Built for AI Agents

Networking your
agents can run themselves

Holesail is a single command install with zero configuration, no accounts, and no dashboards to click through. That makes it the rare networking tool an autonomous coding agent can install, invoke, and tear down entirely on its own — inside a sandbox, a container, or a CI job.

Install
$ npm install -g holesail
Share a local service
$ holesail --live 3000
Connect to a shared key
$ holesail hs://s0003a3d2c1774c336e2c18828662b7b7c66

Why agents get stuck on networking

Most tunneling tools assume a human is in the loop: sign up, verify an email, open a dashboard, copy a token, edit a config file. An agent operating in an ephemeral sandbox usually can't do any of that.

One install, no signup
An agent can install Holesail with a single package-manager command and start using it immediately. No account creation, no email verification, no API key to fetch first.
  • npm, no login required
  • Runs the same in CI as on a laptop
  • Nothing to provision before use
Fully scriptable CLI
Every action is one shell command with a plain-text result — perfect for an agent that reasons in terms of commands and stdout, not clicking through a UI.
  • Single command to share a port
  • Single command to connect
  • Deterministic, parseable output
Works inside sandboxes
Agent runtimes are often containerized and sit behind NAT, CGNAT, or restrictive firewalls with no public IP. Holesail's P2P layer traverses that automatically.
  • No port forwarding
  • No public IP needed
  • Bypasses CGNAT and firewalls
Zero configuration
There's no config file for an agent to generate or maintain, and nothing that can silently drift out of sync between environments.
  • No YAML or JSON to write
  • No Dynamic DNS setup
  • Same command everywhere
Private by default
Connections are end-to-end encrypted and hidden behind a private key the agent generates itself, so sharing a service doesn't mean making it public. Unlike tunnels that hand out a guessable subdomain, a Holesail link is never crawled or indexed by search engines.
  • Key-based access only
  • End-to-end encrypted
  • Not discoverable or indexed by Google
Agent-to-agent tunnels
Two agents — or an agent and a human reviewer — can connect directly, letting one hand the other a live key to reach a running service instantly.
  • Share a live dev server
  • Hand off a running task
  • Direct peer connection, no relay

Built for agent workflows

A few of the ways autonomous coding agents use Holesail today

Preview a build for review
An agent finishes a task inside a sandbox, starts the dev server on localhost, and runs holesail --live 3000 to hand a reviewer a working link — without deploying anywhere or waiting on a hosting provider.
Share a local API with another tool
An agent testing an integration can tunnel its local API to a webhook provider or a second agent, closing the loop on end-to-end tests that would otherwise need a public deployment.
Reach services behind restrictive networks
CI runners and cloud sandboxes are frequently locked behind egress rules and CGNAT. Holesail's P2P NAT traversal lets an agent connect out without any network changes.
Hand off between agents
One agent can share a live key with another agent (or orchestrator) to pass along a running process, a database, or a long-lived task without either one leaving its sandbox.

No human required

What it takes for an agent to share a local service, start to finish

Typical tunneling service
  • Create an account and verify email
  • Generate and store an API token
  • Install and authenticate a CLI
  • Hope the sandbox's network allows outbound access to the relay
  • Watch for rate limits on the free tier
Holesail
  • npm install -g holesail
  • holesail --live 3000
  • Share the generated key

Give your agent a network layer

Drop Holesail into your agent's toolchain and let it share, connect, and hand off on its own.

Download NowView Documentation