--- title: "Railway plugin for ChatGPT, dev.new preview, new observability panels" date: 2026-07-31T18:10:00.000Z number: 0301 url: https://railway.com/changelog/2026-07-31-railway-plugin-for-chatgpt --- # Railway plugin for ChatGPT, dev.new preview, new observability panels ChatGPT learns to drive a train this week: the official Railway plugin is live, so you can deploy and operate your projects from a chat window. We're also sharing an early preview of [dev.new](http://dev.new), a new way to go from prompt to production on Railway. And the Observability dashboard picks up six new panels, synced cursors, and drag-to-zoom. One more change worth calling out: the four core database templates (Postgres, MySQL, MongoDB, and Redis) no longer ship with a public TCP proxy. New databases are private by default, and you can add public access with one click when you need it. Let's get into it! 🚄 # Railway plugin for ChatGPT [Installing the Railway plugin in ChatGPT/Codex](https://cms.railway.com/media/ChatGPT-plugin.mp4) Most of our agent integrations live where code lives: an editor, a terminal, a repo. This one lives where the questions happen. The official [Railway plugin](https://chatgpt.com/plugins/plugin_asdk_app_6a502589384081919c5decf93496c9d1) is live in ChatGPT, and it turns a conversation into a way to run your infrastructure. Install it, connect your Railway account, and put the Railway Agent to work. Ask for an API with a Postgres database behind it, and the agent plans the work, provisions the services, and comes back with a deployed app. It helps you debug, too: "Why is checkout returning 502s?" pulls the HTTP logs and deployment status, and "Set LOG\_LEVEL to debug and redeploy" updates the variable and ships it. Under the hood, the plugin is powered by Railway's hosted MCP server at `mcp.railway.com`, and it bundles the Railway Agent skills, the same operating knowledge our other agent integrations ship with, so ChatGPT knows how to work on Railway instead of guessing. It acts with your personal access level, so ChatGPT can only see and change what you already can. ChatGPT joins [Claude Code](https://claude.com/plugins/railway), [Cursor](https://cursor.com/marketplace/railway), and [Grok](https://github.com/xai-org/plugin-marketplace) on the list of agents with a first-class Railway plugin. Try it out, and let us know what you think on [Central Station](https://station.railway.com/new?type=feedback). # dev.new preview [dev.new](https://cms.railway.com/media/dev-new.mp4) Every Railway project starts with code that already exists, written by you or, increasingly, your agent. \ \ Agents now handle the operating too: many of you ask yours to deploy services, provision resources, or troubleshoot issues (esp. with that new plugin).\ \ But if all you have is an idea, no code, no repo, there hasn't been a way to build it on Railway directly. We thought, can we make a good zero to one experience, that takes advantage of our platform that helps you scale from one to one million? [dev.new](https://dev.new) is that missing piece, now in preview: a zero to deployed experience on the platform. Log in with your Railway account and describe what you want to build. The Railway Agent builds the app while you watch it take shape in a live preview. And because it's the Railway Agent under the hood, the same chat handles more than UI tweaks: ask it to add a database, change the domain, or wire up variables. When you're happy with what you have, publish, and the app deploys into your Railway account. Build a one-off internal tool, or grow the same app into a production system, ultra-fast, using our new in-place builder. Either way, there's no migration cliff between the prototype and the real thing. You'll notice [dev.new](https://dev.new/) is a separate experience from the Railway dashboard, that said, your dev.new projects are Railway Projects under the hood, use the agent or go to the dashboard to perform any advanced configuration. Down the line, the goal is to fold everything into one unified experience. The preview is early and we're actively collecting feedback, so expect some rough edges. Tell us what you build and what breaks in this [Central Station thread](https://station.railway.com/feedback/feedback-dev-new-72c56c4e) or on [Discord](https://discord.gg/railway). # New panels in the Observability dashboard [Observability panels for HTTP metrics + HTTP & DNS logs](https://cms.railway.com/media/O11y-HTTP-panels-DNS-1.mp4) The **Observability** tab in your project is a customizable dashboard. Click **Add block** and you'll find six new panel types: - Three HTTP metric panels: **HTTP Requests** (request counts as stacked bars, broken down by status class), **HTTP Latency** (p50, p90, p95, and p99 lines), and **HTTP 5xx Rate** (the share of requests failing with server errors). - Three log panels: **HTTP Logs**, **DNS Logs**, and **Network Logs**, live log streams pinned to your dashboard. Each log panel supports the same filter syntax as the log explorer, and the filter is saved with the panel. Point an **HTTP Logs** panel at `@httpStatus:404` and you have a standing view of the pages users are hitting that don't exist. The HTTP panels scope to a single service, while **DNS Logs** and **Network Logs** can cover every service in the environment. Charts across the dashboard and metrics views also learn a few new tricks: - **Synced cursors**: hover over any graph and a cursor appears at the same moment in time on every other graph, with a tooltip showing each one's values. Correlating a latency spike with a memory spike no longer means eyeballing two x-axes. - **Custom time ranges**: the time picker now accepts any start and end time, not only the presets. Enter natural language like "2 hours ago", shorthand like `15m`, or pick a date from the calendar. Setting an end time freezes the window, so live updates don't pull the charts forward while you study an incident. - **Drag to zoom**: hold `Shift` and drag across a graph to zoom every panel to the selected window. Your browser's back button returns you to the previous range. Previously, the dashboard covered resource metrics and runtime logs, HTTP data lived in each deployment's own view, and time ranges came as a fixed set of presets. Incidents rarely align themselves with a preset. Now you can zoom to the five minutes that matter and see every signal side by side. Let us know what you think on [Central Station](https://station.railway.com/new?type=feedback). # Fixes and improvements - We fixed the template editor canvas locking up after deleting a service. The page previously froze in a render loop and had to be reloaded.