Connect your tools

Copy-paste config for every IDE, agent, and service — pre-filled with this hub's address. Manage API keys at Config → API Keys.

✓ 2 nodes live — cluster routing is active. Your tools automatically use the best available node.

🔓 Open mode — no API key needed. Set api_key in homelab.yaml to require tokens and enable per-service attribution.

Which endpoint to use?

IDE / coding agent / script on this machine http://localhost:42420/v1 Daemon — loads models on demand, routes to cluster silently

Always-on service on the LAN (n8n, HA, Docker) http://hub.home.browngregory.com/v1 Hub — direct cluster placement, no extra hop

Ollama-native tool (Open-WebUI, AnythingLLM) http://hub.home.browngregory.com/ollama Hub Ollama proxy — same placement, Ollama wire format

MCP-aware agent (Cursor, Claude Desktop, Goose) http://localhost:8092/sse Full cluster control — load/unload models, stream events

IDEs & coding assistants

🖱

Cursor

AI-first code editor

OpenAI API

Cursor → Settings → Models → OpenAI API Key

none

Override URL

http://localhost:42420/v1

Models (type to add any Ollama model name)

qwen3:8b
qwen3.5:27b
qwen3.5:4b

⇌

Continue.dev

VSCode + JetBrains AI plugin

OpenAI API

~/.continue/config.json (add to models array)

{
  "models": [
    {
      "title": "homelab — qwen3:8b",
      "provider": "openai",
      "model": "qwen3:8b",
      "apiBase": "http://localhost:42420/v1",
      "apiKey": "none"
    }
  ],
  "tabAutocompleteModel": {
    "title": "homelab — qwen3.5:4b",
    "provider": "openai",
    "model": "qwen3.5:4b",
    "apiBase": "http://localhost:42420/v1",
    "apiKey": "none"
  }
}

✏

aider

AI pair programmer (terminal)

OpenAI API

~/.aider.conf.yml

openai-api-base: http://localhost:42420/v1
openai-api-key: none
model: openai/qwen3:8b

Or as env vars

export OPENAI_BASE_URL=http://localhost:42420/v1
export OPENAI_API_KEY=none
aider --model openai/qwen3:8b

⚡

avante.nvim

Neovim AI assistant

OpenAI API

init.lua / lazy.nvim config

require("avante").setup({
  provider = "openai",
  openai = {
    endpoint = "http://localhost:42420/v1",
    model    = "qwen3:8b",
    api_key  = "none",
  },
})

Coding agents

🪿

Goose

Block's open-source coding agent

MCP + OpenAI

~/.config/goose/config.yaml

GOOSE_PROVIDER: openai
GOOSE_MODEL: qwen3:8b
OPENAI_HOST: http://localhost:42420
OPENAI_API_KEY: none

extensions:
  homelab:
    type: sse
    url: http://localhost:8092/sse
    name: homelab
    timeout: 300
    enabled: true

🤖

Claude Desktop

Anthropic's desktop AI app

MCP only

claude_desktop_config.json → mcpServers

{
  "mcpServers": {
    "homelab": {
      "url": "http://localhost:8092/sse",
      "type": "sse"
    }
  }
}

Claude Desktop uses its own cloud model for inference. The homelab MCP server gives it cluster management tools (load/unload models, view nodes, run agent tasks).

Services & automation

⚙

n8n

Workflow automation

OpenAI API

n8n → Credentials → OpenAI API

Base URL:  http://hub.home.browngregory.com/v1
API Key:   none

Use the hub URL (not daemon) since n8n always has LAN access. The hub's placement engine picks the best cluster node per request.

🏠

Home Assistant

Home automation platform

OpenAI API

configuration.yaml

conversation:
  agent: homelab

homeassistant_conversation:
  chat_model: qwen3:8b
  api_base: http://hub.home.browngregory.com/v1
  api_key: none
  max_tokens: 1024

∞

AnythingLLM

Document chat + RAG platform

Ollama API

Settings → LLM Provider → Ollama

Ollama Base URL: http://hub.home.browngregory.com/ollama
# Uses the hub's Ollama proxy — routes to best cluster node.
# Then pick any model from the dropdown.

🌐

Open-WebUI

Full-featured web chat interface

Ollama API

Docker env / Settings → Connections

OLLAMA_BASE_URL=http://hub.home.browngregory.com/ollama
# All models on all live nodes appear in the model selector.

🔗

Python (openai library)

Direct API calls from scripts

OpenAI API

Script on this machine — point at the daemon

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:42420/v1",
    api_key="none",
)
response = client.chat.completions.create(
    model="qwen3:8b",
    messages=[{"role":"user","content":"hello"}],
)

Always-on service on the LAN — point at the hub

client = OpenAI(base_url="http://hub.home.browngregory.com/v1", api_key="none")

🐚

httpx / curl (any HTTP client)

Scripts, agents, custom integrations

OpenAI API

Set once as env vars, use everywhere

export OPENAI_BASE_URL=http://hub.home.browngregory.com/v1
export OPENAI_API_KEY=none

# Then in your code, just use the SDK or call the URL directly.
# X-Homelab-Source lets the hub attribute the request to your service:
import httpx
resp = httpx.post(
    f"{OPENAI_BASE_URL}/chat/completions",
    headers={
        "Authorization": f"Bearer {OPENAI_API_KEY}",
        "X-Homelab-Source": "my-service",
    },
    json={"model": "qwen3:8b", "messages": [...]},
)

Quick test

Check hub health

curl http://hub.home.browngregory.com/health
# → {"ok":true,"port":42421,...}

List available models

curl http://hub.home.browngregory.com/v1/models | python3 -m json.tool

Run a chat completion

curl http://hub.home.browngregory.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3:8b","messages":[{"role":"user","content":"hello"}]}'