Use Dolphin LLM in Continue for VS Code and JetBrains

Add the model to config.yaml

Open Continue's settings (the gear in the Continue panel) and choose to open the config file, or edit ~/.continue/config.yaml directly. Add a model:

name: My config
version: 0.0.1
schema: v1
models:
  - name: Dolphin LLM
    provider: openai
    model: uncensored
    apiBase: https://dolphinllm.com/v1
    apiKey: YOUR_KEY
    roles:
      - chat
      - edit
      - apply
    defaultCompletionOptions:
      contextLength: 262144
      maxTokens: 2048
      temperature: 0.3

Save the file. Continue reloads it, and Dolphin LLM appears in the model menu of the chat panel. Create your key in your account; if you do not have one yet, sign up and $0.50 of trial credit is added when you first open the account.

Keep the key out of the file

If you share or sync your config, do not paste the key into it. Recent versions of Continue read secrets from ~/.continue/.env:

# ~/.continue/.env
DOLPHIN_API_KEY=YOUR_KEY

and the model block then uses apiKey: ${{ secrets.DOLPHIN_API_KEY }}. If your version does not resolve it, it reports a missing key; paste the key directly and keep the file private.

Which roles to give it

• chat: questions about code, explanations, refactoring plans. The 262,144-token window holds about 490 pages of code and text, so several files fit in one question.
• edit and apply: rewriting a selection or applying a suggested change to the file.
• autocomplete: not recommended. Tab completion fires a request on almost every pause in typing, which adds up in cost and waits on network latency. A small local model is better for that.
• embed: not available. This API has no embeddings endpoint, so codebase indexing needs a separate embeddings provider or a local one.

Tips for coding work

• Use a low temperature (0.2 to 0.4) for code, higher only for brainstorming.
• Attach only the files you need. Every attached line is input tokens on each request in that chat.
• Start a new chat when you switch tasks, so the old history is not sent again.
• Continue streams replies; the API supports streaming with token usage at the end.

Cost

A typical request with 8,000 tokens of code and question and a 1,000-token answer costs 8,000 × $0.15 / 1,000,000 + 1,000 × $0.50 / 1,000,000 = $0.0017. A day with 100 such requests is $0.17, about $5 a month. The trial covers about 290 of them. See the pricing page.

Troubleshooting

• 404 not_found. apiBase must end in /v1; Continue adds /chat/completions itself.
• 401 invalid_key. Check the key or the secret name; restart the editor after changing .env.
• 429 concurrency on a trial key. The trial allows 2 parallel requests. Turn off other Continue features that call the same model, or top up for 8 parallel requests.