Skip to main content
You need one thing before you start: an API key from a model provider, for example OpenAI, Anthropic, Google or OpenRouter. nemu does not resell model access, it routes to the accounts you already have.

​
1. Connect a provider

In the console, open Providers and add one. Pick Automatic and choose from the models.dev catalog if your provider is a known one. nemu then knows its base URL, its protocol and its model list, and keeps the pricing and context windows current for you. Pick Custom if you are pointing at something the catalog does not know about, such as a local server or a private deployment, and give it a base URL and a compatibility mode. Paste the provider key. It is sealed and never shown again, and never leaves the console.

​
2. Enable a model

Open Models and add one under that provider. The important field is the gateway name. That is the string your code will send, and it is yours to choose. A model whose real identifier is nvidia/nemotron-nano-9b-v2 can be openrouter/nemotron to you. That name is the only name the gateway accepts. The provider’s own identifier does not work from outside, so nothing you run can reach a model you did not deliberately expose.

​
3. Create a key

Open API keys and create one. It looks like sk_xxxxx-xxxxx. followed by a 40 character secret, and it belongs to the workspace you are in. Copy it now. It is not shown again.

​
4. Send a request

To see exactly which names are accepted:
Every id in that response is a string you can put in model.

​
5. Point a real client at it

Most tools take a base URL and a key, so nothing else has to change.
Full guides: Claude Code, Codex, How the gateway works.

​
What to do next

Open Metrics and you will see the request you just made, with its model, latency, token counts and cost. From there, the pieces that are worth knowing about are mappings for renaming models without touching your code, quotas for capping spend, and workflows for routing a request through logic before it reaches a model.