While there might be a GPU shortage, there certainly isn’t a shortage of coding agents. Whether it’s Claude Code, Codex, OpenCode, Gemini CLI, GitHub Copilot CLI, Droid, Pi, or the next amazing agent harness you just read about on HN, there always seems to be a new one just around the corner.

Developers must track not only new models themselves, but also how those models perform across different harnesses, since benchmarked performance can vary dramatically. It’s not too hard to try a specific harness with a couple of different models, but with more than a few options it turns into a large M-by-N-sized mess to deal with every time you want to switch.

This complexity can stifle experimentation, even among the most curious of developers. But it is critical, especially during a time of such rapid change in the industry.

We know how important it is to encourage experiments while keeping strong controls in place. That’s why we built Aperture, an LLM gateway that centralizes credentials, routes requests through one control point, ties usage to identity, and provides visibility into usage and costs across an organization. Developers can keep experimenting without managing every credential and configuration by hand.

Today, we’re making things even easier by releasing Aperture CLI to help developers quickly configure, try, and test AI agents in a simpler, more secure way.

One CLI, any agent

Aperture CLI isn’t a coding agent. Rather, it knows how to configure major coding agents with supported providers available in Aperture’s LLM gateway. On launch, Aperture CLI automatically discovers available Aperture instances connected to the same Tailscale tailnet. It also finds configured LLM endpoints in the gateway. It pairs that provider and model list with the coding agents installed on your system. From there, you can swap agents, endpoints, and models in just a few keystrokes.

Use Aperture CLI without the Tailscale client

In addition to making it simple to switch agents, we’ve set up Aperture CLI so it works on devices where the traditional Tailscale VPN client cannot be installed, or where switching tailnets is a barrier.

Aperture CLI includes a bridge mode, which connects directly to the tailnet that contains your Aperture instance. After it connects to Tailscale, just its proxy connection appears as a node on your tailnet, rather than the entire machine. From then on, Aperture CLI will bridge the Aperture instance to a local port on the machine, and configure any coding agent to use it.

Using bridge mode simplifies the process of rolling out Aperture to your team or organization, since it works alongside an existing VPN installed on the machine without interference.

Try Aperture and Aperture CLI today

If you’re already an Aperture user, installing the Aperture CLI alpha is as simple as running go install github.com/tailscale/aperture-cli/cmd/aperture@latest and then aperture. While it's currently in alpha, we have tested it across macOS, Linux, and Windows. If you run into any problems, please let us know by creating a GitHub issue.

If you don’t already use Aperture (or Tailscale), you’ll need to sign up for an Aperture instance here and add a hosted instance to your tailnet first. Aperture is available for free on the Tailscale Personal plan.