--- title: "Railway Sandboxes, use your ChatGPT subscription with Railway Agent, MySQL PITR" date: 2026-09-17T21:05:00.000Z number: 0308 url: https://railway.com/changelog/2026-09-18-railway-sandboxes --- # Railway Sandboxes, use your ChatGPT subscription with Railway Agent, MySQL PITR Your agents have work to do. We've got computers for them. This week, Railway Sandboxes are generally available. You can also now use your existing ChatGPT subscription to power Railway Agent. We're also bringing point-in-time recovery to MySQL. Both features are available to try in [Priority Boarding](https://railway.com/account/feature-flags). Let's get into it! 🚄 # Railway Sandboxes are generally available [Run, checkpoint, and fork Linux VMs alongside your Railway infrastructure.](https://cms.railway.com/media/cffd275c-678f-446c-af59-69f41180a4bb-Railway_draft2-2-social.mp4) Railway Sandboxes have graduated from Priority Boarding and are available on every plan! Create an isolated Linux VM for a coding agent, a test run, or code execution in your application. Run commands, work with files, and destroy the sandbox when the task is done. You can manage the whole lifecycle through the [TypeScript SDK](https://docs.railway.com/sandboxes#typescript-sdk), [CLI](https://docs.railway.com/cli/sandbox), dashboard, or API. Claude Code, Codex, OpenCode, and Pi come preinstalled, along with tools like Node.js, npm, and Git. Save a checkpoint of your prepared files, fork a sandbox to try several approaches in parallel, or connect it to your environment's private network to work with your existing services and databases. Open your project's **Sandboxes** tab or run `railway sandbox create` to get started. Usage draws from the same included credit as your services. Check out [Railway Sandboxes](https://railway.com/sandboxes), follow the [quickstart](https://docs.railway.com/sandboxes/quickstart), and tell us what you're building on [Central Station](https://station.railway.com/new?type=feedback). # Use your ChatGPT subscription with Railway Agent [Connect a ChatGPT subscription and select a model in Railway Agent](https://cms.railway.com/media/49c08aea-d63f-4f6b-8fee-3df9295573bd-Sign-In-with-ChatGPT.mp4) Your ChatGPT subscription can now power Railway Agent. Try it in [Priority Boarding](https://railway.com/account/feature-flags). Connect your account to use models like GPT-6 Astra, GPT-5.6 Sol, and GPT-5.6 Terra in the Railway Agent TUI and the agent in your Railway dashboard. When you select a subscription model, usage counts toward your ChatGPT plan instead of Railway's model billing. Your plan's model availability and usage limits apply. Open **Account Settings → Agents** and click **Connect** under **OpenAI Subscription**. Complete sign-in, then select a model from the **OpenAI Subscription** group. Connecting your account doesn't change your selected model. Your connection follows you across workspaces, and you can disconnect it from the same settings page. Give it a try and share your feedback on [Central Station](https://station.railway.com/new?type=feedback). # MySQL point-in-time recovery [Choose a timestamp and restore MySQL into a separate service](https://cms.railway.com/media/7677ecc2-20c9-41d9-93e6-f0d2a794f472-mysql-PITR.mp4) Recover MySQL to the moment before a bad migration or accidental delete. MySQL point-in-time recovery lets you recover to a timestamp within your available archive window, and it's available in Priority Boarding for standalone services and MySQL HA clusters. Open your database's **Backups** tab and click **Enable PITR**. Railway configures a storage bucket, takes a full backup, and continuously archives MySQL's binary logs. The archive retains 7 days of history, with coverage starting from the first backup after you enable it. Enabling PITR redeploys the database. To recover, select a timestamp and click **Restore to this moment**. Railway creates a separate MySQL service and replays the archive to that point. Your original database keeps serving traffic while you inspect the restored data and decide when to switch over. Read the [point-in-time recovery guide](https://docs.railway.com/volumes/point-in-time-recovery), try it through [Priority Boarding](https://railway.com/account/feature-flags), and share your feedback on [Central Station](https://station.railway.com/new?type=feedback). # Fixes and improvements - We improved the template editor so you can switch an existing service between a GitHub repository and a Docker image while keeping its variables, volumes, and settings. Previously, changing source types required recreating the service. - We improved build failure messages for invalid variable names. The deployment now explains that a variable needs renaming and how to fix it, replacing a generic build error.