Continue (VS Code / JetBrains) with UberLLM: config.yaml that works

Steps

1. Make a key on your keys page and add credit.
2. Save the key where Continue looks for secrets, in ~/.continue/.env:

ULL_API_KEY=ull_your_key

The editor extensions do not see variables from your shell, so use this file (or .continue/.env in the project).

3. Open ~/.continue/config.yaml and add the models:

name: My Config
version: 1.0.0
schema: v1

models:
  - name: moonshotai/kimi-k3 (UberLLM)
    provider: openai
    model: moonshotai/kimi-k3
    apiBase: https://api.uberllm.dev/v1
    apiKey: ${{ secrets.ULL_API_KEY }}
    useResponsesApi: false
    useLegacyCompletionsEndpoint: false
    capabilities: [tool_use]
    roles: [chat, edit, apply]

  - name: deepseek/deepseek-v4.1-flash (UberLLM)
    provider: openai
    model: deepseek/deepseek-v4.1-flash
    apiBase: https://api.uberllm.dev/v1
    apiKey: ${{ secrets.ULL_API_KEY }}
    useResponsesApi: false
    useLegacyCompletionsEndpoint: false
    capabilities: [tool_use]
    roles: [chat, edit, apply]

4. Reload the window and pick the model in Continue's model menu.

The two route lines

Continue picks a route per model, and these two lines keep every UberLLM model on chat completions:

• useLegacyCompletionsEndpoint: false: autocomplete otherwise uses the old /v1/completions route, which UberLLM does not serve (404). This one is required.
• useResponsesApi: false: Continue sends any model whose name contains gpt-5 (or o1, o3…) to the Responses API, even with a custom apiBase. UberLLM serves that route too, so this line is no longer needed to avoid errors; it keeps all your models on the one route the config was tested on.

Autocomplete

Add a separate fast model with the autocomplete role:

  - name: Autocomplete (UberLLM)
    provider: openai
    model: deepseek/deepseek-v4.1-flash
    apiBase: https://api.uberllm.dev/v1
    apiKey: ${{ secrets.ULL_API_KEY }}
    useResponsesApi: false
    useLegacyCompletionsEndpoint: false
    roles: [autocomplete]

Autocomplete through a chat model works, but quality varies by model more than with chat. Try two or three from the fast table below. It sends many small requests, so keep an eye on Dashboard → Activity the first day.

Troubleshooting

• 404 on chat: apiBase does not end in /v1.
• 404 on autocomplete only: useLegacyCompletionsEndpoint: false is missing.
• 401: the secret is not found. Check the name matches (ULL_API_KEY) and the .env file is in one of the places above, then reload the window.
• Agent mode does nothing: add capabilities: [tool_use].
• 402 insufficient credits: add credit on the credits page.

Keep reading

• Continue doc
• Cline setup · Aider setup