Cline or Continue: agent or assistant, which to pick in VS Code

The query cline vs continue usually means: I have VS Code, I want to plug in my own model, and I need to decide which of the two to install. They really do solve different problems. Cline is built as an agent that takes a whole task. Continue is more a set of tools beside the code: chat, editing a selected fragment, autocomplete and, in newer versions, an agent mode. We cover how they differ in practice, how to connect your models, and how spending differs.

The descriptions come from the docs.cline.bot and docs.continue.dev documentation as of October 18, 2026. We do not compare answer quality and do not claim one tool is better than the other: that depends on tasks and the model.

Cline: an agent with a plan and approvals

Cline reads project files, edits them and runs terminal commands, but with your consent. The main ideas from the docs:

• Plan and Act. In Plan mode Cline can read code and discuss the approach, but does not change files or run commands. When the plan suits you, switch to Act, and the conversation context is kept. The docs recommend starting with a plan.
• Auto Approve. Approval is decided separately for each action type: read files, edit, run commands, use the browser. You can allow the safe ones and be asked about the rest.
• Checkpoints. They let you roll back code changes while keeping the conversation.
• MCP, rules, skills. External tools and project instructions are connected separately.
• Besides the editor extension there is a CLI and a desktop app.

So Cline makes sense for multistep tasks: "work out why the build fails, fix it and run the tests".

Continue: an assistant around the code

The Continue docs list four modes: Agent for development tasks together with AI, Chat for questions, Edit for modifying fragments without leaving the file, and Autocomplete for line suggestions as you type. There is also a console client, cn. Everything is configured through configuration (the config.yaml format), which describes models, rules, MCP servers and prompts.

A distinctive trait of Continue is that model roles are separated. You can assign one model to chat, another to autocomplete, a third to applying edits. This matters for cost, as covered below.

Continue is worth choosing when your main work is writing code yourself and you need help in specific spots: explain a section, rewrite a function, complete a line.

Comparison

The "Rollback" and "Approvals" rows for Continue are worded cautiously: details depend on the version, so check the current docs.

Connecting your own models through Zukko

Both tools can work with an OpenAI-compatible API, and so with one gateway key. For such clients the address is https://api.zukko.pro/v1, as described in the VS Code extensions documentation↗.

Cline. In settings pick the OpenAI Compatible provider, then Base URL https://api.zukko.pro/v1, your fetch_... key and a model id such as claude-sonnet-5. The Cline docs separately describe a model parameters block: max output tokens, context window size, image support and prices. Fill it in from GET https://api.zukko.pro/v1/models. A step-by-step guide is in the Cline article↗.

Continue. Models are declared in config.yaml. The structure is roughly as below, but field names and the schema version depend on the extension version, so check the config.yaml Reference section in the Continue docs:

name: Zukko
version: 1.0.0
schema: v1
models:
  - name: Claude Sonnet 5
    provider: openai
    model: claude-sonnet-5
    apiBase: https://api.zukko.pro/v1
    apiKey: fetch_your_key
    roles: [chat, edit, apply]
  - name: Autocomplete
    provider: openai
    model: gpt-6-luna
    apiBase: https://api.zukko.pro/v1
    apiKey: fetch_your_key
    roles: [autocomplete]

Do not keep the key in the file itself if the file may land in a repository. See the Continue documentation for how to substitute secrets. For the setup process read Continue.dev: connecting to the gateway↗. Whether a given model suits autocomplete and how fast it answers, test yourself: latency matters more here than in chat.

One caveat about Claude. Both tools use the OpenAI format, and in that format the gateway bills all Claude input at the input price and gives no cache discount. For GPT models cache reads are visible in both formats. For the calculations below we take this worst case.

What each mode costs

Their spending profiles differ. An agent pays for context volume, autocomplete for the number of small requests.

An agent session in Cline. 1.5M input tokens and 150K output tokens. Zukko prices on October 18, 2026, $ per 1M tokens, no cache.

• claude-sonnet-5 (0.4 / 2): 1.5 × 0.4 + 0.15 × 2 = 0.6 + 0.3 = $0.90.
• gpt-6.1-sol (0.27 / 1.32): 1.5 × 0.27 + 0.15 × 1.32 = 0.405 + 0.198 ≈ $0.60.
• deepseek-v4.1-flash (0.013 / 0.038): 1.5 × 0.013 + 0.15 × 0.038 = 0.0195 + 0.0057 ≈ $0.025.

Autocomplete in Continue. Say 2,000 requests per day, 1,500 input tokens and 60 output tokens each. That is 3M input tokens and 0.12M output tokens per day.

• gpt-6-luna (0.04 / 0.21): 3 × 0.04 + 0.12 × 0.21 = 0.12 + 0.0252 ≈ $0.145 per day, about $3.19 over 22 working days.
• qwen3.8-flash (0.038 / 0.075): 3 × 0.038 + 0.12 × 0.075 = 0.114 + 0.009 = $0.123 per day.

The request counts and sizes are illustrative; yours will differ. They show the main thing: if you use autocomplete, assign the cheapest suitable model to it and keep the strong one for chat and the agent. In Continue this is done with roles, in Cline by a separate model choice. Fresh prices are always in the GET https://api.zukko.pro/v1/models response.

What to pick

You want "do the task for me and ask when in doubt": Cline. You want "suggest as I write and help in spots": Continue. You can install both: they do not interfere, and the key and balance are shared. Among extensions close to Cline with the same gateway setup there is Kilo Code. To try, create a key in the dashboard↗ and set a daily limit.

FAQ

Cline alternatives: what else is there?

A related VS Code extension is Kilo Code; its setup is described in the Zukko documentation↗. Roo Code, often named alongside, was shut down on May 15, 2026 according to its developers. Among other tools there is Continue.

Can Continue be used as an agent?

The Continue docs list an Agent mode. Whether it suits you compared with Cline, check on a real task.

Which base URL should I use?

For both, https://api.zukko.pro/v1, because they work through the OpenAI-compatible format. The address without /v1 is for Claude Code and the Anthropic SDK, which is a different case.

Why did spending grow in Cline?

An agent sends the model the accumulated context at every step. Watch spend per key in the dashboard and cap the key with a daily limit.