Skip to main content

Continue

Connect Continue to Aurora for IDE chat and editing. This guide uses Continue's YAML configuration and disables the Responses API.

Select Inference Endpoint
Choose preferred region.

Install and prepare​

Install the Continue IDE extension and create an Aurora token.

Put the token in ~/.continue/.env:

AURORA_API_TOKEN=<your-token>

Configure the model​

Open the local configuration from Continue's model/configuration selector. For a new ~/.continue/config.yaml, use:

name: Aurora
version: 1.0.0
schema: v1
models:
- name: DeepSeek V4 Flash (Aurora)
  provider: openai
  model: deepseek/deepseek-v4-flash
  apiBase: 
  apiKey: ${{ secrets.AURORA_API_TOKEN }}
  useResponsesApi: false
  roles:
    - chat
    - edit
    - apply

For an existing configuration, append just the model entry to models. Keep the full Aurora model ID; provider: openai already selects the adapter, so no openai/ model prefix is needed. The URL includes /v1.

Try chat and agent mode​

Reload the local configuration, choose DeepSeek V4 Flash (Aurora), attach an existing README.md through Continue's context picker, and ask:

Summarize the attached README in three bullets.

Expect a streamed response based on the attachment. For Agent mode, verify the model supports tool calling, then add this key to its model entry:

capabilities:
  - tool_use

Switch to Agent mode and ask it to read a different small project file and summarize it without edits. Expect a tool result followed by the summary. The capability setting tells Continue to expose tools; it does not add tool support to the model.

Troubleshooting​

  • Secret not found: Check ~/.continue/.env, the exact secret name, and reload the extension. The YAML uses Continue's secrets syntax, not a shell export.
  • /responses errors: Keep useResponsesApi: false; do not enable Responses-only features on this entry.
  • Agent mode unavailable: Check tool_use and the selected model's support.
  • Autocomplete or indexing fails: This guide configures chat/edit/apply roles. Autocomplete and embeddings are separate model roles; an Aurora chat connection does not configure an embedding endpoint.

See shared troubleshooting, Continue's OpenAI-compatible configuration, and the YAML reference.