Layercode CLI

The CLI ships with tunneling that exposes your local backend to a public URL. That means you can test a real voice agent on your machine, on a real phone call, without touching cloud infrastructure. No DNS. No certificates. It removes the usual fiddly step of setting up reverse proxies just to hear your agent talk.

Instead of starting from an empty project, you get starter backends that already wire up the audio pipeline and agent loop. What does that save you? A weekend of debugging, usually. They give you a known-good baseline to modify, which is faster than building from scratch. You keep full control of the code, so nothing is locked into a black box.

Deployments spread across edge servers in multiple regions. For voice, that matters more than for a normal web app: a few hundred milliseconds of network delay makes a conversation feel unnatural, so keeping the compute close to whoever is speaking matters more than raw throughput. Running closer to the caller keeps the back-and-forth responsive. Latency is the whole game here.

Everything runs through terminal commands, so the setup can be scripted, versioned, and repeated. Teams that automate their deployment pipelines can fold the CLI into existing tooling rather than adopting a separate dashboard. It fits developers who'd rather type a command than click through onboarding screens. That's the whole pitch.

The combined tunnel, sample backend, and deploy path is built to get you to a working call quickly. The pitch is minutes, not days. For a simple agent that holds up. You spend your time on the agent's behavior instead of the plumbing underneath it.

You connect your own speech recognition, language model, and text-to-speech providers. That keeps the tool useful as pricing and model quality shift, and it means you aren't tied to one vendor's stack. Swapping a provider is a configuration change rather than a rewrite. No lock-in.