|Docs

Bring Your Own Models

Bring Your Own Models lets you run Railway's AI features on a model account you already have. Link your workspace's Anthropic Claude Platform organization, or sign in with your ChatGPT subscription or your OpenRouter account, then pick a model, and the Railway Agent uses it everywhere it runs:

Requests on your own account are billed by that provider, not by Railway. Railway models stay available alongside yours, so you can switch between them at any time.

Providers

Railway supports three providers, and each one is available in every workspace.

ProviderWhat you connectConnected for
Anthropic Claude PlatformYour workspace's Claude organizationEvery member of one workspace
ChatGPTYour ChatGPT subscriptionYour account, in every workspace
OpenRouterYour OpenRouter accountYour account, in one workspace

Get started

Turning on a provider takes two steps: connect the provider, then choose a default model.

1. Connect a provider

Navigate to Account Settings, then Agents, with the workspace you want to configure selected. Each provider has a card in the Providers section under Models.

The connection flow differs by provider:

2. Choose a default model

Select a model from Default Model on the same page. The list groups models by account: Railway, OpenAI Subscription, OpenRouter API Key, and Claude Platform.

The default model applies to Railway Agent chat, the Slack and Discord bots, and the cloud agents and dev.new apps you own in the selected workspace. The model picker in chat overrides it for a single message.

Where your models run

Connecting a provider makes its models available on every surface that runs the Railway Agent. A few rules decide which account a request uses.

Dashboard chat and chat bots

Railway Agent chat uses your default model unless you pick another one in the chat's model picker. The Slack and Discord bots use the default model of the person who mentioned @Railway.

Cloud agents and dev.new

Cloud agents running the Railway Agent harness, and apps on dev.new, use their owner's model settings and credentials. That holds even when another workspace member uses the agent or app, so a teammate's prompts to your agent run on your account.

Inside a cloud agent, your models appear in the harness's model list, labeled (ChatGPT), (Personal key), or (Claude Platform).

The harness talks to each provider over a specific API, which limits the OpenRouter models it can run. OpenRouter's OpenAI and Anthropic models are available in dashboard chat and the bots but not in cloud agents. To use GPT or Claude models in a cloud agent, use a linked Claude Platform organization, ChatGPT sign-in, or a Railway model.

Available models

The model list in Default Model is the authoritative list for your account. Railway curates the models offered for each provider:

ProviderModels
Anthropic Claude PlatformClaude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 5.5, Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5
ChatGPTGPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna
OpenRouterGPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna, Claude Opus 5.5, Claude Sonnet 5.5, Grok 4.7, Muse Spark 1.3, GLM 5.3, Kimi K3

Reasoning level is set under Default Reasoning Level when the selected model supports it.

Billing

Railway doesn't charge for tokens processed on your own account. Anthropic bills Claude Platform requests to your Claude organization, ChatGPT requests count against your ChatGPT plan, and OpenRouter requests are paid from your OpenRouter credits.

Railway models keep their standard pricing, so switching back to a Railway model resumes Railway billing for those requests.

Troubleshooting

Most problems show up as a status line on the provider's card in Agents settings, so start there. Each provider's page covers its own errors.