Cline for VS Code: The Open-Source Coding Agent
Cline is an open-source coding agent that runs inside VS Code, and it started life as an extension called Claude Dev. It reads your project, proposes a plan, then edits files and runs terminal commands after you approve each step. You bring your own model key or pay Cline for inference. If you are choosing between Cline and the agent built into VS Code, the main differences are the model you can use and where instructions live. You will also find install steps below. The last part shows how rules files and the Memory Bank carry context between tasks.
What is Cline, in one sentence?
Cline is an Apache-2.0 coding agent from Cline Bot Inc. that runs in your editor and terminal and asks for approval before it edits a file or runs a command.
The GitHub README calls it "the open source coding agent in your IDE, terminal, & desktop." The repository had 71.0k stars when we checked on October 28, 2026. The VS Code Marketplace listing showed 5,768,831 installs on the same day. The listing still uses the extension ID saoudrizwan.claude-dev, a leftover from the old name. The v2.0.0 release notes explain that Claude Dev was a tribute to Claude 3.5 Sonnet, and version 2.0 gave the project its own brand.
Cline now ships on several surfaces. The docs overview lists the VS Code extension, a JetBrains plugin, a CLI, an SDK and a desktop app. The latest extension release on GitHub was v4.1.26, published on October 26.
How does Cline differ from the agent built into VS Code?
People searching "cline vs vscode agent" usually mean the agent that ships with VS Code and GitHub Copilot. Both can edit code and call MCP servers. The main differences are who supplies the model and where the instructions live.
| Cline | VS Code built-in agent | |
|---|---|---|
| License | Apache-2.0 extension (the JetBrains plugin is not open source) | Part of VS Code with GitHub Copilot |
| Models | Your own key for Anthropic, OpenAI, Google, OpenRouter, Bedrock, Ollama and others, or Cline's provider | Copilot models (Copilot Free is available), or your own key |
| Approval | Every file edit and command needs approval unless you turn on auto-approve | Per-session permissions and checkpoints |
| Instruction files | .clinerules, global rules, and detected AGENTS.md and .cursorrules files | .github/copilot-instructions.md, root AGENTS.md, CLAUDE.md as an option |
| Undo | Checkpoints stored in a shadow Git repo | Checkpoints |
| Other surfaces | JetBrains, CLI, SDK, desktop app | VS Code only |
Sources: Cline's README and rules docs. VS Code's agent docs and setup page. Instruction files are on VS Code's custom instructions page.
VS Code's setup page says signing in without a plan enrolls an eligible account in Copilot Free. A bring-your-own-key option can give chat access without a Copilot plan. If you already pay for Copilot, the built-in agent costs nothing extra. Cline makes more sense when you want to pick the model per task or use the same agent from the terminal.
How do Plan and Act modes work?
Cline splits a task into two modes. In Plan mode, the docs say Cline "cannot modify any files or execute commands." It reads the code and asks questions, then drafts a strategy with you. Switching to Act mode lets it modify files and run commands to carry out that plan.
Approval is the default in both modes. The README states that every file edit and terminal command requires your approval. The tools reference describes a common setup that auto-approves read and search tools. Edits and commands still wait for a click.
Checkpoints let you undo Cline's edits. Per the checkpoints docs, Cline stores snapshots of your workspace in a shadow Git repository that stays separate from your own. It is on by default. You can restore the files and the task conversation separately or together, so a bad edit does not cost you the discussion.
Cline also supports MCP. Per the MCP docs, you add servers from the MCP Servers panel as JSON or as remote servers. The CLI has a cline mcp command. The pricing page lists an MCP Marketplace in the free tier.
What does Cline cost?
The extension is free. The pricing page says you "only pay for AI inference on a usage basis." The docs overview lists three ways to pay for inference:
- the Cline provider, billed by usage, with some free model options
- ClinePass, a flat $9.99 a month for open models such as GLM, Kimi, DeepSeek and Qwen
- your own API key with any supported provider
The ClinePass page claims "2-5x the usage on popular open coding models" compared with paying API rates. That is Cline's own figure. The page also says you don't need ClinePass to use Cline. An Enterprise plan adds SSO and audit logs, among other controls, at a custom price. No team tier price is published.
How do you install Cline in VS Code?
The install guide has four steps:
- Open the Extensions view with Ctrl+Shift+X (Cmd+Shift+X on macOS).
- Search for "Cline" and install the extension published by Cline.
- Open Cline from the sidebar.
- Choose how to pay for the model, as described in the pricing section above.
The same steps work in editors that use Open VSX, such as Cursor and Windsurf. For the terminal, the guide installs the CLI with npm install -g cline and signs in with cline auth.
How does Cline remember a project between tasks?
Each Cline task starts from what you give it plus what it reads. Two mechanisms carry knowledge forward, and a third keeps long tasks going.
Rules are, in the docs' words, "Markdown files that provide persistent instructions across all conversations." Workspace rules live in .clinerules/ at the project root. Global rules live in Documents/Cline/Rules. Workspace rules win when the two conflict. Cline also detects AGENTS.md and the rules files of other editors, so a team can share one instruction file across tools.
The Memory Bank is a documentation method, not a built-in feature. The Memory Bank guide has Cline keep six Markdown files, such as activeContext.md and progress.md. A rule tells Cline to read them at the start of each task and update them as work moves. You start it by asking Cline to "initialize memory bank."
Auto Compact handles long tasks. Cline summarizes a conversation when it nears the model's context limit, then continues. That behavior is described in the Auto Compact docs. The deprecations page lists Focus Chain as deprecated in the VS Code extension.
All of this stays inside Cline. Workspace rules and the Memory Bank are kept per repository.
What should you check before relying on Cline?
Most popularity numbers come from Cline. The homepage says "Trusted by 11M+ developers," a figure that covers every surface. The Marketplace count we could verify is about 5.77 million installs. ClinePass usage multiples are also vendor claims with no independent test we could find.
Approval on every edit is slow, which tempts people to turn on auto-approve. In our reading, checkpoints restore workspace files. They cannot undo a command that touched something outside the workspace, such as a database or a remote server. Keep command approval on for anything with side effects.
Costs depend on your model. Cline sends your files and conversation to the provider you choose, so long Act sessions on frontier models can run up a bill.
The bottom line
Install Cline if you want an open-source agent in VS Code where you choose the model and approve each change. Start in Plan mode and keep command approval on. Write a short .clinerules file before your first real task. If you already pay for Copilot and use one model, try VS Code's built-in agent first, because it needs no extra setup.
How do you keep coding agent memory across tools?
Cline's rules and Memory Bank work well inside one repository. The trouble starts when you also use Claude Code in the terminal or Codex for background tasks. Each tool keeps its own files. A fact Cline wrote to activeContext.md never reaches the other agents. The Memory Bank also needs the model to remember to update it, and it only tracks what the rule asks for.
In memU, a scheduled background task slices new session logs into jobs. The agent distills each job into memory and skill Markdown, then commits it. memU's MemoryService makes no LLM calls itself: it stores, embeds and retrieves what the agent prepared.
A standing instruction in CLAUDE.md or AGENTS.md tells the agent to retrieve before answering. All hosts share one backend, so a fact one host saved is available to the others. Named hosts include Claude Code and Codex, plus Cursor. Cline is not on the named list, so check the support matrix in the memU repository before you count on it.
To try the hosted version, which is free and syncs across devices, get an API key from memu.so and send your agent one message:
Read https://memu.pro/SKILL.md, follow its instructions to install and configure memU
memU is open source under Apache-2.0. Self-hosting needs an embedding key and stays on one device.
FAQ
Is Cline the same as Claude Dev?
Yes. Cline was released as Claude Dev and renamed with version 2.0. The VS Code Marketplace still lists it under the ID saoudrizwan.claude-dev.
Is Cline free?
The extension is free and open source under Apache-2.0. You pay for model inference, for example with your own API key. Cline also sells usage-based inference and the $9.99 ClinePass for open models.
Does Cline work with local models?
Yes. The README lists Ollama and LM Studio as providers, along with any OpenAI-compatible API. Small local models may struggle with long multi-file tasks.
Where does Cline store its rules?
Workspace rules go in a .clinerules/ folder at the project root. Global rules go in the Documents/Cline/Rules folder on every desktop OS.
Can Cline run outside VS Code?
Yes. Cline also has a JetBrains plugin and a CLI installed with npm. An SDK and a desktop app complete the set. It also installs in Cursor and Windsurf from Open VSX.