, we plan to improve the multi-diff editor integration so opening blame can show the full commit, including multiple changed files when applicable, instead of only the current file.

Git actions and SCM tree selection in diff editors now work correctly, with proper originalResource context key support, correct resourceScheme handling, and the SCM view now supports viewsWelcome content — when no repository is present, extensions can contribute welcome content such as “Initialize Repository” buttons.

Deprecated @theia/git removed: the long-deprecated @theia/git extension code has been fully removed from the codebase. Git support is now entirely provided by the VS Code built-in git extension.

Terminal Command History via Shell Integration

Theia 1.70 introduces terminal command history via shell integration. The terminal now tracks individual commands along with their output, making it easy to review what happened in a session. This works out of the box with bash and zsh, as well as with task terminals across all platforms.

For users, terminal command history can be enabled via the experimental setting terminal.integrated.enableCommandHistory, and visual command block dividers can be enabled via the experimental setting terminal.integrated.enableCommandSeparator, giving a clear separation between commands in the terminal. For extension authors, two new events — onTerminalCommandStart and onTerminalPromptShow — allow hooking into the command lifecycle. Most importantly, this is a foundational improvement that enables smarter AI terminal interactions in future releases, as AI agents can now precisely retrieve the output of specific commands.

Terminal command block dividers between executed commands.

Search Improvements

Improved Fuzzy Search: Substring Matches First: Search and filtering throughout Theia has been meaningfully improved by prioritizing substring matches over scattered character-wise fuzzy matches. Previously, typing a search term could surface unexpected results where individual characters happened to match across a long string. Now, results where the search term appears as a contiguous substring are ranked higher, making the Keyboard Shortcuts editor, Settings editor, File Explorer filter, and Quick Open file picker (Ctrl+P) all behave more predictably. As a bonus, clearing the filter in the Keybindings Editor now properly re-sorts the table to its initial order instead of leaving results in the previous filter order.

Configurable Search Exclusion Patterns: Theia 1.70 introduces the search.exclude preference, matching VS Code’s setting for controlling which files and folders are excluded from search results. This works alongside the existing files.exclude preference, and both are aggregated through a new centralized WorkspaceSearchFilterService. A redesigned “Use Exclude Settings and Ignore Files” toggle in the search view (replacing the old “Include Ignored Files” button) makes it easy to temporarily bypass all exclusion patterns. The Quick Open file picker (Ctrl+P) also respects these patterns, ensuring a consistent experience across all search surfaces.

Debugging Improvements

Breakpoint Identity Refactoring: The debugger’s breakpoint management has been comprehensively refactored so that BreakpointManager is now the sole owner of all DebugBreakpoint instances, inspired by VS Code’s sessionData pattern. Each breakpoint carries per-session adapter data, enabling correct multi-session behavior — when two debug sessions target the same file, breakpoints remain verified as long as any session has acknowledged them. Source-map collapsing is handled gracefully with editor-level deduplication, and breakpoint identity is preserved across position changes caused by editing. This is a breaking change for downstream adopters who extended the previous marker-based breakpoint system, but it resolves long-standing issues with breakpoint state management.

Extension Host Debugging in Electron: Debugging VS Code extensions hosted in Theia’s Electron app is now fully functional. Previously broken due to process spawning, window loading, and debug session lifecycle issues, the hosted plugin debug experience now works end-to-end: a new Electron window opens for the hosted instance, breakpoints in extension code are hit, and the lifecycle is properly coupled — stopping the debug session closes the hosted window, and vice versa. New debug.focusWindowOnBreak and debug.focusEditorOnBreak preferences control focus behavior when the debugger hits a breakpoint.

Additional Improvements

The 1.70 release includes many more noteworthy changes beyond the scope of this article, e.g.:

Terminal improvements: terminal tab titles have been cleaned up to show the shell name when idle (e.g., “zsh”) and the running command name during execution, with shell-internal prefixes stripped. User-set terminal names are now preserved across shell title changes. Task terminals in tree mode are routed to a dedicated Tasks page within the terminal manager.

Auto-save with error awareness: the new files.autoSaveWhenNoErrors preference prevents auto-save when a file has diagnostic errors, matching VS Code’s behavior.

Window zoom status bar: a new zoom status bar item appears when the window zoom level is not at default, with interactive tooltip controls for quick zoom adjustments. Secondary windows now correctly inherit and synchronize the zoom level.

Hover tooltip improvements: hover tooltips throughout Theia now allow clicking and text selection, so you can copy content from tooltips without them dismissing.

Remote workspace support: when connected to a remote or dev-container, users can now open local workspaces and folders via a “Show Local” button in the file dialog. Devcontainer support also gains workspaceMount and workspaceFolder configuration options.

For a complete overview of all changes in this release, please refer to the 1.70 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