Remote IDE Setup

This guide explains how to connect your local IDE and CLI tools to an Ollama instance running on a remote DevBox server.

Table of Contents

• Overview
• Connection Methods
• IDE Configuration
• CLI Configuration
• Shell Functions
• Recommended Models
• Troubleshooting

Overview

By default, Ollama runs on 127.0.0.1:11434 (localhost only) on the remote server. This guide covers two methods to access it from your local machine:

1. SSH Port Forwarding: Best for temporary connections and maximum security
2. Tailscale Network Binding: Best for always-on access and convenience

Prerequisites

Before starting, you need your server's Tailscale IP:

# Run on your server
tailscale ip -4

# Or from your laptop via SSH
ssh your-server "tailscale ip -4"

Set the environment variable for all scripts:

export OLLAMA_SERVER_IP="100.x.x.x"  # Replace with your Tailscale IP

Connection Methods

Method 1: SSH Port Forwarding

Best for: Maximum security, temporary connections, testing.

Setup:

# Forward remote port 11434 to local port 11434
ssh -L 11434:127.0.0.1:11434 user@remote-server -p 5522 -N

# Run in background
ssh -L 11434:127.0.0.1:11434 user@remote-server -p 5522 -N -f

# Persistent tunnel (requires autossh)
autossh -M 0 -L 11434:127.0.0.1:11434 user@remote-server -p 5522 -N

Client Configuration:

• IDE endpoint: http://localhost:11434
• CLI: No configuration needed (uses localhost by default)

Method 2: Tailscale Network Binding

Best for: Always-on access, convenience, multiple clients.

Step 1: Configure Ollama Docker

Edit ~/docker/ollama-openwebui/docker-compose.yml on your server:

services:
  ollama:
    environment:
      - OLLAMA_HOST=0.0.0.0:11434
    ports:
      - "100.x.x.x:11434:11434"  # Replace with your Tailscale IP

Step 2: Restart Ollama

cd ~/docker/ollama-openwebui
docker compose down ollama && docker compose up -d ollama

Step 3: Verify Connection

curl http://100.x.x.x:11434/api/tags

Client Configuration:

• IDE endpoint: http://100.x.x.x:11434 (your Tailscale IP)
• CLI: export OLLAMA_HOST=http://100.x.x.x:11434

Security Comparison

Method	Security	Setup	Maintenance	Network Exposure
SSH Tunnel	High	Simple	Active tunnel required	Localhost only
Tailscale (IP-bound)	High	Moderate	None	Tailscale network only
Bind to 0.0.0.0:11434	Unsafe	Simple	None	All interfaces

Warning: Never use ports: - "11434:11434" without an IP prefix. This exposes Ollama on all network interfaces, including public IPs.

IDE Configuration

Zed Editor

Edit ~/.config/zed/settings.json:

For SSH Tunnel:

{
  "language_models": {
    "ollama": {
      "api_url": "http://localhost:11434",
      "low_speed_timeout_in_seconds": 120
    }
  },
  "assistant": {
    "version": "2",
    "default_model": {
      "provider": "ollama",
      "model": "qwen3"
    }
  }
}

For Tailscale:

{
  "language_models": {
    "ollama": {
      "api_url": "http://100.x.x.x:11434",
      "low_speed_timeout_in_seconds": 120
    }
  },
  "assistant": {
    "version": "2",
    "default_model": {
      "provider": "ollama",
      "model": "qwen3"
    }
  }
}

Automated Setup Script:

export OLLAMA_SERVER_IP="100.x.x.x"  # Your Tailscale IP
bash scripts/laptop/zed-setup.sh

Keyboard Shortcuts:

Shortcut	Action
Ctrl+Enter / Cmd+Enter	Open Assistant panel
Ctrl+Shift+A	Insert AI suggestion
/ in Assistant	Slash commands menu
Escape	Close Assistant

VS Code with Continue Extension

1. Install the Continue extension
2. Open Continue settings (Ctrl+Shift+P, "Continue: Open Settings")
3. Add Ollama as a provider:

{
  "models": [
    {
      "title": "Remote Ollama",
      "provider": "ollama",
      "model": "qwen3",
      "apiBase": "http://100.x.x.x:11434"
    }
  ]
}

Multiple Models Configuration

For IDEs that support multiple models:

{
  "language_models": {
    "ollama": {
      "api_url": "http://100.x.x.x:11434",
      "low_speed_timeout_in_seconds": 120,
      "available_models": [
        {
          "name": "qwen3",
          "display_name": "Qwen3",
          "max_tokens": 8192,
          "supports_tools": true
        },
        {
          "name": "codellama",
          "display_name": "Code Llama",
          "max_tokens": 4096,
          "supports_tools": false
        }
      ]
    }
  }
}

CLI Configuration

Ollama CLI

SSH Tunnel Method:

# No configuration needed - CLI uses localhost:11434 by default
ollama list
ollama run qwen3

Tailscale Method:

# Set environment variable
export OLLAMA_HOST="http://100.x.x.x:11434"

# Make permanent (add to ~/.bashrc or ~/.zshrc)
echo 'export OLLAMA_SERVER_IP="100.x.x.x"' >> ~/.bashrc
echo 'export OLLAMA_HOST="http://${OLLAMA_SERVER_IP}:11434"' >> ~/.bashrc
source ~/.bashrc

# Use normally
ollama list
ollama run qwen3

API Testing

# List available models
curl http://localhost:11434/api/tags | jq '.models[].name'

# Test generation
curl http://localhost:11434/api/generate -d '{
  "model": "qwen3",
  "prompt": "Hello, world!",
  "stream": false
}'

Shell Functions

Add these functions to your ~/.bashrc or ~/.zshrc for quick AI access:

# Quick ask - get just the answer
ask() {
    if [ -z "\$1" ]; then
        echo "Usage: ask \"your question here\""
        return 1
    fi
    local prompt="$*"
    curl -s "$OLLAMA_HOST/api/generate" -d "{
        \"model\": \"qwen3\",
        \"prompt\": \"$prompt\",
        \"stream\": false
    }" | jq -r '.response'
}

# Code-specific questions
askcode() {
    if [ -z "\$1" ]; then
        echo "Usage: askcode \"your code question\""
        return 1
    fi
    local prompt="You are a coding assistant. Answer concisely with code when appropriate. Question: $*"
    curl -s "$OLLAMA_HOST/api/generate" -d "{
        \"model\": \"qwen3\",
        \"prompt\": \"$prompt\",
        \"stream\": false
    }" | jq -r '.response'
}

# Chat with streaming
chat() {
    if [ -z "\$1" ]; then
        echo "Usage: chat \"your question\""
        return 1
    fi
    local prompt="$*"
    curl -s "$OLLAMA_HOST/api/generate" -d "{
        \"model\": \"qwen3\",
        \"prompt\": \"$prompt\",
        \"stream\": true
    }" | while IFS= read -r line; do
        echo "$line" | jq -r '.response // empty' | tr -d '\n'
    done
    echo ""
}

# List remote models
alias ollamals='curl -s $OLLAMA_HOST/api/tags | jq -r ".models[].name"'

Automated Setup Script:

export OLLAMA_SERVER_IP="100.x.x.x"
bash scripts/laptop/ollama-setup.sh

Recommended Models

Model	Size	Use Case	Tool Support
qwen3	Varies	General coding, agentic tasks	Yes
devstral	Medium	Code completion	Yes
codellama	7B-34B	Code generation	Limited
deepseek-coder-v2	16B-236B	Advanced coding	Yes

Note: Local models have limitations compared to cloud models (Claude, GPT-4) for complex agentic tasks requiring multi-step operations and file manipulation.

Troubleshooting

Port Already in Use (SSH Tunnel)

Diagnosis:

sudo lsof -i :11434

Solutions:

1. Stop local Ollama: sudo systemctl stop ollama
2. Use different local port: ssh -L 11435:127.0.0.1:11434 remote -N
3. Kill existing tunnel: pkill -f "11434:127.0.0.1:11434"

Connection Failures

Check network connectivity:

# Verify Tailscale
tailscale status
ping 100.x.x.x

# Test port accessibility
curl http://100.x.x.x:11434/api/tags

Verify OLLAMA_HOST:

echo $OLLAMA_HOST
# Should output: http://100.x.x.x:11434

Check tunnel status:

ps aux | grep "ssh.*11434"

IDE Not Detecting Models

1. Verify API connection: curl http://endpoint/api/tags
2. Restart IDE completely
3. Check IDE logs for connection errors
4. Verify firewall settings

Slow Performance

Local models require significant compute resources:

• Large models (30B+) are slow on CPU
• Verify GPU usage: docker logs ollama 2>&1 | grep -i gpu
• Consider smaller models for faster responses
• Increase timeout in IDE: "low_speed_timeout_in_seconds": 120

Zed-Specific Issues

"Failed to connect to language model":

1. Check Tailscale connection:

   curl http://100.x.x.x:11434/api/tags
   

2. Ensure Tailscale is running: tailscale status
3. Check Zed logs: Ctrl+Shift+P, type "zed: open log"

Models not showing up:

1. Restart Zed completely
2. Check Assistant Panel settings (open with Ctrl+Enter)
3. Manually refresh: curl http://100.x.x.x:11434/api/tags

References

• Zed LLM Providers
• Ollama Documentation
• Tailscale Documentation

---

Last updated: 2026-05-16 (v1.0.0)