GitHub Copilot alternatives with your own API key: what to choose

A search for "github copilot alternatives" usually comes down to one of two reasons: you want to choose the model yourself, or you dislike the subscription limits and terms. There is a class of tools that work differently: you connect your own API key and pay for the tokens you actually use. Below we compare Cline, Kilo Code, Continue and Cursor, show how to connect a gateway to them, and estimate spend with an example.

We do not quote the terms and limits of Copilot, Cursor or other subscriptions: they change, so see the current plan on the service's website. This article is about the "key plus balance" model.

How the own-key model differs from a subscription

The downside is honest: spend is not fixed. An agent that reads many files makes many requests, and every one carries the accumulated context. Protection against surprises is a dollar limit on the key and watching spend in the dashboard.

What a "Copilot alternative" really means

Copilot combines several functions, and alternatives cover different ones:

• autocomplete of a line as you type;
• chat about the project code;
• agent mode: the model reads files, edits them and runs commands by itself.

Cline and Kilo Code are primarily agentic VS Code extensions. Continue covers chat, edits and autocomplete (the model for it is configured separately; see the Continue documentation). Cursor is a separate editor based on VS Code. If you specifically need line-by-line autocomplete, check it in the documentation of the extension before switching.

Cline and Kilo Code

Both connect to Zukko in the same way: provider, base URL, key, model ID. The main rule: for OpenAI Compatible the address is with /v1, for the Anthropic provider without.

Roo Code used to be on this list too, but according to its documentation the extension was shut down on May 15, 2026. Do not consider it for new installs; if you already have it, plan a move to Cline or Kilo Code.

Step-by-step instructions for each extension, including the context window fields, are in the VS Code extensions documentation↗. Exact field names in the UI change between versions, so if something differs, go by the extension's own documentation.

Continue

Continue is configured through a config.yaml file. For an OpenAI-compatible server the Continue documentation uses the provider: openai and apiBase fields:

models:
  - name: Zukko Sonnet
    provider: openai
    model: claude-sonnet-5-5
    apiBase: https://api.zukko.pro/v1
    apiKey: fetch_...

Check the file structure and model roles (chat, edit, autocomplete) in the Continue documentation, which gets updated.

Cursor

Cursor connects via Settings → Models: turn on OpenAI API Key, paste the fetch_... key, turn on Override OpenAI Base URL with https://api.zukko.pro/v1, then add a model and press Verify. Cursor has a quirk: it will not let you add a custom model with a name it already knows, so popular Zukko models have second names such as cursor-sonnet-5 and cursor-opus-55. The table of names is in the Cursor documentation↗.

Two caveats. According to the Zukko docs, overriding the provider is not available on Cursor's free plan, and some editor features may still run only on Cursor's own models. Check the current plan terms on Cursor's website. A comparison of Cursor with other editors is in Cursor alternatives↗.

What about Codex and Claude Code

If you are looking for Codex alternatives, look at terminal agents: Codex CLI, Claude Code, opencode. They are not editor extensions but separate programs in the terminal. The difference between terminal and editor tools is covered in Claude Code and Cursor↗.

Estimating spend with an example

Suppose that during a day you give the agent 20 tasks, each taking about 8 requests. Input context is 25,000 tokens per request, output 1000. That is 160 requests: input 160 × 25,000 = 4.0M tokens, output 160 × 1000 = 160 thousand.

Prices as of October 18, 2026, $ per 1M tokens; calculation without any cache discount:

This is an upper estimate. For Claude the cache discount comes only with the Anthropic format (for example Claude Code, or an extension connected through the Anthropic provider): there the repeated part of the context is billed at the cache read price, separately from fresh input and writes. For GPT cache reads are counted in both formats. The real total depends on how the extension builds its requests, so work for a day and compare with the data in the dashboard ("Usage"). Current prices: GET https://api.zukko.pro/v1/models.

How to choose

1. You only need chat and edits inside a familiar environment: look at Continue.
2. You need an agent that edits files by itself in VS Code: Cline or Kilo Code; choose by interface and convenience.
3. You want a separate editor: Cursor, after checking the terms of its plan.
4. You work in a terminal or on a server: Claude Code, Codex, opencode.

Do not decide from descriptions. Install two or three candidates, connect the same key, give them the same task and compare the result and the spend.

To start, create a key in the dashboard and give it a daily dollar limit so that experiments stay within budget.

FAQ

Does the own-key model have limits like "N requests per month"?

The gateway has no plan-based request limit: spend is bounded by the balance and by the limits you set on a key yourself (daily and monthly, in dollars). If a model hits a provider restriction you get a 429. For Copilot's own limits see GitHub's website.

Which is cheaper, a subscription or tokens?

It depends on your workload. Calculate your typical day with the formula above and compare with the subscription plan on the service's website. We do not quote subscription prices because they change.

Do I need a separate key for each extension?

It is more convenient. The dashboard then shows how much each one spent, and you can set the limit and the list of allowed models for each key separately.

Does this work from Russia?

You can top up the balance from Russia by SBP, MasterCard or crypto.