Kilo Code Claude / GPT / Chinese Model API Setup Guide
As of 2026-10-07, you connect Kilo Code to a Claude API like this: Settings → Providers tab → scroll to the bottom → Custom provider, set Provider API to Anthropic Messages, Base URL https://api.qcode.cc/api (no trailing slash), your cr_ key, and type the model ID by hand. For GPT and Chinese models, pick OpenAI Compatible with Base URL https://api.qcode.cc/openai/v1. This page collects the steps, model IDs and common errors from the QCode docs and the official Kilo docs.
Updated 2026-10-07
Four things to know before you start
Provider API for Claude
Base URL https://api.qcode.cc/api. The QCode docs recommend stopping at /api and letting the extension append /v1/messages.
GPT and Chinese models
Base URL https://api.qcode.cc/openai/v1, the same cr_ key, model IDs typed by hand, e.g. gpt-6-sol or glm-5.2.
Where the setting lives
Settings (gear) → Providers tab → bottom of the list → Custom provider. The QCode docs note it is no longer the Roo Code style API Provider dropdown.
Latest stable (2026-10-01)
The newest stable release on GitHub as of 2026-10-07; it added a manual Fetch Models button to the Custom Provider dialog. A v7.8.7 pre-release followed on 2026-10-07.
How Kilo Code connects to a custom API
As of 2026-10-07, connecting Claude in Kilo Code means adding a Custom provider at the bottom of the Providers tab: Provider API Anthropic Messages, Base URL https://api.qcode.cc/api, your cr_ key, and a hand-typed model ID. For GPT and Chinese models, use the same dialog with OpenAI Compatible and https://api.qcode.cc/openai/v1. The official Kilo README describes it as an open source coding agent for VS Code, JetBrains and the CLI. The QCode docs say it is a fork of Roo Code, which in turn forked Cline, but the interfaces have diverged, so steps from those two tools do not carry over.
Is Kilo Code still maintained? (as of 2026-10-07)
As of 2026-10-07, Kilo Code is actively updated: the GitHub repository Kilo-Org/kilocode is not archived, uses the MIT license and last received a push on 2026-10-07 (UTC). The latest stable release is v7.8.3 from 2026-10-01 (UTC); its notes say Add a manual Fetch Models button to the Custom Provider settings dialog. A v7.8.7 pre-release appeared on 2026-10-07, and JetBrains plugin 7.1.9-rc.1 on 2026-09-28 (UTC). On ownership, the official Kilo docs read Kilo has been acquired by Anaconda. The QCode docs say the custom endpoint entry point has been redesigned; this page follows the new interface.
Release and docs timeline
The QCode Kilo Code guide and endpoints page are updated, with steps written for the new Custom provider interface (the Kilo guide was last verified on 2026-09-18).
Kilo Code v7.8.3 ships as a stable release (UTC); its notes list a manual Fetch Models button in the Custom Provider dialog.
Kilo Code publishes the v7.8.7 pre-release (UTC) and the repository receives pushes the same day; v7.8.3 remains the newest stable release.
Confirmed vs not documented
Confirmed (verbatim in the docs)
All of the following can be matched word for word in the QCode docs, the official Kilo docs or on GitHub: the entry point is Settings → Providers tab → Custom provider at the bottom; the dialog fields are Provider ID, Display name, Provider API, Base URL, API key, Models and optional Headers, and after Submit the models appear in the model picker; Kilo states that Anthropic models use Anthropic Messages, while OpenAI and xAI use OpenAI Responses; Claude uses https://api.qcode.cc/api, GPT and Chinese models use https://api.qcode.cc/openai/v1, both with the same cr_ key; Claude under OpenAI Compatible returns model_not_available_on_endpoint; v7.8.3 shipped on 2026-10-01.
Not documented or not verified
Four points this page leaves open: 1) whether models are auto-fetched under Anthropic Messages: Kilo says it auto-fetches when the Base URL exposes an OpenAI-compatible models endpoint; not tested here, so this page assumes you type model IDs. 2) whether an extra trailing slash causes a 404 right away: the QCode Kilo guide says this was not tested; the endpoints page says in general it produces //v1/messages and a 404. 3) the JetBrains plugin steps are not in the QCode docs; this page covers the VS Code extension only. 4) the official Kilo docs name the config file kilo.json in one place and kilo.jsonc in another; this page gives no file-based setup.
How to choose
Anthropic Messages vs OpenAI Compatible
Go by model family: Claude only works with Anthropic Messages (https://api.qcode.cc/api), because the QCode OpenAI endpoint does not accept Claude models; GPT uses OpenAI Compatible (https://api.qcode.cc/openai/v1). GLM, Kimi, DeepSeek and Qwen work on both endpoints according to the QCode endpoints page, and the Kilo guide shows the OpenAI Compatible setup.
Kilo Code vs Roo Code / Cline
Same code lineage, different interfaces: Kilo uses the Custom provider dialog in the Providers tab, while Roo Code and Cline still use an API Provider dropdown. The QCode docs warn that the steps do not carry over; follow the setup page for the tool you actually use.
Six steps to Claude (plus GPT and Chinese models)
1) Open the settings (gear icon) in the Kilo Code sidebar, go to the Providers tab, scroll to the bottom and click Custom provider. 2) Enter a unique Provider ID (any name works; the docs use qcode as the example) and any Display name. 3) Set Provider API to Anthropic Messages, Base URL to https://api.qcode.cc/api (no trailing slash), and API key to your QCode key starting with cr_. 4) Type the model ID under Models, e.g. claude-sonnet-5; claude-sonnet-5-5 and claude-opus-5-5 are typed the same way. Headers can stay empty. 5) Click Submit; the models appear in the model picker. Send one message in the chat to verify. 6) For GPT and Chinese models, add another Custom provider: Provider API OpenAI Compatible, Base URL https://api.qcode.cc/openai/v1, the same cr_ key, and models such as gpt-6-sol, gpt-5.6-terra or glm-5.2.
On QCode
On QCode, one cr_ key works across protocols, and the request path decides the protocol: claude-sonnet-5, claude-sonnet-5-5 and claude-opus-5-5 go through Anthropic Messages at https://api.qcode.cc/api; gpt-6-sol, gpt-5.6-terra, glm-5.2, deepseek-v4.1-flash and kimi-k3 go through OpenAI Compatible at https://api.qcode.cc/openai/v1. The same key works on all three access domains: us.qcode.cc serves North America and Europe, and for networks in mainland China the docs suggest switching the domain to asia.qcode.cc with the same key. Billing is per token; see /models for each model's price. Request logs are visible at probe.qcode.cc after entering your key.
Frequently asked questions
Which Provider API should I pick for Claude in Kilo Code?
Anthropic Messages. In the Custom provider dialog, set Base URL to https://api.qcode.cc/api, enter your cr_ key and type a model ID such as claude-sonnet-5 under Models. The official Kilo docs say Anthropic models use Anthropic Messages, and the QCode docs warn against calling Claude through OpenAI Compatible.
What does model_not_available_on_endpoint mean?
You picked the wrong protocol; the key is fine. With Claude under OpenAI Compatible, QCode returns an error like Model 'claude-sonnet-5' is not available on this endpoint. The docs state this check runs before authentication, so it appears even with an invalid key. Switch to Anthropic Messages with Base URL https://api.qcode.cc/api.
Should the Base URL include /v1 or a trailing slash?
For Anthropic Messages, stop at /api (https://api.qcode.cc/api) and let the extension append /v1/messages; for OpenAI Compatible, use https://api.qcode.cc/openai/v1 as the QCode docs show. Leave off the trailing slash in both cases: the QCode Kilo guide says it did not test whether an extra slash causes a 404, and the endpoints page says in general it produces //v1/messages and a 404.
claude-sonnet-5-5 is not in the model dropdown. What now?
Type the model ID yourself. The QCode docs say that when the built-in list does not include an ID, you enter it in the custom model ID field; the official Kilo docs also say you can still enter model IDs manually if auto-detection fails. After fixing the Base URL or API key, click Fetch models next to the model list to retry. If you see Model Not Found, check that the model ID is valid.
The key is set but the connection still fails. What should I check?
First send a POST test to https://api.qcode.cc/api/v1/messages with curl in a terminal: if the JSON response contains content, the endpoint is fine and the problem is in the extension settings. With Invalid API key, make sure the key starts with cr_ and has no leading or trailing spaces. Do not judge by opening https://api.qcode.cc/api directly: it returns an HTML intro page (HTTP 200), which is neither an error nor proof that the path works.
Is Kilo Code still maintained?
Yes. As of 2026-10-07, the GitHub repository Kilo-Org/kilocode is not archived and last received a push on 2026-10-07 (UTC); the latest stable release v7.8.3 shipped on 2026-10-01, and a v7.8.7 pre-release followed on 2026-10-07.
Sources
Setup steps, Provider API types and addresses: the QCode docs Kilo Code guide (updated 2026-09-25) and the endpoints and API formats page (updated 2026-09-25, covering which models use which endpoint, the exact error text and the three access domains), both retrieved on 2026-10-07. Kilo interface and fields: the official Kilo docs (kilo.ai/docs) OpenAI Compatible and Anthropic pages, retrieved the same day. Maintenance status and releases: the GitHub repository Kilo-Org/kilocode README, repository metadata and Releases, retrieved the same day.
Connect QCode in Kilo Code
One cr_ key: Claude through Anthropic Messages, GPT and Chinese models through OpenAI Compatible. Billed per token; see /models for each model's price.
Related reading
Cline OpenAI-Compatible API Guide
Three fields in Cline to connect it to QCode.
Cherry Studio Claude API setup guide
Claude via the Anthropic type, GPT and Chinese models via the OpenAI type.
Claude Code custom endpoint setup
The two environment variables, the settings.json form and apiKeyHelper.
Steps and quotes on this page were checked on 2026-10-07 against the QCode docs, the official Kilo docs and the GitHub release notes; the official pages prevail, and upstream changes may happen without notice. Extension labels change between versions, so go by the version you have; model availability is listed on /models.