Continue OpenAI-compatible API: config.yaml, apiBase and Chat Completions

Continue stores model settings in config.yaml. For ChinaGenAI, the important distinction is the endpoint: the site exposes Chat Completions at /v1/chat/completions, not the Responses API.

Use the OpenAI provider with a custom API base

Continue documents apiBase for OpenAI-compatible services. This config uses the site's free route label and keeps the site key in Continue's secret resolver.

Add the provider to config.yaml

name: ChinaGenAI
version: 0.0.1
schema: v1

models:
  - name: ChinaGenAI free route
    provider: openai
    model: free
    apiBase: https://chinagenai.com/v1
    apiKey: ${{ secrets.CHINAGENAI_SITE_KEY }}
    useResponsesApi: false

Open Continue's local YAML config from its config menu. The current Continue docs use config.yaml; the older JSON config is marked deprecated.
Keep the apiBase at the API root ending in /v1. Do not append /chat/completions; Continue builds the request path.
Put the site key in a local dotenv file that Continue can read, then keep that file out of version control. Replace only the placeholder name above; do not paste a real key into a shared project config.
Save the YAML and select the ChinaGenAI model in Continue.

Keep the site key local

Continue resolves ${{ secrets.CHINAGENAI_SITE_KEY }} from a workspace .env, a workspace .continue/.env, a global ~/.continue/.env, or—only for the Continue CLI—the process environment. For an IDE extension, a shell-only export is not enough. A global Continue dotenv file keeps the key outside a project checkout; if you use a workspace dotenv file, add it to .gitignore.

CHINAGENAI_SITE_KEY=YOUR_SITE_KEY

After changing a secret, Continue's documentation says to restart the IDE. A changed config that does not appear may also require a VS Code window reload.

Chat Completions, not Responses

The site endpoint accepts POST https://chinagenai.com/v1/chat/completions. Continue's current OpenAI provider documentation says it can select the Responses API by default for some recognized model IDs, including o-series and GPT-5. This example sets useResponsesApi: false so the request matches the site's Chat Completions route. A path ending in /responses is not supported here.

Start with the free model label shown above. It is the site route label, not a claim that a particular upstream model will answer. Check current key availability and limits before making a request; changing Continue settings does not make a key or raise a quota.

Check the symptom before changing providers

What you see	Check first
`404` or a path containing `/responses`	Keep `apiBase` at `https://chinagenai.com/v1` and set `useResponsesApi: false`. The ChinaGenAI route is Chat Completions only.
`401`	Use a ChinaGenAI site key, not a key from OpenAI or another provider. Check whether site-key issuance is currently available.
The model does not appear	Check the exact model label, the base URL and site status. Model-list discovery and a successful chat request are separate checks.
Agent tools or images do not work	OpenAI-compatible request shape does not guarantee that a selected upstream model supports tools or images. Check the provider's capability information before enabling a feature.
The new YAML or secret is not picked up	Save the config, restart Continue or reload the VS Code window, and verify that the dotenv file is in a location Continue can read.

Continue's apiBase setting and Chat Completions config are documented by Continue. ChinaGenAI's route and feature limits are site-specific; read the connection details before connecting a larger workflow. If you already have a raw response, the local response checker can inspect it without another API request.

Official references

Continue: OpenAI and OpenAI-compatible provider config
Continue: secrets, environment files and troubleshooting

Official Continue documentation checked 20 October 2026. The page is setup guidance; no provider key or live model request was used.

Free API access and limits

Check the site free API access and limits before trying to obtain a site key. Read the current status and request limits before using a site key. It is not a source of official provider keys.