Connect Visual Studio Code

VS Code’s built-in Chat view supports Custom Endpoint models. Cline and other extensions have separate settings; use their own guides when configuring them.

Prepare your service key, an exact ID from our models endpoint, and verified input/output limits for the chosen model.

Add the endpoint

Run Chat: Manage Language Models from the Command Palette. Choose Add Models → Custom Endpoint, give the group a name such as My API, and enter your key in the API-key prompt. Select Chat Completions if asked.

VS Code opens chatLanguageModels.json. Keep the generated apiKey reference to secret storage. Update this provider’s model list using the template below; preserve all other providers.

Replace the full URL with the service Base URL above plus /chat/completions. Replace the ID and both [TBD] values; token limits must be integers without quotes before saving.

[
  {
    "name": "My API",
    "vendor": "customendpoint",
    "apiKey": "${input:KEEP_THE_GENERATED_SECRET_REFERENCE}",
    "apiType": "chat-completions",
    "models": [
      {
        "id": "YOUR_MODEL_ID",
        "name": "My API model",
        "url": "FULL_CHAT_COMPLETIONS_URL",
        "toolCalling": true,
        "vision": false,
        "maxInputTokens": "[TBD]",
        "maxOutputTokens": "[TBD]"
      }
    ]
  }
]

The apiKey line above is explanatory: retain the actual ${input:...} value VS Code generated. Enter the secret through the wizard, rather than inserting a literal key in JSON. id is sent unchanged to our API; name is a display label. This API type uses Bearer authentication automatically.

Only set toolCalling: true for a model with tool support. The input and output limits together must fit its context window. Save, then select My API model in Chat. Restart VS Code if it is missing. See the official Custom Endpoint configuration reference.

Verify and check feature scope

In a fresh chat, request CONNECTION_OK without tools and compare the reply with the matching service request record. Test a small file read separately with normal approvals. Each request uses API balance; saving JSON does not verify the key.

BYOK chat works without a GitHub sign-in or Copilot plan. Inline suggestions, semantic search and embedding features have separate GitHub requirements. Organization policy can also restrict BYOK. Without Copilot sign-in, configure chat.utilityModel and chat.utilitySmallModel to an available model, or set chat.byokUtilityModelDefault to Main Agent Model, if background utility features need a model. Those calls use API balance too. VS Code documents these limits and utility settings.

For Copilot harness sessions in the Agent Host, VS Code currently also requires the experimental chat.agentHost.byokModels.enabled setting. Check this if your model appears in regular Chat but not in that harness.

If a model is missing in agent mode, check its tool-call capability. For authentication errors, reopen the group’s gear menu and re-enter the key. For endpoint or model errors, check the full /v1/chat/completions URL and exact ID, then use our API troubleshooting.