Cline: what this VS Code agent is and what it costs to run

Cline is an AI coding agent that lives in your editor. Unlike a suggestion chat, it reads your project files, proposes edits, runs commands in the terminal, and continues based on the results. If you searched for what Cline is and how to use it, here is a short walkthrough: what it consists of, how the Plan and Act modes work, how to connect your own key and what a task costs.

The description of features follows docs.cline.bot as of October 18, 2026. The product changes quickly, so check UI names against your own version.

What Cline is

According to the documentation, Cline is an agent that can read and write files, run terminal commands, use a browser and build features through conversation. Every action needs your approval until you turn on auto-approve yourself. It runs in several shells on one core:

• a VS Code extension (the main option, including compatible editors such as Cursor and Windsurf);
• a plugin for the JetBrains family;
• a command-line version (CLI) and a terminal interface;
• a desktop app and Kanban for parallel agents;
• an SDK for building your own agent.

Cline does not necessarily ship with a model: you pick how to get access. That can be Cline's own usage-billed provider, their subscription, or your own key at a third-party vendor (BYOK). Terms and prices on Cline's side change, so see the Cline documentation for the current ones. The rest of this article is about the own-key option.

Plan and Act modes

The main idea of Cline is separating thinking from doing.

Plan. Cline reads code, searches the project and discusses the approach, but cannot change files or run commands. The mode suits unfamiliar code, architecture decisions and debugging when it is unclear what is breaking.

Act. Cline carries out the plan: writes files, runs commands, executes tests. The conversation history is kept when you switch, so you do not need to repeat agreements.

The usual order: describe the task in Plan, discuss the approach, switch to Act. The documentation recommends planning before acting. You can enable different models for the modes ("Use different models for Plan and Act" in settings): a stronger one for planning and a cheaper one for execution.

What Cline works with

• Files. Add context with @ mentions and drag and drop.
• Terminal. Commands run through your terminal; the model itself separates safe commands from those needing approval, and you configure what is allowed without asking.
• Checkpoints. They let you roll back code changes while keeping the conversation.
• Rules and skills. Permanent requirements go in .clinerules, rare long instructions become skills.
• MCP. External tools connect to the agent through MCP servers.
• Slash commands. For example, /newtask carries the essence of a task into a clean context, /smol compresses the history, /deep-planning builds a detailed plan.

How to use it: a quick start

1. Install the Cline extension in VS Code from the Marketplace.
2. Open the Cline panel and go to settings (gear icon).
3. Choose the OpenAI Compatible provider.
4. Enter Base URL https://api.zukko.pro/v1, your fetch_... key and a model id such as claude-sonnet-5-5. The address must include /v1: without the suffix requests go to the wrong path and return 404.
5. If settings have a context window field, set the value from the model table (1M tokens for Claude and GPT, 128K for GLM).
6. Open a project, switch to Plan and describe a small task, for example "add input validation to function X and cover it with a test".
7. Read the plan, switch to Act and approve steps until you get used to it.

A detailed walkthrough is in the VS Code extensions docs↗ and in the article Cline: connecting to Claude and GPT with your own key↗. Create the key in the dashboard↗.

What it costs

With your own key the extension itself adds no fee: you pay the model vendor for tokens. And an agent spends noticeably more tokens than a chat, because at every step it sends the accumulated history.

Let us estimate a typical task: 25 agent requests, on average 40,000 input tokens and 1,000 output tokens each. Total input 40,000 × 25 = 1,000,000, output 1,000 × 25 = 25,000 tokens. Zukko prices as of October 18, 2026 per 1M tokens, ignoring cache:

For comparison, at the official prices from the same table, claude-sonnet-5-5 (input $2, output $10) gives 2 + 10 × 0.025 = $2.25 for the same task.

Three caveats. First, this is an illustrative calculation: real tasks can be several times shorter or much longer. Second, glm-5.3-flash has a 128K window and on long tasks it runs out sooner than models with 1M. Third, in the OpenAI format Claude is billed at the full input price, while cheaper cache reads come only with the Anthropic format (Cline can also work through the Anthropic provider, in which case the Base URL is https://api.zukko.pro without /v1; see choosing the format↗). How this difference is calculated is covered in the article on agent cost with cache↗.

We do not rank model quality: rely on price, context and the intelligence index (58 for claude-opus-5-5, 52 for gpt-6.1-sol, 48 for claude-sonnet-5-5) and test on your own tasks. Current prices are available through GET https://api.zukko.pro/v1/models.

How not to overspend

• Start a new task at the beginning of each new topic; a long history costs money at every step.
• Use a cheap model for routine work and keep the strong one for planning.
• Set a daily and monthly dollar limit on the key: with auto-approve on, it stops the agent before it overspends.
• Do not enable auto-approve for all commands on an unfamiliar project.

You can top up the balance via SBP (the payment page adds an 8.5% fee on top), cryptocurrency or a MasterCard card. There is no subscription; you pay for the tokens you use.

FAQ

Is Cline paid?

You can install the extension and use it with your own key, in which case you pay the model vendor for tokens. Cline itself has its own ways to access models with separate terms; check the site and the Cline documentation for their current form.

How is Cline different from a chatbot?

A chat answers with text, while Cline acts: it reads files, writes edits, runs commands and checks the result. That is why it spends more tokens and needs attention to approvals.

Cline or Claude Code, which to choose?

It is a matter of taste: Cline is built into the editor, Claude Code works in the terminal. If you spend the day in VS Code and want to see each edit in the editor, start with Cline; if you work in a terminal or on a server, Claude Code is handier.

Is it safe to give Cline terminal access?

While auto-approve is off, you confirm every action. Keep the project under git so you can roll back edits, and do not give the agent secrets it does not need to see.

To try it, create a key in the Zukko dashboard and connect it in Cline following the steps above.