Skip to main content

OpenAI-compatible remote gateways

OGX's remote::openai inference provider talks to any endpoint that implements the OpenAI Chat Completions (and related) HTTP surface. You do not need a custom provider package for third-party multi-model gateways: set base_url (and an API key) on remote::openai.

This is the same mechanism the starter distribution uses when you export OPENAI_API_KEY and optionally OPENAI_BASE_URL. See also the OpenAI API compatibility overview and the Open Responses / OpenAI compatibility blog post.

Example: DaoXE​

DaoXE is a multi-model, multi-protocol API gateway. For OGX, use its OpenAI-compatible base URL:

https://daoxe.com/v1

DaoXE also exposes other client protocols (for example Anthropic Messages) for non-OGX clients. Inside OGX you configure only the OpenAI-compatible path above.

Availability

DaoXE is not available in mainland China. Create an account and API key at daoxe.com. Model IDs are account-scoped (availability can differ by plan or group). Always discover exact IDs with authenticated GET /v1/models rather than hard-coding a public catalog.

Option A — starter distribution + environment variables​

export OPENAI_API_KEY="your_daoxe_api_key"
export OPENAI_BASE_URL="https://daoxe.com/v1"
uvx --from 'ogx[starter]' ogx run starter

With the starter's default provider_id: openai, registered model IDs look like openai/<upstream_model_id>. List them after the server is up:

curl -s http://localhost:8321/v1/models | python -m json.tool

Option B — explicit config.yaml entry​

Prefer a dedicated provider ID when you also keep a direct OpenAI key, or when you want clearer model prefixes:

version: 2
apis:
  - inference
  - responses
providers:
  inference:
    - provider_id: daoxe
      provider_type: remote::openai
      config:
        api_key: ${env.DAOXE_API_KEY}
        base_url: https://daoxe.com/v1
        # Optional: pin models after discovery
        # allowed_models:
        #   - your_exact_model_id
        refresh_models: true

Start the server with that config (see Customizing config.yaml and Starting the OGX server). Model IDs will be prefixed with the provider ID, for example daoxe/<your_exact_model_id>.

Call OGX with the OpenAI SDK​

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8321/v1", api_key="fake")

# Replace with an ID from GET /v1/models for your server
response = client.chat.completions.create(
    model="daoxe/your_exact_model_id",
    messages=[{"role": "user", "content": "Hello from OGX + DaoXE"}],
    max_tokens=64,
)
print(response.choices[0].message.content)

Call DaoXE directly (without OGX)​

If you only need the gateway (no vector stores, Responses orchestration, etc.), point the OpenAI client at DaoXE:

from openai import OpenAI

client = OpenAI(
    base_url="https://daoxe.com/v1",
    api_key="your_daoxe_api_key",
)

response = client.chat.completions.create(
    model="your_exact_model_id",  # from DaoXE GET /v1/models
    messages=[{"role": "user", "content": "Hello"}],
    max_tokens=64,
)
print(response.choices[0].message.content)

Auditable examples and a low-cost smoke benchmark live at seven7763/DaoXE-AI.

GoalProvider
OpenAI API or any OpenAI-compatible gateway (base_url override)remote::openai
Arbitrary OpenAI-compatible endpoint without a fixed default URLremote::passthrough
Llama API OpenAI-compat surfaceremote::llama-openai-compat

Disclosure​

This page was contributed by a DaoXE maintainer. Configuration uses only stock OGX providers; no DaoXE-specific adapter is required.