---
title: "Local model servers"
description: "Serve models on your machine and expose them through an upstream."
---

> Documentation Index
> Fetch the complete documentation index at: https://docs.noket.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Local model servers

Any HTTP API that speaks the OpenAI or Anthropic protocol can sit behind a Noket upstream. These
guides cover the runtimes people use most, with the addresses and flags each one needs.

| Runtime | Guide | Default address | Protocol |
| --- | --- | --- | --- |
| Ollama | [Ollama](https://docs.noket.dev/connector/local-models/ollama/) | `http://127.0.0.1:11434` | `openai` |
| vLLM | [vLLM](https://docs.noket.dev/connector/local-models/vllm/) | `http://127.0.0.1:8000` | `openai` |
| LM Studio | [LM Studio](https://docs.noket.dev/connector/local-models/lm-studio/) | `http://127.0.0.1:1234` | `openai` |
| Any Messages API server | [Anthropic-compatible servers](https://docs.noket.dev/connector/local-models/anthropic-compatible/) | varies | `anthropic` |

## What Noket needs from a server

- An address the machine running the connector can reach. Loopback is fine; Noket calls it from the connector,
  not from the cloud.
- A model id the server reports at `{base_url}/v1/models`, or an id you can type by hand when the probe
  does not return one.
- The OpenAI or Anthropic wire format. If a server offers both, pick the one whose response format your
  clients expect.
- An API key only when the server requires one. Local servers usually do not; if yours does, pipe the
  key into `noket upstream add` from your secret manager.

A trailing `/v1` in the base URL is allowed. Noket normalizes the path before probing and forwarding.

## Next

- [Add and manage upstreams](https://docs.noket.dev/connector/upstreams/)
- [Probing and model discovery](https://docs.noket.dev/concepts/probing-and-model-discovery/)

Source: https://docs.noket.dev/connector/local-models/index.mdx