Kilo Code and Roo Code: how they differ from Cline

Roo Code is no longer maintained. Its README says the extension was shut down on 15 May 2026, and the GitHub repository is archived. So the question "how do Kilo Code and Roo Code differ from Cline" now comes down to two living extensions and one footnote about the third. Both Cline and Kilo Code accept a custom OpenAI-compatible endpoint. To use Claudexia you need four values: the address https://api.claudexia.tech/v1, a key that starts with sk_cdx_, a model ID and a context size.

Below you will find the comparison, setup for each extension, model choice, token spending and the errors people hit. Project status is as of 10 November 2026, the day we opened the README and documentation of Cline, Roo Code and Kilo Code. We did not run the extensions on a clean machine for this piece, so wherever a settings screen comes from a project's docs rather than from our own run, we say so.

Cline, Kilo Code and Roo Code side by side

Licenses, modes and platforms come from each project's README. Kilo's origin is stated in its own blog post from April 2025: the team forked Roo Code, which was itself a fork of Cline. The story has moved on since. According to Kilo's docs the current extension is built on the Kilo CLI, and the README says the CLI is a fork of OpenCode. Calling Kilo a lightly modified Cline no longer describes the code.

The projects also handle money differently. The code of both living extensions is open, and you pay for tokens either through the project's own account or straight to the model provider. Cline's docs describe three ways in: signing in to its account, a subscription, or your own key. Kilo promises provider pricing with no markup and more than 500 models through its account, and says your own key works with more than 30 providers. The last path is the one that matters here, because the Claudexia key is yours to enter.

A note on Kilo's ownership. On 15 July 2026 Anaconda announced that it is acquiring Kilo Code, and Kilo's documentation pages now carry a banner saying the deal is done. The announcement says Kilo stays available to developers and teams. It does not say what happens to the license or pricing.

What happened to Roo Code

The Roo Code README has a Disclaimer section: the extension was shut down on 15 May, and anyone looking for a replacement is pointed to Cline and ZooCode. GitHub lists RooCodeInc/Roo-Code as archived, with its last push on the same date. The project's documentation site is still up and carries the same shutdown notice.

ZooCode is described in its README as a fork started by the Roo Code community. It lists Apache 2.0, VS Code as the main target, the same Code, Architect, Ask, Debug and custom modes, and a migration guide from Roo. We have not installed it and cannot speak for its stability.

If Roo Code is already in your editor, its provider fields match Cline's. Roo's docs name them API Provider, Base URL, API Key, Model, plus a Model Configuration block with a Context Window field. You can fill them in using the Cline steps below. The project will not get fixes, though, so for a fresh install we would pick Cline or Kilo Code.

Two connection formats, any extension

All three projects share one logic, and our extension guide describes it: the provider type is either OpenAI Compatible or Anthropic. The choice decides whether the address ends in /v1.

OpenAI Compatible covers Claude, GPT and the other catalog models under one key. The Anthropic mode keeps the Claude format and is needed less often. In Cline it is the Anthropic provider with the Use custom base URL option, in Kilo it is the Anthropic Messages choice in the Provider API list. We have not checked how Kilo treats the address in that mode, so the steps below use OpenAI Compatible.

Step 1. Get a Claudexia key

Sign up, top up your balance and create a key in the API Keys section of the dashboard. The minimum top-up is $1, and the top-up dialog shows exact limits for your payment method. Cards and crypto (USDT, BTC, ETH) both work, and in Russia SBP and Russian cards do too. The key looks like sk_cdx_..., and one key covers Claude and GPT.

Do you need it at all? If you already run Kilo through its own account with credits, you can keep doing that. A Claudexia key makes sense when you want Claude models without an Anthropic account, one balance shared by Cline, Claude Code and Cursor, or a payment route that works from where you live.

Step 2. Set up Cline

Open the Cline panel, click the gear and choose OpenAI Compatible in the API Provider field. Fill in the fields as our guide has them:

API Provider:        OpenAI Compatible
Base URL:            https://api.claudexia.tech/v1
API Key:             sk_cdx_YOUR_KEY
Model ID:            claude-sonnet-5
Context Window Size: 1000000

In Cline's docs the context size and output limit sit in a block called Model Configuration. Next to them are switches for image support and per-token prices, none of which you need to get going. Write the model ID without the [1m] suffix: it exists only for Claude Code, and every other program takes the name as it appears in the catalog.

If you want Claude only and prefer the Anthropic format, pick the Anthropic provider, turn on Use custom base URL and paste the address without /v1. A longer walkthrough with a curl check is in Setting up Cline in VS Code with a custom base URL.

Step 3. Set up Kilo Code

The screen differs from Cline's here, and this description comes from Kilo's documentation, not from our own run. Open settings with the gear in the extension panel, go to the Providers tab, scroll to the bottom and click Custom provider. Fill in the dialog:

• Provider ID: any unique identifier, for example claudexia.
• Display name: the label shown in the interface.
• Provider API: OpenAI Compatible.
• Base URL: https://api.claudexia.tech/v1.
• API key: your sk_cdx_... key.
• Models: a model ID such as claude-sonnet-5.

Kilo can fetch the model list on its own when the address serves one. Claudexia has GET /v1/models, but whether the dialog picks it up is something we have not tested, so type the ID by hand if the list stays empty.

The dialog has no context size field. Kilo's docs have you set limits in kilo.jsonc, and the wording is firm: for custom and local models, always set limit.context and limit.output. The Local Config and Global Config buttons in the settings header open the file. The global file lives at ~/.config/kilo/kilo.jsonc, on Windows at C:\Users\<username>\.config\kilo\kilo.jsonc, and the project one sits in the repository root or in .kilo/kilo.jsonc.

If you are on an older build whose settings show an API Provider dropdown, like Cline's, fill the fields as in step 2. That is the screen our own guide describes.

Context size: why the field is mandatory

When the context size is not set, the behavior depends on the extension. Per its documentation, Kilo turns off automatic context management without limit.context and limit.output. The other extensions fall back to a default, and our guide warns that the model may be treated as 128K. claude-sonnet-5 has a 1M context, so set it explicitly.

Still, do not paste 1000000 for every model. In the Premium channel claude-haiku-4.5 has a 200K context and should get 200000. Look up the numbers for each model in the models table: the Max input column goes into context, Max output goes into output.

Which model to pick

The rule is plain: start with Sonnet and bring in Opus when Sonnet fails on a specific task. How to test that without overspending is covered in Picking the model for the task without overpaying. Take exact IDs from the catalog, since a typo ends in Model not found.

How agent modes spend tokens

An agent works in a loop: it reads files, proposes an edit, runs a command and reads the output. Each step goes to the API together with the history so far, so a long task costs noticeably more than a short one. Two habits move the bill the most.

First, do not turn on auto-approve right away. Per Cline's README, every file edit and terminal command waits for your confirmation until you enable auto-approve. While you watch each step, an agent stuck in a loop is visible at once, not on the bill at the end of the day. Second, start a new task when the old one has grown, instead of dragging one conversation through half a day. The other techniques, from matching the model to the job to capping the reply length, are collected in How to cut your Claude Code bill threefold, and they apply to agent extensions in the same way.

Which one to choose

For plugging in your own key there is almost no difference between Cline and Kilo Code: both want an address, a key and a model ID. Pick by other things. If you want Plan and Act modes and a desktop app, take Cline. If you want four agents, a cloud run in the browser and sign-in to an account with hundreds of models, take Kilo Code. Roo Code is not an option for a new project: it is shut down.

If you work in Cursor or Zed you do not need these extensions, since each editor has its own address setting: see Cursor and Zed with Continue. For Claude in a terminal and in VS Code there is a separate Anthropic extension, covered in Claude Code in VS Code.

Common errors

A 401 almost always means the key was pasted with a stray space, got cut off in copying or does not start with sk_cdx_. Copy it from the dashboard again.

No answer and no error: check the /v1 in the address. OpenAI Compatible mode needs it, Anthropic mode must not have it.

Model not found: compare the ID with the catalog letter for letter, with no spaces and no [1m].

History gets cut early: the context size is missing or holds a value from another model.

Tool-call errors. Roo's docs say the model must support OpenAI-style tool calling, and providers that implement it only partly cause such failures. For Claude and GPT from the catalog this is usually fine, but on obscure models test the task on a tiny example first.

FAQ

Is Roo Code still maintained?

No. The README says the extension was shut down on 15 May 2026, the repository is archived, and Cline and ZooCode are named as replacements. Whether an already installed copy keeps answering on your machine, we have not checked.

What is the difference between Kilo Code and Cline?

Per the READMEs: Kilo is MIT licensed and Cline is Apache 2.0. Cline has Plan and Act modes, Kilo has Code, Plan, Ask, Debug and custom agents. Kilo also runs as a cloud agent in the browser, and its model access goes through its account or your own key. For plugging in your own key both ask for the same thing: address, key, model ID.

How do I use my own API key in Kilo Code?

Settings, the Providers tab, then Custom provider at the bottom. Choose OpenAI Compatible as Provider API, put https://api.claudexia.tech/v1 in Base URL, add the key and a model ID. Set the context and output limits in kilo.jsonc.

Does Kilo Code work in JetBrains IDEs like PyCharm and IntelliJ IDEA?

A Kilo Code plugin exists in the JetBrains Marketplace, and the README tells you to search for it under Settings, Plugins in any JetBrains IDE. How custom endpoint entry looks inside the plugin, we have not checked.

Is Kilo Code free?

The code is open under MIT, and the README promises provider pricing with no markup and sign-in without keys. We did not check the limits on free models. With your own key you pay for tokens to whoever issued the key.

Should the base URL end in /v1?

In OpenAI Compatible mode, yes: https://api.claudexia.tech/v1. In Anthropic mode, no: https://api.claudexia.tech.