Continue

Continue reaches ai.ml through its openai provider. Give each model an entry in ~/.continue/config.yaml, with the key held as a Continue secret:

# ~/.continue/config.yaml (or the assistant's config in Continue Hub)
name: ai.ml
version: 1.0.0
schema: v1
models:
  - name: Claude Sonnet 4.6 (ai.ml)
    provider: openai
    model: anthropic/claude-sonnet-4.6
    apiBase: https://api.ai.ml/v1
    apiKey: ${{ secrets.AIML_API_KEY }}
    roles: [chat, edit, apply]
  - name: GPT-5 (ai.ml)
    provider: openai
    model: openai/gpt-5
    apiBase: https://api.ai.ml/v1
    apiKey: ${{ secrets.AIML_API_KEY }}
    roles: [chat, edit, apply]
    # Continue sends gpt-5 ids to /v1/responses; ai.ml serves both, and this
    # keeps every model on /v1/chat/completions
    useResponsesApi: false

• model is the ai.ml id, or an alias. Continue sends it as written.
• Continue sends a model whose name contains gpt-5 (or o1, o3, …) to /v1/responses instead of /v1/chat/completions. ai.ml serves both. useResponsesApi: false keeps such a model on Chat Completions, like every other model.
• Add roles: [embed] to an embedding model's entry (for example mock/embed-1 on a test cell) to use ai.ml for codebase indexing.
• Continue does not read the model list from the API. Each model needs its own entry.

Verified with the Continue CLI 1.5.47, which reads the same config.yaml. Both entries were run, and the GPT-5 entry was run on both routes.