Tailscale Integration¶

Secure remote access to local LLMs via Tailscale Serve and MagicDNS.

Overview¶

Tailscale provides:

Zero-config VPN - Mesh network with WireGuard
MagicDNS - Automatic DNS for all devices
Tailscale Serve - HTTPS proxy for local services
ACLs - Fine-grained access control
Funnel - Optional public internet access

Prerequisites¶

Tailscale installed on server and clients
Tailnet configured
LLM server running on localhost

See Tailscale Installation for setup.

Tailscale Serve¶

Basic Setup¶

Expose Ollama to your Tailnet:

# Expose port 11434 as HTTPS
tailscale serve --bg https+insecure://localhost:11434

# Check status
tailscale serve status

The service is now available at https://your-hostname.your-tailnet.ts.net

With Custom Path¶

# Expose under /api path
tailscale serve --bg --set-path /api https+insecure://localhost:11434

Multiple Services¶

# Ollama on /ollama
tailscale serve --bg --set-path /ollama https+insecure://localhost:11434

# llama.cpp on /llama
tailscale serve --bg --set-path /llama https+insecure://localhost:8080

# Open WebUI on root
tailscale serve --bg https+insecure://localhost:3000

Persistent Configuration¶

Using tailscale serve config file:

# Show current config
tailscale serve status --json

# Reset config
tailscale serve reset

The configuration persists across reboots when using --bg.

DNS Names¶

MagicDNS¶

Access via MagicDNS name:

# Format: hostname.tailnet-name.ts.net
curl https://server.tail12345.ts.net/v1/models

# Or just hostname within tailnet
curl https://server/v1/models

Find Your Hostname¶

# Show tailscale status
tailscale status

# Your device shows as:
# 100.x.x.x    server    your-email@   linux   -

Access Control Lists (ACLs)¶

Basic ACL¶

Restrict access to specific users:

{
  "acls": [
    {
      "action": "accept",
      "src": ["user1@gmail.com", "user2@gmail.com"],
      "dst": ["server:*"]
    }
  ]
}

Tag-Based Access¶

{
  "tagOwners": {
    "tag:llm-server": ["autogroup:admin"],
    "tag:ai-user": ["autogroup:admin"]
  },
  "acls": [
    {
      "action": "accept",
      "src": ["tag:ai-user"],
      "dst": ["tag:llm-server:11434"]
    }
  ]
}

Apply tag to server:

tailscale set --advertise-tags=tag:llm-server

Group-Based Access¶

{
  "groups": {
    "group:developers": ["dev1@example.com", "dev2@example.com"],
    "group:ai-team": ["ai1@example.com", "ai2@example.com"]
  },
  "acls": [
    {
      "action": "accept",
      "src": ["group:ai-team"],
      "dst": ["tag:llm-server:*"]
    }
  ]
}

Tailscale Funnel¶

For public internet access (use carefully):

Enable Funnel¶

# Expose to internet
tailscale funnel 11434

# With background mode
tailscale funnel --bg 11434

Funnel Considerations¶

Aspect	Consideration
Security	Add authentication layer
Cost	No bandwidth limits but consider model costs
Abuse	Rate limiting essential
Privacy	Requests come from internet

Recommended Funnel Setup¶

# Put reverse proxy with auth in front
tailscale funnel 8080  # Proxy port, not direct Ollama

# Proxy handles auth, rate limiting

Client Configuration¶

Environment Variables¶

# For tools expecting OpenAI API
export OPENAI_API_BASE=https://server.tail12345.ts.net/v1
export OPENAI_API_KEY=not-needed

# Or Ollama-specific
export OLLAMA_HOST=https://server.tail12345.ts.net

Tool Configuration¶

# Aider
aider --openai-api-base https://server.tail12345.ts.net/v1

# Continue.dev (config.json)
{
  "models": [{
    "apiBase": "https://server.tail12345.ts.net/v1"
  }]
}

Python Client¶

from openai import OpenAI

# Use Tailscale DNS name
client = OpenAI(
    base_url="https://server.tail12345.ts.net/v1",
    api_key="not-needed"
)

response = client.chat.completions.create(
    model="llama3.3:70b",
    messages=[{"role": "user", "content": "Hello!"}]
)

Docker Integration¶

Tailscale Sidecar¶

version: '3.8'

services:
  tailscale:
    image: tailscale/tailscale:latest
    container_name: tailscale
    hostname: llm-server
    environment:
      - TS_AUTHKEY=${TS_AUTHKEY}
      - TS_STATE_DIR=/var/lib/tailscale
      - TS_SERVE_CONFIG=/config/serve.json
    volumes:
      - tailscale-state:/var/lib/tailscale
      - ./serve.json:/config/serve.json:ro
    cap_add:
      - NET_ADMIN
      - SYS_MODULE
    restart: unless-stopped

  ollama:
    image: ollama/ollama
    network_mode: service:tailscale
    volumes:
      - /tank/ai/models/ollama:/root/.ollama
    depends_on:
      - tailscale

// serve.json
{
  "TCP": {
    "443": {
      "HTTPS": true
    }
  },
  "Web": {
    "${TS_CERT_DOMAIN}:443": {
      "Handlers": {
        "/": {
          "Proxy": "http://127.0.0.1:11434"
        }
      }
    }
  }
}

Troubleshooting¶

Can't Connect¶

# Check Tailscale is running
tailscale status

# Check Serve is configured
tailscale serve status

# Test local connection first
curl http://localhost:11434/

# Test Tailscale connection
curl https://your-server.tail12345.ts.net/

Certificate Errors¶

# Tailscale Serve handles certs automatically
# If issues, check:
tailscale cert your-server.tail12345.ts.net

# Verify HTTPS
curl -v https://your-server.tail12345.ts.net/

ACL Blocking Access¶

# Check ACL status
tailscale debug acl

# Verify your identity
tailscale whois $(tailscale ip -4)

Slow Connection¶

# Check for direct connection
tailscale netcheck

# Verify not relaying
tailscale status
# Should show "direct" not "relay"

Monitoring¶

Access Logs¶

# Tailscale doesn't log by default
# Monitor Ollama logs instead
journalctl -u ollama -f

# Or Docker
docker logs -f ollama

Connection Status¶

# Active peers
tailscale status

# Network quality
tailscale ping server

Security Best Practices¶

Use ACLs - Don't allow unrestricted access
Avoid Funnel for LLMs - Keep private unless needed
Monitor usage - Check logs for unusual patterns
Limit endpoints - Only expose what's needed
Use tags - Easier to manage than user lists