Integrations
The chat in VS Code is GitHub Copilot Chat, and its built-in Custom Endpoint connects it to DEVUP AI. VS Code can use DEVUP AI models in chat and Agent mode through its built-in Custom Endpoint, with no extension required.
Open the native Chat panel in VS Code.
In the chat panel, click the active model name above the input area to reveal the dropdown and select Manage Models.
From the management menu, select Add Models and pick Custom Endpoint.
Select Chat Completions as the protocol format.
Assign an identifier to your endpoint group and supply your API key from the API keys dashboard.
Paste the following configuration inside the models array:
{
"id": "deepseek-ai/DeepSeek-V4-Pro",
"name": "DeepSeek V4 Pro (DEVUP AI)",
"url": "https://api.devupai.com/v1/chat/completions",
"toolCalling": true,
"vision": false,
"maxInputTokens": 1040384,
"maxOutputTokens": 8192
}url is the full endpoint path, not a base URL.vision to match the model you are adding. Models without image input should have it set to false.Save chatLanguageModels.json and pick your new model directly in the chat panel.
{
"id": "deepseek-ai/DeepSeek-V4-Pro",
"name": "DeepSeek V4 Pro (DEVUP AI)",
"url": "https://api.devupai.com/v1/chat/completions",
"toolCalling": true,
"vision": false,
"maxInputTokens": 1040384,
"maxOutputTokens": 8192
}Each entry in the Custom Endpoint models array accepts the following parameters:
| Field | Type | Description |
|---|---|---|
| id | string | Model identifier from DEVUP AI catalog (e.g. "deepseek-ai/DeepSeek-V4-Pro") |
| name | string | Display label shown in VS Code model picker |
| url | string | Full endpoint path: "https://api.devupai.com/v1/chat/completions" |
| apiType | string | "chat-completions" (default) or "responses". Set "responses" per model, with the url https://api.devupai.com/v1/responses, for GPT-5 models such as openai/gpt-5.4-mini. |
| toolCalling | boolean | Set to true to enable workspace tool use and Agent mode |
| vision | boolean | Set to match model capabilities; false for models without image input |
| maxInputTokens | number | Input tokens for the model. maxInputTokens plus maxOutputTokens must not exceed the model's context window: for DeepSeek-V4-Pro (1,048,576 tokens), use 1040384 with 8192. |
| maxOutputTokens | number | Maximum tokens per answer: at most the model's max_output_tokens in GET /v1/models. Read DeepSeek-V4-Pro's with GET /v1/models/deepseek-ai/DeepSeek-V4-Pro. |
In Agent mode, VS Code reads workspace files, runs terminal commands, and applies code edits through tool calling, and streams the model's replies into the chat.
Agent mode on DeepSeek V4 Pro (DEVUP AI): Copilot Chat created hello.py and ran it.
Add a GPT-5 model to the same models array with its own apiType and url:
{
"id": "openai/gpt-5.4-mini",
"name": "GPT-5.4 mini (DEVUP AI)",
"url": "https://api.devupai.com/v1/responses",
"apiType": "responses",
"toolCalling": true,
"vision": true,
"maxInputTokens": 272000,
"maxOutputTokens": 8192
}It answers in Ask mode through the DEVUP AI Responses API. maxInputTokens plus maxOutputTokens stays within its 400K context, and maxOutputTokens within its max_output_tokens in GET /v1/models.
DEVUP AI models added through the Custom Endpoint answer in Copilot Chat after you sign out of GitHub in VS Code. Features that rely on GitHub's own services, such as inline suggestions, need a GitHub account.
For Agent mode, use a model whose catalog entry supports tool calling, for example deepseek-ai/DeepSeek-V4-Pro, and set toolCalling to true in its entry. See Tool Calling and Models.
id field must match the catalog exactly, for example deepseek-ai/DeepSeek-V4-Pro.Yes. VS Code's built-in Custom Endpoint adds DEVUP AI models to the Copilot Chat model picker, in Ask and Agent mode.
No. DEVUP AI models added through the Custom Endpoint answer in Copilot Chat after you sign out of GitHub. Inline suggestions need a GitHub account.
Models with toolCalling set to true, such as deepseek-ai/DeepSeek-V4-Pro.
Add the model with apiType set to responses and the url https://api.devupai.com/v1/responses, as in the GPT-5.4 mini entry on this page.
Their sum must not exceed the model's context window, and maxOutputTokens must not exceed the model's max_output_tokens in GET /v1/models. For DeepSeek-V4-Pro, 1040384 and 8192 work; for GPT-5.4 mini, 272000 and 8192.
Copilot Chat runs inside VS Code and reads its models from chatLanguageModels.json. For the terminal agent, see the GitHub Copilot CLI guide.