Cloud and local inference integration

Connection

Base URL

https://api.eelml.com/model-gateway/v1

Model alias

eelml-cloud-auto

Authentication

Existing EELML bearer token

Streaming

Server-Sent Events when stream is true

Local path

EELML Studio Runtime Service with .eelml packages

1 · Choose the execution path

Use a supported .eelml package for private on-device inference, or select EELML Cloud Auto when the task needs managed online reasoning.

2 · Authenticate for cloud inference

Cloud requests use the same verified EELML account token as the control plane. A subscription alone does not grant restricted cloud-model access. Local inference does not send this token to a model provider.

3 · Discover the cloud service alias

GET /models returns only aliases the current account may discover. Provider names and provider model IDs remain server-side.

4 · Create a cloud completion

POST /chat/completions using eelml-cloud-auto. Set stream to true for SSE or false for one JSON response.

5 · Continue tool calls in Harness

Both execution paths return model decisions to the same Harness. Actual tool execution remains inside the Agent Kernel.

Surface

GET

/models

POST

/chat/completions

LOCAL

Runtime Service

Streaming completion

curl https://api.eelml.com/model-gateway/v1/chat/completions \ -H "Authorization: Bearer $EELML_TOKEN" \ -H "Content-Type: application/json" \ -d '{ "model": "eelml-cloud-auto", "messages": [{"role": "user", "content": "Summarize this task."}], "stream": true }'

Security and product boundaries

• Cloud Credits and provider/model entitlement are independent checks. Purchasing a plan does not automatically enable an expensive provider.
• The gateway fails closed when no authorized route is available; it does not silently choose an unapproved provider.
• Local inference does not use this endpoint. Supported .eelml packages run through the local Runtime Service inside EELML Studio.