Skip to content

Continue for VS Code: DeepSeek V4 Flash

Use Continue's direct compatibility transport against CloudService's Chat Completions endpoint.

Use Chat Completions. Keep useResponsesApi: false. Tool calling is enabled only when authenticated model discovery reports capabilities.tools=true for the active deployment.

Automatic configuration

bash
npm install -g cloudservice@latest
cloudservice login
cloudservice update
cloudservice doctor

The CloudService CLI installs the official Continue extension when VS Code is detected, merges the model into ~/.continue/config.yaml, stores the already-authenticated key in private ~/.continue/.env, and verifies the result. Reload VS Code once after setup.

Store the CloudService key privately

Create or update ~/.continue/.env with a single entry:

text
CLOUDSERVICE_API_KEY=replace-with-your-CloudService-key

Continue's VS Code extension resolves this value through its secrets syntax. Use only the CloudService key for the billing URL you purchased. You do not need any upstream provider credential.

Manual config.yaml

If you would rather edit the file directly, merge this model into ~/.continue/config.yaml. This is the exact block cloudservice update writes for an API-credit key; preserve any unrelated models entries already in the file.

config.yaml
name: CloudService
version: 1.0.0
schema: v1
models:
  - name: CloudService DeepSeek V4 Flash
    provider: openai
    model: deepseek-v4-flash
    apiBase: https://api.yourdomain.example/v1
    apiKey: ${{ secrets.CLOUDSERVICE_API_KEY }}
    useResponsesApi: false
    roles:
      - chat
      - edit
    requestOptions:
      headers:
        User-Agent: CloudService-Continue/1.0
    env:
      apiType: openai-compatible
    defaultCompletionOptions:
      maxTokens: 4096

useResponsesApi: false and the chat/edit roles keep Continue on the Chat Completions contract this deployment supports. For a DeepSeek token-budget key, change only apiBase to https://api.yourdomain.example/token/v1.

API-credit configuration

Sign in with the API-credit key, then run cloudservice update. The CLI configures model deepseek-v4-flash against https://api.yourdomain.example/v1, enables Chat/Edit, keeps the Responses API disabled, and caps output at 4096 tokens. Agent tools are offered only when authenticated discovery says the active deployment supports them.

DeepSeek token-budget configuration

Sign in with the DeepSeek token-budget key and run the same update command. CloudService automatically uses https://api.yourdomain.example/token/v1; do not interchange this URL with the API-credit URL.

Verify in VS Code

Reload the VS Code window, open Continue, choose CloudService DeepSeek V4 Flash, and send Reply with K only. Confirm that the response streams and ends normally. Continue is the universal Chat/Edit route; use Roo, OpenClaw, or Hermes for agent workflows only when authenticated discovery reports capabilities.tools=true. If the profile is missing, rerun cloudservice update rather than hand-editing internal compatibility fields.

Keep the secret out of project files

Do not paste the key into a workspace config.yaml, commit ~/.continue/.env, or replace the CloudService key with an upstream provider credential.

See Continue's OpenAI-compatible provider guide and current config.yaml reference.