Cline vs Roo Code: comparing the two VS Code agents in 2026
Roo Code grew out of Cline, then the paths split. We compare modes, rules and token spend, and cover what the Roo Code shutdown changes.
6 min · · IDE · сравнение
Cline and Roo Code were long treated as twins: Roo grew out of Cline, and their provider settings are almost identical. The "cline vs roo code" question used to come down to taste, with some people preferring Roo's modes and others Cline's simplicity. There is now an update that the comparison cannot skip, and the first section covers it. After that we go through the differences you notice in daily work and connect both tools to one Zukko key.
What is going on with Roo Code
According to the official Roo Code documentation, the extension was shut down on May 15, 2026; the RooCodeInc/Roo-Code repository on GitHub has been archived, with its last changes dated the same day. The documentation points to two alternatives: ZooCode (a fork started by the community) and Cline itself, where Roo originally came from.
What this means:
- If Roo Code is already installed, it may keep working with your key for a while, but there will be no updates or fixes. It will not pick up new models or changes in provider APIs.
- For a new project it is more sensible to look at Cline or at the fork. Check the fork's state yourself: repository activity, recent releases, compatibility with your VS Code version. We have not tested it.
- As of October 18, 2026 Cline is under active development: the cline/cline repository has fresh commits, and the documentation lists the VS Code extension next to a JetBrains plugin, a CLI, a desktop app and an SDK.
The comparison below covers what both tools could do and how their approaches differed. Those differences also matter when choosing between Cline and a Roo fork.
Cline vs Roo Code: what differs in daily work
| Cline | Roo Code | |
|---|---|---|
| Modes | Two: Plan and Act | Code, Ask, Architect, Debug, Orchestrator and custom modes |
| Model per mode | A "Use different models for Plan and Act" switch | Each mode remembers its last used model |
| Project rules | .clinerules/, also reads .cursorrules, .windsurfrules, AGENTS.md | .roo/rules-{slug}/ per mode, custom modes in .roomodes |
| Delegation | Subagents, but only in the CLI and Desktop; the VS Code extension does not use them | Orchestrator hands subtasks to other modes through new_task |
| Where it runs | VS Code, JetBrains, CLI, Desktop | VS Code |
| Tool calling | See the Cline documentation for details | Native tool calling only, no XML fallback |
Modes versus two phases
Cline is simple. In Plan the agent reads code and discusses the approach but does not change files or run commands. In Act it carries out the plan, and the conversation history is kept. That is enough for most work, and switching takes one click.
Roo goes further: each mode has its own set of allowed tools. Ask can only read, Architect can write only markdown files, Code and Debug get everything. You can define your own modes: a role, a list of tools, file patterns that may be edited. That helps in a team, because a "reviewer" mode physically cannot change code. Orchestrator also splits a large task into subtasks, each with its own context window.
Rules and memory
Cline rules live in .clinerules/ or .cline/rules/ at the project root and can be toggled in a separate panel. A nice detail: Cline picks up rules written for other tools, that is .cursorrules, .windsurfrules and AGENTS.md. If your repository already has such a file, you do not need to configure anything again. In Roo, rules are tied to modes: a mode with the slug docs has its own files in .roo/rules-docs/.
Approving actions
Both agents ask permission for every action by default. Cline's Auto Approve works by category: reading project files, editing project files, safe commands, all commands, browser, MCP. The model itself marks a command with a requires_approval flag, which is an estimate by the model and not a guarantee. For someone else's code or a production server, keep manual approval; for your own git repository you can allow reads and edits.
Context and token spend
Both agents send the accumulated conversation to the model at every step. So the bill is decided not by the interface but by how you pay for the repeated context.
An example at prices as of October 18, 2026. Take claude-sonnet-5-5: input 0.4, output 2, cache read 0.05, cache write 0.5 dollars per 1M tokens. A session of 30 requests, with hypothetical volumes (an assumption for the example). Each request: 40,000 tokens read from cache, 4,000 written to cache, 1,000 as fresh input, and the model answers with 800 tokens.
- With cache (Anthropic format): 1,000 × 0.4 + 40,000 × 0.05 + 4,000 × 0.5 + 800 × 2 = 400 + 2,000 + 2,000 + 1,600 = 6,000 micro-dollars, so $0.006 per request and $0.18 per session.
- Without cache accounting (OpenAI format, all input at the input price): 45,000 × 0.4 + 800 × 2 = 18,000 + 1,600 = 19,600, so $0.0196 per request and $0.588 per session.
The practical conclusion: for Claude, connect the agent in Anthropic format when the tool allows it. Your real numbers will differ with project size; check them under "Usage" in the dashboard.
Connecting to Zukko
Create a key on the API keys page. Full instructions are in the VS Code extensions documentation.
Cline, OpenAI-compatible provider
- Gear icon in the Cline panel, API Provider: OpenAI Compatible.
- Base URL:
https://api.zukko.pro/v1, API Key:fetch_.... - Model ID:
claude-sonnet-5-5. If the model settings have a context window field, enter 1000000.
Cline, Anthropic provider. The Cline documentation describes a "Use custom base URL" checkbox there. The address has no suffix: https://api.zukko.pro. Requests then go in Anthropic format, and the cache is billed at the read and write prices.
Roo Code or its fork. Anthropic provider, enable the custom base URL, address https://api.zukko.pro without /v1, key fetch_.... Roo sends it in the x-api-key header, which the gateway accepts. If the model you need is not in the list, switch to OpenAI Compatible with https://api.zukko.pro/v1 and type the model ID by hand. A fork may name menu items differently, so check its documentation.
For OpenAI-format requests, make sure the model supports tool calling: the Roo documentation states that a model without it will not work. More on each tool: Cline setup, Roo Code quick setup.
Who should pick what
- You want a tool that is supported today with minimal setup: Cline. It runs in VS Code, JetBrains, the terminal and a desktop app, and rules carry over from other tools.
- You need strict roles with permission limits and subtask orchestration: look at a Roo fork, but first judge whether it is alive.
- You already work in Roo Code and are happy: no need to rush, but plan the move. Rules from
.roo/have to be copied to.clinerules/by hand. - You do not want to be tied to anything: see Kilo Code, which also connects through an OpenAI-compatible provider.
Choose the model by price, context and the strength index in the model table, then test it on your own tasks. Prices and the model list: GET https://api.zukko.pro/v1/models or the "Models" page in the dashboard. Payment: SBP (the payment page adds an 8.5% fee on top), cryptocurrency or MasterCard cards.
FAQ
Does Roo Code still work?
According to the official documentation, the extension was shut down on May 15, 2026 and the repository has been archived. An installed copy may keep working for a while, but there will be no updates or fixes. For new work, choose a tool that is maintained.
Can I move settings from Roo to Cline?
Not automatically. Instructions from .roo/rules-* are copied into .clinerules/, and the per-mode restrictions have to be replaced by Plan and Act plus manual approval.
Which model should I set in Cline?
claude-sonnet-5-5 is a reasonable start, glm-5.3-flash is cheaper for simple edits, and claude-opus-5-5 is for hard tasks. The strength index in the table is a guide, and the test on your own code decides.
What if requests return 404?
Almost always the /v1 suffix is mixed up. OpenAI Compatible needs it, the Anthropic provider does not.
zukko
Pay 25×
less
for Claude Code & ChatGPT
One fetch_* key and base URL api.zukko.pro — for Cursor, Claude Code, Codex, and any SDK.
Related articles
- GitHub Copilot alternatives with your own API key: what to chooseCline, Kilo Code, Continue and Cursor work with your own key. We compare where they live, how they connect to a gateway and how to estimate spend.
- Zed AI vs Cursor: a fast editor against an AI fork of VS CodeZed and Cursor approach AI differently: one builds around its own editor and the ACP protocol, the other extends VS Code. We compare them and connect your own key.
- JetBrains AI vs Cursor: what a developer on IntelliJ and PyCharm should pickStay in a JetBrains IDE with AI Assistant or move to Cursor? We compare features, own-model support and costs, with step-by-step settings.
- Cline or Cursor: what to pick if you want to pay only for tokensCline is a free extension and you pay the provider for tokens. Cursor is sold as a product. We compare payment, context control and agent features and price a month of work.