Connect a model

Studio ships with no model connected. This is how you connect it, and which model names silently fail.

Studio is deployed without a model provider. Everything else works — you can sign in, build a flow, wire components — right up to the moment you run it. Then the run fails with "An error occurred" and no further explanation.

This page is that one step.

Generate a key

Your NuFi console issues the key, and it bills to you.

Open console.nufi.me/keys.

Click Generate Key. Give it an alias you will recognise later — studio is fine.

Leave the defaults unless you have a reason: $10 budget over 30d, 10,000 tokens/min, 60 requests/min, expiring in 90d.

Copy the key. The full value is shown once — after you close the dialog the console only ever shows the first three and the last four characters.

Point Studio at the gateway

In Studio, open Settings → Model Providers.

Choose OpenAI Compatible. NuFi's gateway speaks the OpenAI protocol, so this is the right entry — not the OpenAI one above it.

Fill in:

Field	Value
Base URL	`https://api.codechi.me/v1`
API Key	the key you just copied

Click Save.

You will know it took because OpenAI Compatible jumps to the top of the provider list with a model count beside it.

Which models actually work

The gateway advertises just over thirty models. Seven of them are retired upstream and return an error when called, and several sort to the top of the dropdown, so the default choice is often a broken one.

These are advertised but do not work:

Nufi-lab/models/gemini-2.0-flash
Nufi-lab/models/gemini-2.0-flash-001
Nufi-lab/models/gemini-2.0-flash-lite
Nufi-lab/models/gemini-2.0-flash-lite-001
Nufi-lab/models/gemini-3-pro-preview
Nufi-lab/models/gemini-robotics-er-1.5-preview
Nufi-lab/models/gemini-omni-flash-preview

Google retired them; the gateway's list has not caught up. Calling one returns "This model is no longer available", which Studio surfaces to you only as "An error occurred".

Safe choices, verified against the live gateway:

Model	Good for
`gemini`	The general alias. A sensible default
`Nufi-lab/models/gemini-2.5-flash`	Fast, cheap, fine for most flows
`Nufi-lab/models/gemini-2.5-pro`	Harder reasoning
`Nufi-lab/models/gemini-3.5-flash`	Newer, fast
`Nufi-lab/models/gemini-2.5-flash-image`	Flows that produce images

Checking it worked

The fastest confirmation is a run, not a setting. Open any flow, click Playground, and send one message. An answer with a duration beside it means the whole path — Studio, key, gateway, model — is good.

If instead you get "An error occurred", the model is the first thing to change. See Troubleshooting.

Generate a key

Point Studio at the gateway

Which models actually work

Checking it worked

On this page