Paperclip + Local AI Server Setup Guide

Run Paperclip with a remote AI server (Ollama) + local dev model (LM Studio)
Fully local, network-accessible, and agent-compatible.

Overview

This guide shows you how to:

• Run Paperclip locally
• Connect it to a dedicated AI server (Ollama)
• Add a local development model (LM Studio)
• Enable LAN access (no RDP required)
• Fix agent access issues caused by authentication

Architecture

Dev Machine (LM Studio)
↓
Paperclip Server (LAN Accessible)
↓
AI Server (Ollama - Heavy Models)

Requirements

Server Machine

• Ubuntu / Linux
• Node.js 20+
• Ollama installed

Dev Machine

• LM Studio running
• Network access to server

AI Agent Server

• Ubuntu / Linux

• pnpm

On AI Agent (Paperclip Server)

This is the system you have to run the Paperclip Application and Agents (For me this is a spare Intel Nuc Running Ubuntu 24 LTS)

Step 1 install opencode and build config:

npm install -g opencode-ai

check its running:

opencode --version

now you need to build to build a config.json for opencode to point to your AI services (make sure your AI server / Dev Server AI is online), navigate to

~/.config/opencode/opencode.json"

you will now need to build out your config to tell the server where to connect to, here is an exmaple for two options for opencode, a dedicated AI server and also a dev system. do this with nano and obviously in the config change to your ip's

nano config.json

use a config like the one below and then ctrl + X to save it.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "ollama-server": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Ollama Server",
      "options": {
        "baseURL": "http://192.168.101.162:11434/v1"
      },
      "models": {
        "qwen3.5:35b": {
          "name": "Qwen 3.5 35B"
        },
        "qwen3.5:9b": {
          "name": "Qwen 3.5 9B"
        }
      }
    },
    "lmstudio-dev": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "LM Studio Dev",
      "options": {
        "baseURL": "http://192.168.101.24:1234/v1"
      },
      "models": {
        "local-model": {
          "name": "LM Studio Model"
        }
      }
    }
  },
  "model": "ollama-server/qwen3.5:35b"
}

save that and then do a check your systems are online,

# Ollama:
curl http://192.168.101.162:11434/v1/models
# LM Studio:
curl http://192.168.101.24:1234/v1/models

Hopefully you will get a response from each call saying they can see the models, if not you need to check they are both bound to 0.0.0.0 and both systems have network access, im not going to say how as there is literally hunderds of simple guides on it.

Step 2 — Install pnpm

corepack enable
corepack prepare pnpm@latest --activate

step 3 — Install Paperclip

git clone https://github.com/paperclipai/paperclip.git
cd paperclip
pnpm install

Step 4 — Configure Server Mode

Run:

pnpm paperclipai configure --section server

Then Select:

pnpm paperclipai configure --section server

During setup:

Host:

0.0.0.0

his allows access from your network (LAN)

Allowed Hostnames:

Add BOTH:

127.0.0.1
<YOUR_SERVER_IP>

Example:

127.0.0.1,192.168.76.248

Why this matters
Without this:
Browser access → ❌ blocked
Agents → ❌ blocked
Paperclip enforces strict hostname validation.

Step 5 - Secure the System:

this is to restrict access to LAN only again check the IP for your network:

sudo ufw allow from 192.168.101.0/24 to any port 3100
sudo ufw deny 3100

The allow IPs to paperclip, this needs to be the current address the 192 address of this system and the 127.0.0.1, to make sure both agents and LAN have access to the system.

pnpm paperclipai allowed-hostname <IP>

Step 6 — Start the Server

Start the models on your systems Ollama or LMStudio

pnpm paperclipai run

Access from browser:

http://192.168.76.248:3100

Now you should be able to continue as intended build out your company and agents, make sure in integrations to select OpenCode and select the loaded model method or a static model that will be listed.

common issues:

❌ "Model not found"
Check:

• model name matches /v1/models
• config uses correct ID

❌ Agents cannot access API
Fix:

• disable auth mode
• use custom mode
• ensure hostname is allowed

Credits

Built through real-world setup + debugging.
This guide exists because the docs don’t cover this properly.