AI Bridge <p>Bring Claude Code into your JetBrains IDE for everyday development. Explore your codebase, plan changes and review proposed edits from the same workspace where you write code.</p> <h2>Features</h2> <ul> <li><strong>Work with project context.</strong> Send selected code or the current file to a conversation with <code>Alt+K</code>, and preview the context before submitting a message.</li> <li><strong>Review changes in the IDE.</strong> Inspect proposed edits in the native diff viewer and open related files and terminals directly.</li> <li><strong>Organize your work.</strong> Use separate chat tabs, task templates and project queues. Reorder pending messages or combine them while the current task continues.</li> <li><strong>Keep useful conversations.</strong> Search past messages, save bookmarks and record decision notes.</li> <li><strong>Find your own messages.</strong> Your messages are tinted in a color you choose, and a column of dots at the edge of the chat jumps to each one.</li> <li><strong>Recover from dropped connections.</strong> Temporary network and server errors are retried automatically, and the reply continues where it stopped.</li> <li><strong>Control long-running tasks.</strong> Set project concurrency limits, monitor local CPU and memory usage, and reduce background rendering with low resource mode.</li> <li><strong>Choose how Claude works.</strong> Select the model and reasoning effort, schedule messages and configure automatic continuation after usage limits.</li> <li><strong>Use your preferred language.</strong> The interface includes English and Turkish.</li> </ul> <h2>Requirements and setup</h2> <p>Use a JetBrains IDE based on IntelliJ Platform 2023.1 or newer with a JCEF-enabled runtime, Node.js and an installed, configured Claude Code CLI. Open the plugin from the <strong>Tools</strong> menu or press <code>Ctrl+Shift+C</code> to start a conversation.</p> <p>Use your existing Claude Code authentication or provider configuration. Messages, code context and tool requests are processed through the provider and services you configure; their usage terms and charges apply.</p> <h2>About</h2> <p>Independently maintained by Birkan Kader. Claude and Claude Code are Anthropic products. This plugin is not an official Anthropic or JetBrains product.</p> <h2>License</h2> <p>GNU AGPL v3. The license, copyright notices and corresponding source accompany the distribution. Provided without warranty.</p><p>Corresponding source: <a href="https://birkankader.com/plugins/ai-bridge/ai-bridge-0.7-source-69b1fc9e96f6.zip">Source ZIP</a></p> <h3>0.7 (2026-10-17)</h3> <ul> <li>New setting: Web page and shader tools (Settings → General, experimental, off by default). The agent can open a web page, a local file or your dev server in the Chrome or Edge already on your computer, headless, and gets its layout as text: the elements with their positions and styles, neighbours that are almost but not quite aligned with the styles that differ, text that is cut off, boxes that stick out of their parent, and the console's errors and warnings. It can also click, type, run a script on the page and take screenshots.</li> <li>A shader tool draws a GLSL fragment shader (a Shadertoy mainImage, GLSL ES 3.00 or 1.00) on the CPU, so no GPU is needed, and reports compile errors with their lines, the output's value ranges, NaN and values outside 0–1, and the colors at chosen points. A shader that never finishes is stopped after 20 seconds.</li> <li>For models that can't see images, screenshots and shader output come as the same text description as attached images.</li> <li>Nothing is downloaded: the browser runs with a throwaway profile, which is removed afterwards.</li> <li>Computer use on Windows: a window layout tool reads a desktop app's window as text, without a screenshot: each control with its type, name, state and position, neighbours that are almost but not quite aligned, controls that are cut off by their container, and controls that overlap. It reads WPF, WinForms, Win32 and other apps through UI Automation, and Swing apps through the Java Access Bridge (if it is off, run jabswitch -enable once). macOS and Linux are not supported yet.</li> </ul> <h3>0.6 (2026-10-17)</h3> <ul> <li>New setting: Models that can't see images (Settings → General). For a model listed there, such as glm-* or *, an attached image goes as a text description instead: the text on it, the layout of its shapes with their positions, shapes that are almost but not quite aligned, and its colors. A model without image support used to fail on the image, and every message after it failed too because the image stayed in the conversation.</li> <li>The text is read with the OCR built into the operating system (Windows OCR, Vision on macOS, tesseract on Linux). Nothing is sent anywhere and no GPU is needed. On Windows, install the OCR language pack of your language for its letters to be read right.</li> <li>The chat shows each description under the message, so you can see what the model got.</li> <li>Computer use: for these models the screenshot tool answers with the same description of the screen, read from the screen's full resolution.</li> </ul> <h3>0.5 (2026-10-16)</h3> <ul> <li>Send now on a queued message no longer waits for a long thinking block. While the model is thinking or writing, its reply is interrupted and your message runs at once. A running tool finishes first, and the message goes in as it returns.</li> <li>DeepSeek Harness and OpenCode: Send now during a tool call delivers the message when that tool returns, instead of after the whole answer.</li> </ul> <h3>0.4 (2026-10-16)</h3> <ul> <li>DeepSeek Harness uses the model providers you set up in dsh (a custom model API with its own URL and models, a DeepSeek endpoint, the default model). dsh saves them in the profile you ran, such as web, and AI Bridge runs the acp profile, so chats used to start on deepseek-official and failed with "no API key for provider route deepseek-official". AI Bridge now passes those settings to dsh with --patch and starts chats on your default model. Your dsh files are not changed.</li> </ul> <h3>0.3 (2026-10-16)</h3> <ul> <li>DeepSeek Harness from a source checkout (Location set to the deepseek-harness folder) now starts in your project the same way pnpm dsh does in the checkout, without building it first. Before, it stopped with exit code 1 because its packages were looked up as built files.</li> <li>When a harness stops, the chat shows the error it printed instead of only the last lines of its stack trace.</li> </ul> <h3>0.2 (2026-10-16)</h3> <ul> <li>Commits the agent makes no longer get a Co-Authored-By trailer. If your own Claude settings set attribution or includeCoAuthoredBy, that choice still applies.</li> <li>Settings, Connection lists every agent harness (Claude Code, OpenCode, DeepSeek Harness) with its status, version and location. A missing harness installs with one click, and each one takes a Location: an executable, an npm install folder or a deepseek-harness source checkout.</li> <li>A harness that is not on PATH is still found: AI Bridge checks its Location, then its own install, then PATH.</li> <li>When a harness cannot start, the chat says why (command not found, or the exit code and error output) instead of "The agent closed its output".</li> <li>The chat header, message author and input placeholder show the harness the chat runs on instead of always "Claude".</li> </ul> <h3>0.1 - AI Bridge (2026-10-16)</h3> <ul> <li>AI Bridge is now its own plugin (ID com.birkankader.ai-bridge) and its versions start again at 0.1. Builds up to 0.39 used a different ID and do not update to it: install AI Bridge from https://birkankader.com/plugins/ai-bridge/updatePlugins.xml, then uninstall the previous build so the two do not run side by side.</li> <li>On first start, settings, saved accounts, harness conversations, project settings and custom chat titles are copied from the previous build. Its own files are left untouched.</li> <li>New: choose OpenCode or DeepSeek Harness as the agent for new chats (Settings, Model, Agent harness). Both run through their official ACP command and are installed separately.</li> <li>Experimental: computer use (Settings, General), off by default. The agent gets screenshot, mouse and keyboard tools from a local MCP server; in the default mode every action asks for permission.</li> <li>The panel no longer looks blurry or pixelated after switching tabs, restoring a minimized window or moving it to a display with a different scale.</li> <li>Windows: Bash tool output uses UTF-8 in IDE installs as well, so non-ASCII characters are no longer garbled.</li> <li>Interface languages: Korean, Chinese and Japanese were removed; nine languages remain.</li> </ul> com.intellij.modules.platform org.jetbrains.plugins.terminal Arcade Machine Three office games in your IDE: Turkish Wordle, live drawing rooms, daily Sudoku. Connect to your office's Arcade Machine LAN server for shared daily challenges and live leaderboards. <p>A colorful new three-game menu: mint Wordle, coral Drawing and blue Sudoku. Refreshed game surfaces, clear controls and readable light/dark themes.</p><p>Higher / Lower has been removed from the client, server and rankings. Update the server and plugins together. Existing player data and progress in the remaining games are preserved.</p> com.intellij.modules.platform Haftanin Seyirligi Weekly movie and TEDx voting board for teams, powered by the shared Arcade Machine server. <p> Ekibin haftalık film ve TEDx panosu. Öner, oy ver, haftanın seçimini izle. Ortak ekip sunucusu ile öneriler, oylar ve izleme durumu senkronize edilir. </p> com.intellij.modules.platform