Use the local API

Use the local API

LM Nexus offers two different local OpenAI-compatible paths:

1. the integrated Nexus /v1 API;
2. the direct endpoint of a running managed llama-server.

They are useful for different reasons.

Use the integrated Nexus /v1 API

1. Open API Serving

In Model Manager, select API Serving.

2. Enable API serving

Turn on:

Enable OpenAI-compatible /v1 API serving

API Serving is off by default.

After enabling it, the page checks /v1/health and shows the current API base URL, running-model count, endpoint list, and copyable examples.

3. Check the endpoint

With the normal local backend address, the simplest smoke request is:

curl http://127.0.0.1:8000/v1/models

The API Serving page also provides copyable examples for:

• curl;
• Python OpenAI client;
• Linux/WSL CLI;
• PowerShell CLI;
• NEXUS_API_URL.

4. Use a running profile id

For /v1/chat/completions, use a running Model Manager profile’s model_id.

The Servable models section on the API Serving page shows the configured profiles and links back to Profiles.

What the Nexus /v1 path does

The integrated path lets LM Nexus resolve Model Manager profiles and proxy OpenAI-compatible requests to the managed runtime.

Conceptually:

your client
    |
    v
LM Nexus /v1
    |
    v
managed llama-server

Use this path when you want the request to participate in Nexus Model Manager serving behavior.

Connect directly to the managed runtime

When a profile is running, Model Manager exposes the assigned port of its underlying llama-server.

The direct path is:

your client
    |
    +-----------------> managed llama-server

A direct request therefore looks like:

curl http://127.0.0.1:PORT/v1/models

Replace PORT with the assigned port shown by the running profile.

Use direct access when you specifically want to test or use the runtime without the Nexus /v1 serving layer.

Security note

The Enable API Serving switch controls whether LM Nexus accepts new /v1 requests.

It is not authentication, authorization, bind control, or a firewall.

Directly addressed managed-runtime ports are a separate path and are not protected by that switch.

Treat these endpoints as trusted-local development surfaces unless you have explicitly added an appropriate network/security boundary outside LM Nexus.

If the direct endpoint works but Nexus /v1 does not

That strongly suggests the managed runtime itself is healthy and the problem is somewhere in the LM Nexus serving path.

Check:

1. API Serving is enabled;
2. /v1 health status is Ready;
3. the requested model_id matches a servable profile;
4. the backend base URL is correct;
5. Nexus logs show how the request was resolved.

See Raw server access for the architecture distinction.