Skip to main content

​
LiteLLM Proxy Configuration

LiteLLM Proxy is a gateway that speaks the OpenAI API and forwards requests to upstream providers. Your organization or Novaplan can operate the proxy; coordinate its configuration and network access with Novaplan before connecting it to a managed workspace. Use LiteLLM Proxy when you want one place to manage keys, spending limits, and fallbacks for every model your organisation uses, instead of configuring each provider separately in Novaplan AI.
LiteLLM Proxy is the only provider in Novaplan AI that supports all five model types: text generation, embeddings, image generation, text to speech, and speech to text. You configure each one separately, and they can point at the same proxy.

​
Before you start

You need a running LiteLLM Proxy instance that the managed Novaplan AI workspace can reach. The LiteLLM Proxy quick start covers proxy setup for its operator. Add at least one model to its config.yaml; its default port is 4000.

​
Required Fields

​
Endpoint URL *

The address where your LiteLLM Proxy is reachable, including the scheme and port. Application default: http://host.docker.internal:4000. This hostname works only if the managed workspace can resolve and reach it. Ask Novaplan for the actual reachable proxy URL, for example https://litellm.example.com. A localhost address on your computer is not reachable from the managed workspace by default.

​
API Key *

The master key or virtual key from your LiteLLM Proxy. This is the key you set as master_key in the proxy’s configuration, or a virtual key you generated with it. This is not the upstream provider’s key. Your OpenAI or Anthropic keys stay in the proxy’s own configuration; Novaplan AI never sees them.

​
Model Name *

The model name exactly as it appears under model_name in your LiteLLM Proxy configuration, not the upstream provider’s name for it. For example, if your proxy config contains:
then enter my-gpt here. The proxy administrator can list the names it exposes from a machine that can reach it, for example:

​
Optional Fields

​
Model Friendly Name

A label shown in the Novaplan AI interface so you can tell several configurations apart. If you leave it blank, the model name is used.

​
Context Length

The size of the model’s context window, in tokens. Novaplan AI uses this number to decide how much retrieved content it can safely include in a prompt. Set it to the value published for your model. If you leave it blank, Novaplan AI uses a conservative default.

​
Is Multimodal

Turn this on if the model accepts images as well as text. It is on by default.

​
Is Reasoning

Turn this on if the model performs extended reasoning before answering. It is on by default.

​
Troubleshooting

“Connection refused” when saving. Novaplan AI cannot reach the endpoint. Ask the proxy administrator and Novaplan to check its status and the network address from the managed workspace. “Model not found”. The model name does not match any model_name in the proxy configuration. List the available names with the curl command above. Authentication errors. The API Key does not match the proxy’s master_key, or the virtual key has expired.