How to Connect AI Agents Securely with Tailscale

TL;DR

Tailscale connects computers and VPS instances through a private encrypted network, allowing one AI agent to reach and manage agents on other machines without exposing ports to the public internet. Install it on each device, authenticate through an identity provider such as GitHub, verify the connection with terminal commands, and use access control lists to limit which machines each agent or person can reach.

Transcript

My name is David Andre and here's how to use tail scale with AI agents. So first, what even is tail scale? Tail scale is a private network that connects all your machines together. And this matters a lot in the AI era because it allows a single agent to control any number of other agents, no matter if they're local on your computer or on a VPS. And... Read More

Key Insights

• Tailscale is a private network that connects local computers and VPS instances, allowing AI agents on separate machines to communicate and be managed together. The presenter positions this shared network as a practical way to reduce repetitive maintenance across increasingly common multi-agent environments.
• A central AI agent can manage agents running on other connected machines, regardless of whether they are local or hosted on separate VPS instances. This arrangement can reduce manual work when a remote agent crashes, needs a model change, or requires another routine intervention.
• Private encrypted tunnels allow connected machines to communicate without exposing ports to the public internet. The presenter describes every open port as a potential point of exposure because automated bots continually scan internet-facing systems, making a private networking layer valuable for VPS protection.
• Tailscale traffic travels directly between connected machines in a peer-to-peer arrangement rather than through an intermediary server. The service is described as being built on WireGuard, which the presenter says keeps communication fast when agents make many API calls across the private network.
• Identity-based access ties network membership to accounts such as Google or GitHub rather than requiring SSH keys for Tailscale access. If a laptop is stolen or an employee leaves, an administrator can revoke that specific device so it immediately loses access to the entire private network.
• Access control lists define which devices, agents, or team members may reach particular machines. Although every device can communicate with every other device by default, restrictive rules can contain the impact of a compromised computer, agent, or user by limiting its permitted destinations.
• The free Tailscale plan is presented as supporting unlimited devices and up to six users, making the networking layer free for many individuals and small teams. Separate infrastructure expenses still apply because users must pay for any VPS instances that host their remote agents.
• The basic setup starts at Tailscale's website, where the user signs in, adds a local device, selects Linux for the second machine, and receives an installation command. Terminal commands such as "tailscale up" and "tailscale status" are used to start the service and inspect its state.

Questions & Answers

Q: What is Tailscale for AI agents?

Tailscale is a private network that connects computers and VPS instances so AI agents running on different machines can communicate. In the setup described, an agent on a MacBook can command or repair agents on remote servers. Each connected machine appears in one console with a recognizable name and can be reached by that name from anywhere in the world.

Q: How does Tailscale simplify multi-agent management?

Tailscale places machines hosting different AI agents on one private network, allowing a single agent to reach the others. Instead of manually connecting to each VPS whenever an agent crashes, needs an update, or requires a model change, an operator can use an agent on one connected computer to manage the remote systems and reduce repeated manual intervention.

Q: How does Tailscale protect VPS instances?

Tailscale connects machines through private encrypted tunnels, so the VPS instances can operate without ports exposed to the public internet. The presenter describes open ports as points of exposure because bots continually scan internet-facing machines. Access can also be revoked for an individual device, and access control rules can prevent participants from reaching systems they do not need.

Q: How do Tailscale access control lists work?

Access control lists specify which computers, agents, or people may communicate with particular machines on the private network. By default, every device on the tailnet can talk to every other device. Rules can narrow that access, such as allowing an agent to reach only one VPS, which limits possible damage if a device, user, or agent is compromised.

Q: How do you begin setting up Tailscale on two machines?

Start at the Tailscale website, select the option to get started, and authenticate with an available identity provider such as GitHub. Add the local computer first, then choose Linux when adding a VPS. Tailscale provides an installation command for the Linux server. On the local machine, the presenter uses "tailscale up" and "tailscale status" to start and check the service.

Q: Does Tailscale require SSH keys for network access?

The Tailscale access described in the transcript is tied to a real identity, such as a Google or GitHub account, so SSH keys are not required for that network access. Traditional SSH is still used in the demonstration to make the initial connection from the MacBook to the VPS before Tailscale and the AI-assisted workflow are operated on the remote machine.

Q: What does the Tailscale free plan include?

According to the presenter, the free Tailscale plan includes unlimited devices and supports up to six different users. That allowance is described as sufficient for many solo users and small teams, so the networking layer can cost nothing in those cases. Users still pay separately for any VPS infrastructure used to run remote AI agents.

Q: Should AI agents receive VPS login credentials directly?

The presenter says giving VPS credentials directly to Codex would be faster but is not the best practice. In the demonstration, he connects to the VPS himself using SSH, confirms the connection, and enters the root password manually. He then asks Codex to operate the connected environment, keeping the initial credential entry under direct human control.

Summary & Key Takeaways

• Tailscale creates a private network connecting local computers, remote VPS instances, and the AI agents running on them. A central agent can then reach other agents by recognizable machine names from anywhere in the world, reducing the manual effort required to diagnose crashes, change models, or perform updates across a multi-agent setup.

• Security comes from private encrypted tunnels, identity-based authentication, device revocation, and configurable access control lists. Machines can operate without ports exposed to the public internet. Administrators can also restrict a person, computer, or agent to specific destinations, limiting potential damage if any single participant in the network becomes compromised.

• The demonstrated setup begins by creating a Tailscale account, adding the local computer, choosing Linux for a VPS, and running the provided installation command. The presenter connects to an Ubuntu VPS through SSH before asking Codex to help operate it, while recommending that users avoid directly sharing credentials with an AI agent.