Related articles
- Claude Code Now Runs Code Reviews in the Background and Gets Opus 5
- Best Coding Agents 2026 — Decision Guide
- Gitlawb Zero Adds Undo, Side Conversations, and Configurable Sandbox
Cline, the open-source coding agent, shipped two updates this week that change how you interact with it. The extension — now at v4.0.11 — added Claude Opus 5 across every provider it supports. And for the first time, there is a standalone desktop app for macOS that lets you chat with Cline without opening a project folder.
If you have been using Cline only as a VS Code extension, these changes give you two new ways to work with it: a full desktop application and direct access to the most capable language model currently available.
The Cline desktop app: chat without opening a folder
Cline Desktop v0.0.4, released July 24, introduces something the extension could not do: workspace-free chat sessions. You launch the app and start talking to Cline immediately, without pointing it at a repository or project directory. This is useful for quick questions, brainstorming, or exploring ideas before you have code to work with.
The app is signed and notarized for Apple Silicon and Intel Macs. It checks for updates on launch and every two hours in the background, downloading new versions silently and prompting you for a one-click restart. Every future release arrives through the same mechanism — you download the DMG once and never touch it again.
What else is new in the desktop app
The list of changes in v0.0.4 is substantial:
- Drag and drop files directly onto the chat window to attach them. Cline can read and discuss the contents without you copying and pasting code.
- Image attachments display inline in the chat transcript, so screenshots, diagrams, and error messages appear exactly where you sent them.
- One-time routines are now supported alongside recurring ones. You can schedule a task to run once, and the app shows navigation to jump directly to a routine’s run history.
- Agentic compaction by default means long conversations stay within the context window more intelligently. Instead of losing earlier messages, the app compresses older context while preserving the important parts.
- Headless routines default to YOLO mode, so automated tasks run without manual approval prompts — useful for CI-style workflows you want to schedule from the desktop.
- CLI tool resolution is fixed, so commands like
gh(the GitHub CLI) are found correctly by resolving your login shell’s PATH instead of the app’s own environment. - A new custom overlay title bar adds in-app navigation, and channel setup has been redesigned as expandable cards.
If you have been running Cline through VS Code and wishing you could just open a chat window, this is the release that makes that possible.
Claude Opus 5: every provider, 1M context
Version 4.0.11 of the Cline extension adds Claude Opus 5 to every supported provider. That means Anthropic direct, Amazon Bedrock, Google Vertex AI, OpenRouter, and the Claude Code provider all get the new model, including the 1-million-token context window variants.
Claude Opus 5 is Anthropic’s most capable model. With 1M tokens of context, you can feed it an entire large codebase — thousands of files, tens of thousands of lines — and ask questions about it without hitting context limits. For coding agents, this matters because compaction (summarizing and dropping earlier context) is one of the main ways agents lose track of what they were doing. A larger context window means less compaction and fewer mistakes.
What the 1M context window means in practice
Think of context as the agent’s working memory. A standard 128K or 200K token window forces the agent to periodically summarize and discard older messages when conversations get long. With 1M tokens, you can have extended coding sessions — multi-file refactors, long debugging conversations, or iterative code review — without the agent forgetting what you discussed 50 messages ago.
The tradeoff is cost. Claude Opus 5 pricing sits at $10 per million input tokens and $50 per million output tokens in fast mode. For a typical coding session, this works out to a few dollars per conversation. For very long autonomous loops where the agent is reading and writing hundreds of files, costs can add up. Cline now corrects the pricing for the 1M context variants — earlier versions overstated costs for requests above 200K tokens.
Moonshot Kimi K3 support
Also in v4.0.11: Cline added support for Moonshot Kimi K3, a model from the Chinese AI lab Moonshot AI. The release fixes a problem where Kimi K3 returned empty responses by enabling native tool calling for the model. If you are experimenting with non-Anthropic providers, Kimi K3 is now a working option.
How to get started
For the desktop app: Download the DMG from the Cline GitHub releases page. It is available for macOS only for now. Launch it, go through the onboarding flow, and start chatting.
For Claude Opus 5 in the extension: Update to Cline v4.0.11 or later. If you already have Claude selected as your provider, Opus 5 should appear in the model picker. Select it to start using the 1M context window. Make sure you have API credits — Opus 5 is a premium model and is not included in free tiers.
For YOLO mode on headless routines: If you are running Cline routines from the desktop app, headless mode now defaults to auto-approve. You can override this in settings if you want manual approval for automated tasks.
What this means for Cline users
Cline has been steadily expanding beyond being just a VS Code extension. The desktop app makes it a standalone tool you can use independently of any editor. Combined with Claude Opus 5 support across every provider, it gives you access to the same frontier model that Claude Code uses — without committing to Anthropic’s ecosystem.
For developers who want to compare how different coding agents handle the same model, this is a useful moment. Claude Code, Cline, and OpenClaw all now support Claude Opus 5 with large context windows. The differences between them — approval workflows, sandbox behavior, context management, and pricing — are worth evaluating on your own codebase before picking one.