You could be beachside.

Your agents are out there doing the work. Need another one? Describe it in a sentence and it stays on the job, around the clock.

WORKING NOW23 running
  • opencode-workspaceFreestyle · remote dev workspaceup 3d 4h
  • claude-code-workspaceboxd · remote dev workspaceup 11d 4h
  • perf-hill-climbDaytona · spawns agents to hill-climb×14up 1d 7h
  • pr-reviewArchil · wakes for each PR×6running
For your agent

Ask your agent for a machine. It makes one.

Claude Code, Codex or any agent that reads skills can start a sandbox, launch Herdr or Rex in it, push your checkout and hand the long job off, then clean up when it's done. You watch from Herdr's sidebar or a Rex tab, even from your phone.

First, teach your agent. Two commands:
$ npm install -g @beachside/cli
$ beachside skill setup
Installs the Beachside skill for Claude Code, Codex and other agents that read skills.
  • Stable JSON output and exit codes
  • An optional sandbox cap so a runaway agent can't run up a bill
  • Optional TTLs, and rm --all to sweep up
agent session · excerpt

Spin up a Herdr box on Freestyle and run the integration tests there. I'm heading out.

$ beachside create --name tests --ttl 4h --with herdr --json
{ "name": "tests", "provider": "freestyle", "state": "running",
  "addons": [{ "name": "herdr", "state": "ready", … }] }
$ hc machine add ubuntu@beachside-tests.tail1234.ts.net --temp --install --name tests
$ beachside push tests
/home/ubuntu/work

Done. The tests are running in tests on Freestyle. It shows up in your Herdr sidebar as tests, and it removes itself in 4 hours.

Everything around the sandbox, already built.

The parts every agent team ends up rebuilding: credentials, tracing, workspaces, CI and cleanup. Open source and working today.

One SDK, one CLI

A typed TypeScript API and a CLI made for agents: stable JSON output and exit codes, sandbox caps, time limits, and one command to sweep up. Apache-2.0.

Tracing, metrics next

Every sandbox call is an OpenTelemetry span, sent to Sentry, Datadog or any OTLP backend you already run. Metrics are coming alongside tracing.

Credential broker

Real keys never enter the sandbox. Each sandbox gets a short-lived grant, 15 minutes by default, revoked when it's destroyed. GitHub Actions signs in with OIDC workload identity, so CI holds no long-lived secrets.

Model gateways

Point sandbox model clients at OpenRouter, Kilo AI Gateway, Cloudflare AI Gateway, Vercel AI Gateway, OpenAI, Anthropic or your own HTTPS endpoint, all through the broker.

Herdr and Rex workspaces

One command starts a Herdr or Rex workspace in a sandbox and joins it to your tailnet with Tailscale. Open it from your Mac or phone. It keeps running when your laptop sleeps.

CI in fresh sandboxes

Run every CI job in its own sandbox, from GitHub Actions or any CI system, with warm bases, time limits and many jobs at once.

Snapshots, forks and volumes

Capture and restore a sandbox, fork a running one, and keep data on volumes between sandboxes.

Previews and SSH

Reach a service in a sandbox through protected or expiring preview links, or connect with a standard SSH client.

Runs on the providers you already use.

Today you pick the provider. Soon, Beachside can help fail over between providers, suggest the best provider for your workload, and more.

SUPPORTED

EXPERIMENTAL

On the road to 1.0.

What's coming next. Beta members help decide the order.

  1. 01MetricsUsage and health metrics next to the traces you already get.
  2. 02Provider failoverIf a provider fails, carry on with another.
  3. 03Workload-aware provider picksBeachside suggests the best provider for each workload.
  4. 04Orca supportOrca workspaces in a sandbox, on whichever provider you pick.
  5. 05Pi extension and durable sessionsOne sandbox per Pi session, still going when you close your laptop.
  6. 06Artifact handlingGet the files your agents make back out of the sandbox.
  7. 07Protected, shareable URLsShare an app in a sandbox behind a login or token, not an open link.
  8. 08Config and env syncingThe same dotfiles, agent config and env vars in every sandbox.
  9. 09Workspace and config backupsBack up a sandbox's workspace and config, and restore them later.
  10. 10Agent inboxes and message boardsAgents in a swarm message each other and post to shared boards.
  11. 11Persistent memory serviceMemory your agents keep across sandboxes and sessions.
  12. 12Webhook trigger gatewayStart an agent when a webhook fires, like a new issue or pull request.
  13. 13Lifecycle and cost guardrailsCaps, TTLs and spend tracking across providers.
  14. 14More providersNew adapters, with the same CLI and SDK on every one.

Laptop closed. Agents still working.

The beta is open for sign-ups. We'll send an invite when your spot is ready.

Start with the SDK