Continue setup: OpenAI provider with a custom apiBase

Continue is an open-source AI assistant for VS Code and JetBrains IDEs. Its openai provider works with any OpenAI-compatible endpoint once you set apiBase, so our gateway slots straight in. Each model you list becomes selectable in Continue’s chat panel, and all usage is charged to your prepaid balance.

What you need

• The Continue extension installed in VS Code or a JetBrains IDE.
• A funded account: any balance above $0, the sign-up credit included, pays for every model; when it runs out, add $1 or more (sign up).
• An API key from the console, starting with sk-.

Step 1: open your config.yaml

Continue keeps its local configuration in ~/.continue/config.yaml (on Windows, %USERPROFILE%\.continue\config.yaml). You can open it from Continue’s settings in the IDE or edit it in any text editor. If the file already has a models: list, add the entries below to it rather than replacing it.

Step 2: add the gateway models

name: My Config
version: 0.0.1
schema: v1

models:
  - name: Claude Sonnet 5.5 (One AI API Key)
    provider: openai
    model: claude-sonnet-5-5
    apiBase: https://api.oneaiapikey.com/v1
    apiKey: ${{ secrets.ONEAIAPIKEY_API_KEY }}
    roles:
      - chat
      - edit
      - apply

  - name: GPT-6.1 Sol (One AI API Key)
    provider: openai
    model: gpt-6.1-sol
    apiBase: https://api.oneaiapikey.com/v1
    apiKey: ${{ secrets.ONEAIAPIKEY_API_KEY }}
    roles:
      - chat
      - edit

  - name: GLM-5.3 Flash (One AI API Key)
    provider: openai
    model: glm-5.3-flash
    apiBase: https://api.oneaiapikey.com/v1
    apiKey: ${{ secrets.ONEAIAPIKEY_API_KEY }}
    roles:
      - chat
      - summarize

The fields that matter:

• provider: openai is correct even for Claude and GLM. It selects the OpenAI request format, not the vendor.
• model is the exact id from our catalogue.
• apiBase includes /v1. Continue appends /chat/completions on its own.
• apiKey uses Continue’s secrets syntax so the key itself stays out of the file.
• roles decides where each model shows up: chat for the chat panel, edit for inline edits, apply for applying code blocks to files, summarize for summaries.

YAML is indentation-sensitive. Use spaces (not tabs) and keep each model’s fields aligned under its - name: line.

Step 3: provide the key

The secrets reference above resolves from a local .env file. Create ~/.continue/.env with one line:

ONEAIAPIKEY_API_KEY=sk-your-key-here

If your Continue version doesn’t pick the value up, you can paste the key directly into apiKey instead, but then treat config.yaml as a secret: don’t sync it to a public dotfiles repo.

Recommended models

Rates are USD per 1M tokens:

claude-sonnet-5-5 is an excellent chat and edit model. glm-5.3-flash keeps summaries and quick questions cheap. These chat models aren’t designed for tab autocomplete, so we don’t recommend giving them the autocomplete role. More options are on the models page.

Step 4: verify it works

Save config.yaml; Continue reloads it automatically. Open the Continue panel, pick one of the new models from the model dropdown and send a short question. A normal reply means everything is wired up, and the request will show in your console’s usage log within seconds.

If nothing works, test the key outside the IDE:

curl -sS https://api.oneaiapikey.com/v1/models -H "Authorization: Bearer sk-your-key-here"

Troubleshooting

The new models don’t appear in the dropdown. Continue couldn’t parse the file. Check indentation and that the three top-level keys (name, version, schema) are present. Continue usually surfaces config errors in its panel.

401 Unauthorized. The secret didn’t resolve or the key is wrong. Confirm the variable name in .env matches the one in secrets. exactly, then reload the IDE window.

404 Not Found. apiBase must be https://api.oneaiapikey.com/v1, without a trailing slash or endpoint path.

“Model not found” or “no available channel”. The model value doesn’t match a catalogue id. Note that name is just a label; only model is sent to the API.

Long answers cut off or time out. Continue streams by default, which keeps long generations alive. In your own scripts, use "stream": true for long outputs, since proxies cut idle non-streaming requests at roughly 100 seconds.

Next steps

Want an autonomous agent in the same editor? See Cline or Kilo Code. All guides are listed under tool guides, and every rate is on the pricing page.