What's Changed

• [Perf] Add more detailed message to FileSystemPlugin
• [Perf] Add mandatory free-text field & description to ask_user_question tool
• [Fix] TextDifference.WaitForAcceptanceAsync throws NullReferenceException
• [Fix] TerminalRun.CommandLine may be incomplete
• [Fix] emoji is not supported by TerminalParser
• [Feat] Add display for cloud service quota limit
• [Fix] Button background flickers in dark mode
• [Fix] ChangeLogView margin

Full Changelog: v0.8.0-canary.20260612.3...v0.8.0-canary.20260612.4

What's Changed

• [Feat] Add support for SKILL.md by @DearVa in #423
• [Feat] Add Claude Fable 5
• [Perf] Web search & mcp buttons & more UI layout
• [Perf] Add more i18n for cloud service
• [Fix] SoftwareUpdateControl version display error
• [Fix] Handle file delete failure in FileDownloadService

Full Changelog: v0.8.0-canary.20260611.2...v0.8.0-canary.20260612.3

Contributors

• @DearVa

What's Changed

refactor: decouple Everywhere.Terminal project
feat: add Claude Opus 4.8 & Gemini 3.5 Flash & MiniMax-M3 to preset
feat: Add more parameter settings beyond temperature and top_p (#378)
fix: toast dismiss will cause exception
fix: ChatPluginFileReference cannot be serialized and saved

Full Changelog: v0.8.0-canary.20260608.1...v0.8.0-canary.20260611.2

What's Changed

• feat: add {DefaultSystemPrompt} prompt variable with recursive resolution by @sstar16 in #404

• feat: Update channel by @DearVa in #417

• feat: Modern UI by @AuroraZiling and @DearVa in #419

• feat: per-schema model options & claude 4.8 support (#410, #412)

• feat: add support for AnySearch

• perf: optimize markdown selection & performance

• refactor: decouple I18N into separated projects by @DearVa in #401

• refactor: add deduplicater for ChatPlugin (#397)

• fix: ensure ToastHost and DialogManager initialization always occurs by @Fillip74 in #403

• fix: ensure emoji dictionary loads for English locale by @Fillip74 in #413

• fix: support empty reasoning_content rendering

• fix: Chat titles being truncated too short (#399)

• fix: memory leak

New Contributors

• @sstar16 made their first contribution in #404

Full Changelog: v0.7.8...v0.8.0-canary.20260608.1

Contributors

• @Fillip74
• @DearVa
• @AuroraZiling
• @sstar16

⚠️ Important Notice

Due to refactoring of the Execute In Terminal tool, the new version may cause some unexpected issues. If you encounter any problems, please don't hesitate to report them to us. We will prioritize fixing critical issues as soon as possible.

✨ Features & 🚀 Improvements

• 🔥 Completely refactored the Execute In Terminal tool with PTY technology. This achieves comprehensive terminal integration and unified cross-platform behavior. As a result, the Windows installer size is reduced by ~80MB, and the execution stability is significantly improved, laying a solid foundation for features like real-time output display and upcoming command security reviews.
• 🔥 Introduced runtime environment detection to the MCP Manager. If dependencies like uv or Node.js are missing, Everywhere will notice you with in-app downloading and configuration.
• 🔥 Massively optimized the content truncation mechanism for all tool outputs, enforcing a hard limit of ~40K tokens per tool execution. This prevents "token bombs" from crashing the model context, especially when Read File or Extract Web Content.
• 🔥 Added the "Continue and Retry" button: When an error interrupts the chat, you can now resume the conversation while preserving the partially generated message from the current turn, instead of starting a new branch.
• 🔥 Added configuration options for Default Subagent and Image Understanding Subagent. This allows the primary assistant to delegate tasks to lightweight models to save costs, or to bridge the gap if your primary assistant lacks multimodal capabilities.
• Added GPT-5.5 to the preset mode and removed obsolete Gemini and Kimi models.
• Added auto-download updates in background: When network conditions are ideal and a new version released after 1 day, Everywhere will download updates in the background with speed limits. Intrusive update notifications have been removed.
• Added more advanced options for reasoning parameters (#361).
• Added loading hints while the LLM is generating tool parameters, which is particularly helpful for tools requiring long, complex inputs.
• Optimized window transition and navigation logic for a smoother UX.
• Improved the visual styling and layout of the Chat History drawer.
• Refined the login workflow and fetching logic for official models in Everywhere Cloud, ensuring a more stable connection experience. More user-friendly tooltips were also added.
• Settings for Temperature and Top-p are now automatically hidden if the currently selected model does not support them.
• Stabilized the logic for handling signatures during LLM calls to reduce unexpected validation errors, and added more detailed error messages to help diagnose signature-related failures.

🐛 Bug Fixes

• Fixed an issue where the Web Search toggle button remained visible in the chat input area even after tool-calling permissions were explicitly disabled.
• Fixed an issue where API error messages were sometimes not displayed correctly in the UI when using the OpenAI Responses schema.
• Fixed an issue where the input box for the Ask User Question tool would fail to wrap text under specific circumstances.
• Fixed an issue where sub-agents could incorrectly invoke the Ask User Question tool, which previously caused the entire task execution flow to halt unexpectedly.
• Fixed an issue in Light Mode where some code block syntax highlighting colors were difficult to read.
• Fixed minor typos and text description errors in the UI (#373 - Thanks to @Fillip74).

Full Changelog: v0.7.7...v0.7.8

Contributors

• @Fillip74

✨ Features & 🚀 Improvements

everywhere-cloud-search

• 🔥 Everywhere Cloud Services now includes an out-of-the-box Web Search API. Existing third-party web search configurations remain unaffected. We also refactored the related entry points and configuration pages to enhance the overall user experience.
• 🔥 Completely refactored the "Web Content Extraction" tool, improving the extraction success rate while significantly reducing token consumption. A dedicated configuration entry has also been added.
• 🔥 Massively optimized the "File System" tool by introducing safety features like timeout protection and context truncation. Prompts and parameters were also refined to ensure more stable tool invocation by LLMs.
• Added an "Allow for this session" option for the Terminal tool, along with an "Auto-approve" toggle in the settings for on-demand use. (⚠️ Note: Data is invaluable; please be cautious when granting automatic terminal execution permissions!)
• The chat window can now be resized to a smaller minimum dimension, offering more flexibility when desktop space is limited.
• Removed the dependency on 0Harmony.dll to prevent false positive detections by antivirus software.
• Added more helpful tooltips and hint texts throughout the application.

🐛 Bug Fixes

• Added extra safeguards to ensure the background Everything search process is reliably and completely terminated when closing the application.
• Fixed an issue where the cache directory was incorrectly located in some scenarios.
• Fixed an issue where models like MiniMax and Xiaomi Mimo would throw a missing signature error when used under the Anthropic schema.

Full Changelog: v0.7.6...v0.7.7

🐛 Bug Fixes

• Fixed an issue where the main window opened every time when started.

Full Changelog: v0.7.5...v0.7.6

✨ Features & 🚀 Improvements

• 🔥 Completely refactored the Chat History page, making it much smoother and more convenient to view, manage, and delete chats.
• 🔥 Completely refactored the "web-snapshot" tool, improving the success rate of webpage visits and the information density of the extracted content.
• Added settings to customize the theme color.
• Optimized the "run-subagent" tool: Sub-agents will now inherit the current visual context and session-level tool approval permissions from the main conversation.
• Enhanced tool execution and bubble interactions: Added more detailed message for tool calls; extended the auto-collapse delay for the "Deep Thinking" bubble; and disabled the auto-collapse behavior for tool execution bubbles to let you easily track the process.
• Improved the visual display of software update push notifications.
• Disabled unnecessary UI animations to reduce performance overhead and improve responsiveness.

🐛 Bug Fixes

• Fixed an issue where configurations could not be properly saved or loaded in certain system languages due to decimal separator format differences (e.g., period vs. comma) (#358).
• Fixed an issue where the "manage-todo-list" tool did not correctly update UI.
• Fixed a crash when rendering specific LaTeX formulas.
• (Windows) Fixed an issue where the SetText for automation operation will always fail.

Full Changelog: v0.7.4...v0.7.5

🐛 Bug Fixes

• 🔥 Fixed an issue where multi-turn conversations would fail when both "Tool Calling" and "Deep Thinking" were enabled simultaneously. This previously caused models like DeepSeek and Kimi to fail functioning properly.
• Fixed an issue where the automatic selection of the "System Assistant" might fail to work in certain situations.
• Fixed an issue where the items parameter in the "Manage Todo List" tool was incorrectly set as required, which previously caused the model's initial tool call to fail and led to confusion.

Full Changelog: v0.7.3...v0.7.4

✨ Features

• 🔥Added DeepSeek V4 Pro and DeepSeek V4 Flash to the preset mode, and removed the outdated DeepSeek Chat and DeepSeek Reasoner models.
• Added an "Disabled" level to the "Reasoning Effort" option in the chat window. For supported models (such as DeepSeek V4, certain Claude models, Xiaomi Mimo, etc.), you can now use this level to completely disable the model's deep thinking process.

🐛 Bug Fixes

• 🔥 Fixed an issue where the chat window could not be brought up via shortcuts when the "Automatically Add Element" option was disabled. Special thanks to all the users who helped us troubleshoot and isolate this issue! (#345, #350, #354)
• Fixed an issue where the send button remains disabled after selecting a strategy, forcing users to input text before being able to send.
• Fixed an issue where generic strategies failed to display in the list.
• Fixed an issue where the input box was not correctly populated with the original text when editing a user message.
• Fixed an error that occurred when Gemini models attempted to execute parallel tool calls.
• Fixed an issue where auto-generated chat topics would occasionally appear as garbled text under specific circumstances.