# NextChat with QCode: Windows 11, one screenshot per step

Follow the 9 steps below and you will have NextChat desktop installed on Windows 11, connected to QCode, and chatting with Claude and GPT.


**Short answer**: Open `Settings` in the NextChat desktop app, scroll to the `Model Provider` card, set `Endpoint address` to `https://api.qcode.cc/openai`, `API Key` to your key starting with `cr_`, and `Custom model name` to `gpt-6.2-sol@OpenAI`; go back to the chat page, select `gpt-6.2-sol (OpenAI)` and you can chat.

| Item | Fill in |
|---|---|
| Where to configure | `Settings → Model Provider` |
| Endpoint address (OpenAI-compatible) | `https://api.qcode.cc/openai` |
| Endpoint address (Anthropic / Claude) | `https://api.qcode.cc/api` |
| Custom model name | `gpt-6.2-sol@OpenAI,claude-opus-5-5@Anthropic` |
| Key format | `cr_xxxxxxxx` |

Checked on 2026-11-22 · NextChat 2.15.8 · Windows 11


NextChat (formerly ChatGPT-Next-Web) is an open-source desktop chat client that runs entirely on your own machine once installed, with no sign-in to any model vendor's account. Point the endpoint address at QCode in the settings and one key lets you call both Claude and GPT models.

> **Tip**: The NextChat desktop app stops at **v2.15.8**, and later versions no longer ship a Windows installer. The setup for this version still works today — just follow the steps below.


- **Applies to**: Windows 10 / 11 (64-bit)
- **Software version**: NextChat desktop 2.15.8
- **Screenshots updated**: 2026-11
- **Time needed**: 10 minutes
- **You will need**: a QCode API key starting with `cr_`; a network that can open `github.com` and `api.qcode.cc`



- The PC runs Windows 10 / 11 (64-bit) and the browser opens normally
- You have already signed up at QCode.cc and copied your `cr_` API key from the [Dashboard](https://qcode.cc/en/dashboard). No account yet? [Sign up for QCode](https://qcode.cc/en/auth/sign-up?utm_source=docs.qcode.cc&utm_medium=tut_before&utm_campaign=knowledge_base&src=docs-ide-nextchat-before) first, pick a plan on the [pricing page](https://qcode.cc/en/pricing?utm_source=docs.qcode.cc&utm_medium=tut_before&utm_campaign=knowledge_base&src=docs-ide-nextchat-before), then create the key in the [Dashboard](https://qcode.cc/en/dashboard).
- No OpenAI or Anthropic account is needed, and you do not sign in to any model vendor account — the NextChat desktop app works as soon as you open it
- The PC can reach `github.com` (to download the installer) and `api.qcode.cc` (to send requests)
- Set aside about 10 uninterrupted minutes


## Download and install

The NextChat desktop installer is published only on GitHub Releases. The last version with a Windows installer is **v2.15.8** (released 2024-11-11); the newer v2.16.0 / v2.16.1 provide source code only and have no Windows installer, so v2.15.8 is the one to install.

- **Recommended entry**: <https://github.com/ChatGPTNextWeb/NextChat/releases/tag/v2.15.8>
- **Installer file name**: `NextChat_2.15.8_x64-setup.exe` (about 6.1 MB, NSIS installer — beginners pick this one)
- **MSI build**: `NextChat_2.15.8_x64_en-US.msi`, suited to company-wide deployment
- **Do not download**: `Source code (zip)` / `Source code (tar.gz)` are source code; `NextChat_2.15.8_x64-setup.nsis.zip` is a zipped installer, not a portable build

Pick the download route for your region:

**Other regions**

Open the release page <https://github.com/ChatGPTNextWeb/NextChat/releases/tag/v2.15.8> directly, scroll down to the `Assets 16` list and click `NextChat_2.15.8_x64-setup.exe` to download. When the download finishes, the browser offers to run the installer.


**Mainland China**

GitHub sometimes refuses connections, or the download speed swings up and down. First click `NextChat_2.15.8_x64-setup.exe` on the release page directly; if the page will not open or the progress stalls, refresh the page and click download once more. The installer is only about 6 MB, so wait for the browser to finish downloading before you run it.


## Follow the screenshots: Windows, one step at a time

The 9 steps below each do one thing: first connect the OpenAI-compatible endpoint on QCode (to use GPT), then switch to the Anthropic endpoint (to use Claude). The orange numbers on the screenshots match the ①②③ in the text one to one.



### Open the download page and get the Windows installer

Open <https://github.com/ChatGPTNextWeb/NextChat/releases/tag/v2.15.8> in the browser, scroll down to the `Assets 16` list and click the `NextChat_2.15.8_x64-setup.exe` that ① points at. The list loads only after you scroll down a little; the file name shows up after a moment.

![NextChat v2.15.8 GitHub Release page opened, scrolled down to the Assets list, clicking the Windows installer NextChat_2.15.8_x64-setup.exe (1)](https://docs.qcode.cc/static/images/tut/nextchat/202611/zh/01-releases-download.webp)


Done right you will see: the browser starts downloading a 6.1 MB file named `NextChat_2.15.8_x64-setup.exe`, with the `.exe` extension.


**Stuck? No Windows installer in the Releases list**

First check that you opened the **v2.15.8** page: v2.16.0 and v2.16.1 have no Windows desktop installer, only source code. Once the expanded list says `Assets 16` rather than `Source code` above it, you will see `NextChat_2.15.8_x64-setup.exe`.




### Run the installer and keep the default path

Double-click the `NextChat_2.15.8_x64-setup.exe` you just downloaded and click `Next >` on the welcome page. On `Choose Install Location`, click the `Next >` that ① points at to keep the default path — it installs into the current user's `AppData\Local\NextChat` folder and needs no administrator rights. Then click `Install` on `Choose Start Menu Folder` and `Finish` on the last page.

![](https://docs.qcode.cc/static/images/tut/nextchat/202611/zh/02-installer-location.webp) on the welcome page and reaching the install location page: by default it installs into the current user's AppData\Local\NextChat folder (no administrator rights needed), so keep the default and click Next > (1)" width="551" height="438" loading="lazy">


Done right you will see: the `Destination Folder` field shows `C:\Users\<your username>\AppData\Local\NextChat`, clicking `Next >` moves on to `Choose Start Menu Folder`, and no administrator password window appears along the way.


**Stuck? Windows shows "Windows protected your PC"**

This is the normal prompt when the installer has no code-signing certificate. Click `More info` in that window, then `Run anyway` and the installation continues. If a company policy blocks it on a work PC, ask IT to allow it.




### First launch: open `Settings` at the bottom left

NextChat starts automatically after installation: no sign-in, straight to the chat page. If it does not open by itself, double-click the NextChat icon on the desktop. Click the gear icon at the bottom left where ① points to open `Settings`.

![After installation the NextChat desktop app opens by itself and goes straight to the main chat interface (no sign-in, no blocking popup); click the gear icon at the bottom left to open Settings (1)](https://docs.qcode.cc/static/images/tut/nextchat/202611/zh/03-first-launch.webp)


Done right you will see: the whole right column becomes the `Settings` page, with a `Language` dropdown at the top and a small `当前版本: v2.15.8` (current version) line.




### Scroll down to the `Model Provider` card

The settings page is one long page from top to bottom, with no tabs. Scroll down with the mouse wheel and find the `Model Provider` card: ① is the `Model Provider` dropdown, keep the default `OpenAI`; ② is the `Endpoint address` input, with `API Key` and `Custom model name` below it.

![Scrolled down the settings page to the Model Provider card: desktop v2.15.8 has no separate Custom Interface switch and shows Model Provider directly (default OpenAI) (1), plus the Endpoint address, API Key and Custom model name inputs (2)](https://docs.qcode.cc/static/images/tut/nextchat/202611/zh/04-settings-provider.webp)


Done right you will see: the card lists `Model Provider`, `Endpoint address`, `API Key`, `Custom model name` and `模型 (model)` from top to bottom, and `OpenAI` is already the selected provider.


**Stuck? There is no "Custom Interface" switch in Settings**

NextChat desktop v2.15.8 has no `Custom Interface` switch and does not need one. Changing `Endpoint address` in the card is all it takes to point the app at QCode.




### Fill in the endpoint address, API key and custom model name

First put `https://api.qcode.cc/openai` into the `Endpoint address` field at ①, with no `/v1` at the end — NextChat appends `v1/chat/completions` itself, and one extra segment turns into a duplicated path. Then paste the key starting with `cr_` into `API Key` at ② (it shows as asterisks after typing). Finally put `gpt-6.2-sol@OpenAI` into `Custom model name` at ③ and click a blank spot outside the card to save.

![Filling in under the OpenAI provider: endpoint address https://api.qcode.cc/openai (no /v1, NextChat appends v1/chat/completions itself) (1), the QCode API key (masked) (2), and the custom model name gpt-6.2-sol@OpenAI (3)](https://docs.qcode.cc/static/images/tut/nextchat/202611/zh/05-settings-openai-config.webp)

**Need an API key?**

If you do not have a QCode account yet, [Sign up](https://qcode.cc/en/auth/sign-up?utm_source=docs.qcode.cc&utm_medium=tut_key_step&utm_campaign=knowledge_base&src=docs-ide-nextchat-key) first. If you prefer to pay by usage, buy a [Pay-as-you-go Pack](https://qcode.cc/en/pay-as-you-go?utm_source=docs.qcode.cc&utm_medium=tut_key_step&utm_campaign=knowledge_base&src=docs-ide-nextchat-key) first, then create the key in the [Dashboard](https://qcode.cc/en/dashboard). Copy the whole string that starts with `cr_` and paste it into `API Key`, with no spaces before or after. You can pay by credit/debit card (Visa / Mastercard), Apple Pay / Google Pay, or cryptocurrency — see [Payment methods](/docs/reference/payment-methods).



Done right you will see: the `Endpoint address` field shows `https://api.qcode.cc/open…` (the field is narrow and cuts off the end — that is normal styling), `API Key` turns into a row of asterisks, and `Custom model name` holds `gpt-6.2-sol@OpenAI`.




### Select the model and check it on the chat page

Below the same card, find the `模型 (model)` dropdown, choose `gpt-6.2-sol`, then click `Close` at the top right to go back to the chat page. Then click the robot icon in the toolbar above the input box to open the model picker and confirm that `gpt-6.2-sol (OpenAI)` at ① is pinned to the top and already selected.

![Selecting gpt-6.2-sol in the 模型 (model) dropdown below the settings and closing Settings; clicking the model name in the toolbar above the input box to open the model picker and confirming gpt-6.2-sol (OpenAI) is selected (1)](https://docs.qcode.cc/static/images/tut/nextchat/202611/zh/06-select-model-openai.webp)


Done right you will see: the custom `gpt-6.2-sol (OpenAI)` at the top of the model list, and the model name above the input box reads `gpt-6.2-sol` as well.


**Stuck? The model you just added is not in the dropdown**

Go back to `Custom model name` in Settings and check the spelling character by character: it must be written as `model-name@provider`, for example `gpt-6.2-sol@OpenAI`. After editing, click a blank spot outside the card to save, then go back to the chat page and open the model picker again.




### Send your first sentence and watch the reply stream in

Type `Hello, introduce yourself in one sentence.` into the input box at the bottom where ① points and press `Enter` to send, or click `Send` at the bottom right. A few seconds later the reply appears where ② points, with the text streaming out character by character.

![Typing Hello, introduce yourself in one sentence. into the bottom input box and pressing Enter to send (1); a few seconds later a normal reply from gpt-6.2-sol arrives (2), which means the QCode OpenAI-compatible endpoint is connected](https://docs.qcode.cc/static/images/tut/nextchat/202611/zh/07-chat-gpt-reply.webp)


Done right you will see: your question and one complete reply, with `gpt-6.2-sol` labeled above the reply, and a new conversation title appears in the session list on the left.


**Stuck? Sending turns into a red error, or nothing happens at all**

First check whether `Endpoint address` has an extra `/v1` or `/v1/chat/completions`; the correct value stops at `https://api.qcode.cc/openai`. If it says the key is invalid, go back to the Dashboard and copy the `cr_` key once more, and paste it without spaces or quotes.




### Switch to the Anthropic endpoint and add the Claude model

To use Claude as well, change `Model Provider` at ① to `Anthropic`. Then clear the `Endpoint address` at ② and enter `https://api.qcode.cc/api` (no `/v1`); put the same QCode key into `Interface key` below, and leave `接口版本 (claude api version)` at its default `2023-06-01`. Finally change `Custom model name` at ③ to `gpt-6.2-sol@OpenAI,claude-opus-5-5@Anthropic`, separating the two models with a comma.

![To use Claude models, switching Model Provider to Anthropic (1), filling the endpoint address with https://api.qcode.cc/api (2), entering the same QCode API key (masked) as the interface key, and changing the custom model name to gpt-6.2-sol@OpenAI,claude-opus-5-5@Anthropic (3)](https://docs.qcode.cc/static/images/tut/nextchat/202611/zh/08-settings-anthropic-config.webp)


Done right you will see: `Model Provider` shows `Anthropic`, `Endpoint address` holds `https://api.qcode.cc/api`, `Interface key` shows as asterisks, and `接口版本 (claude api version)` is still `2023-06-01`.


**Stuck? The GPT model disappeared after switching to Anthropic**

That happens because the whole custom model name line was replaced. Write it as `gpt-6.2-sol@OpenAI,claude-opus-5-5@Anthropic`, with exactly one comma and no spaces between the two parts, and both models show up in the list again.




### Start a new chat and ask Claude something

Click `新的聊天` (New chat) on the left, then `直接开始` (Start directly) to open a blank conversation. Click the robot icon above the input box, choose the pinned `claude-opus-5-5 (Anthropic)`, then type the same `Hello, introduce yourself in one sentence.` and send it. The reply that appears at ① comes from `claude-opus-5-5`, which means the Anthropic endpoint is connected too.

![After closing Settings, clicking 新的聊天 (New chat) → 直接开始 (Start directly), clicking the model icon above the input box to select claude-opus-5-5 (Anthropic) at the top, and getting a normal reply from claude-opus-5-5 after sending a test message (1), which means the QCode Anthropic endpoint is connected as well](https://docs.qcode.cc/static/images/tut/nextchat/202611/zh/09-chat-claude-reply.webp)


Done right you will see: `claude-opus-5-5` labeled above the reply, a complete self-introduction in the body, and no red error.


**Stuck? The Claude model errors out or says the model does not exist**

Claude only works through the Anthropic endpoint: first confirm `Model Provider` is `Anthropic` and `Endpoint address` is `https://api.qcode.cc/api`, then confirm the custom model name contains `claude-opus-5-5@Anthropic`. Sending a Claude model to the OpenAI-compatible address is rejected.




## Verify it works

Check the four things below to confirm NextChat is solidly connected to QCode:

- The `Endpoint address` in Settings does not revert to `https://api.openai.com` or `https://api.anthropic.com` after saving;
- The model picker on the chat page lets you select the custom model names, for example `gpt-6.2-sol (OpenAI)` and `claude-opus-5-5 (Anthropic)`;
- Sending a sentence returns a streamed reply — not a red error, and no `Request failed`;
- Closing NextChat and opening it again keeps both the settings and the chat history (the data lives on this machine).


### All done

You have installed the NextChat desktop app and connected both the GPT and Claude model families through QCode. From here you can:

- [Endpoints and API formats](/docs/getting-started/endpoints-and-api-paths)
- [Chinese models](/docs/usage/cn-models)
- [Troubleshooting guide](/docs/reference/troubleshooting)
- Next: compare the plans on the [pricing page](https://qcode.cc/en/pricing?utm_source=docs.qcode.cc&utm_medium=tut_done&utm_campaign=knowledge_base&src=docs-ide-nextchat-done)
- Go back to your [Dashboard](https://qcode.cc/en/dashboard) to check usage and balance


## Common errors

| Error message | Why | What to do |
|---|---|---|
| `除默认地址外，必须包含 http(s)://` | This hint appears under the `Endpoint address` line when the address in the field is missing `https://` or is only a domain name | Enter the full `https://api.qcode.cc/openai` (OpenAI-compatible) or `https://api.qcode.cc/api` (Anthropic), starting with `https://` and with no `/` at the end |
| `样例: https://api.anthropic.com` | After switching to Anthropic, `Endpoint address` still holds the official example address and requests never reach QCode | Clear the line and change it to `https://api.qcode.cc/api`; use the same `cr_` key and keep the interface version at `2023-06-01` |
| `发现新版本: v2.16.1` | The desktop app stops at v2.15.8, and the top of the settings page quietly notes that a newer version exists | Ignore it: it does not interrupt anything and does not affect saving `Endpoint address` or `API Key`; finish the QCode setup below and chatting works normally |
| `ERR_CONNECTION_TIMED_OUT` | The network dropped while opening the GitHub download page or downloading the installer, so the download never finished | Wait for the download to finish or download once more; after installing, still fill `https://api.qcode.cc/openai` into `Endpoint address` — do not change the endpoint address because of a download error |

## FAQ

### What happens when my key expires?

Two cases:

- A **personal plan** key cannot be renewed when it expires — you have to reopen a plan and switch to a new key;
- For a **pay-as-you-go pack**, buy another pack after the key expires and the same key starts working again automatically.

After switching to a new key, go back to `Settings → Model Provider`, clear the contents of `API Key` (called `Interface key` under Anthropic) and paste the new key starting with `cr_`.

### Why is there no `/v1` in the endpoint address?

NextChat appends `v1/chat/completions` to the endpoint address itself. So the OpenAI-compatible address only needs to go as far as `https://api.qcode.cc/openai`, and Claude's Anthropic address only as far as `https://api.qcode.cc/api`. Adding `/v1` produces a duplicated path and the request fails straight away.

### What are `@OpenAI` and `@Anthropic` in the custom model name?

That is how NextChat specifies which provider a model goes through. With only `gpt-6.2-sol`, the list shows `gpt-6.2-sol (gpt-6.2-sol)`; written as `gpt-6.2-sol@OpenAI` it shows `gpt-6.2-sol (OpenAI)`, and Claude models can be pinned to the Anthropic endpoint the same way. Separate multiple models with commas.

### The desktop app keeps saying a new version is available — do I need to upgrade?

No. The NextChat desktop app stops at 2.15.8 and later versions no longer ship a Windows installer, so you can ignore that notice for good. If you want newer features, use the web version or deploy it yourself; the connection setup is the same as here.

### Where are these settings stored? Will they survive a new computer?

The config files, keys and chat history all live on this machine and are kept when you close and reopen the app. After moving to a new computer or reinstalling the system you need to fill in `Endpoint address`, `API Key` and `Custom model name` once more.

## Connection overview

| Item | Details |
|---|---|
| Available models | Claude ✅ (through the Anthropic endpoint) · GPT ✅ (through the OpenAI-compatible endpoint) |
| Where to configure | The `Model Provider` card on the NextChat settings page |
| OpenAI-compatible address | `https://api.qcode.cc/openai` (no `/v1`) |
| Anthropic address | `https://api.qcode.cc/api` (no `/v1`) |
| Custom model name syntax | `model-name@provider`, several separated by commas, for example `gpt-6.2-sol@OpenAI,claude-opus-5-5@Anthropic` |
| Key format | `cr_xxxxxxxx` |
| Applies to | NextChat desktop 2.15.8 (Windows 10 / 11 64-bit) |

## Available models

| Model | How to write it | Notes |
|---|---|---|
| Claude Opus 5.5 | `claude-opus-5-5@Anthropic` | Recommended Claude default, strongest overall capability |
| Claude Haiku 5.5 | `claude-haiku-5-5@Anthropic` | Lightweight alternative, faster responses |
| GPT 6.2 Sol | `gpt-6.2-sol@OpenAI` | Recommended GPT default, the daily chat workhorse |
| GPT 6.2 Astra | `gpt-6.2-astra@OpenAI` | Highest-tier alternative, for complex tasks |

For the full model names of Chinese models see [Chinese models](/docs/usage/cn-models); they use the same `model-name@OpenAI` syntax and go into `Custom model name`, separated by commas.

## Other ingress points

The three ingress domains do the same job and share the same key — only the network route differs. If the network is unstable, try another one; the path syntax stays the same:

| Node | Anthropic (Claude) | OpenAI-compatible (GPT) |
|---|---|---|
| Global (default) | `https://api.qcode.cc/api` | `https://api.qcode.cc/openai` |
| Asia | `https://asia.qcode.cc/api` | `https://asia.qcode.cc/openai` |
| North America / Europe | `https://us.qcode.cc/api` | `https://us.qcode.cc/openai` |

## Deploy the web version yourself

Besides the desktop app, NextChat can be deployed as a web version that points at QCode through environment variables. Fill in three variables on the deployment platform:

```bash
BASE_URL=https://api.qcode.cc/openai
OPENAI_API_KEY=cr_xxxxxxxx
CODE=your-access-password
```

`BASE_URL` also stops at `/openai`, with no `/v1`; `CODE` is the access password for the web version, so that others cannot open your site.

## Related docs

- [Endpoints and API formats](/docs/getting-started/endpoints-and-api-paths) — protocols and model families side by side
- [Chinese models](/docs/usage/cn-models) — Chinese models available on the OpenAI-compatible endpoint
- [Troubleshooting guide](/docs/reference/troubleshooting) — 401, 404, timeouts and more
- [Billing](/docs/reference/billing) — plans and how to check usage