# Docs


- **Get Started**
- [Introduction](/docs): A TypeScript SDK that lets users sign in with their ChatGPT account so your app can stream AI responses on their plan.
- [Quickstart](/docs/quickstart): Install the SDK, mount the server handler, render the sign-in button, and stream a response.
- [For AI agents](/docs/ai): llms.txt, per-page markdown, and how to point coding agents at these docs.
- Concepts
  - [How it works](/docs/concepts/how-it-works): The device-code login flow, the session state machine, and why polling is serverless-safe.
  - [Sessions & tokens](/docs/concepts/sessions): The session cookie, the session store, token encryption, and refresh mechanics.
  - [Response proxy](/docs/concepts/response-proxy): What happens between streamText() in the browser and the ChatGPT-backed Codex endpoint.
  - [Model discovery](/docs/concepts/models): Treat the model list as account data, pick a sensible default, and survive client_version drift.
  - [Security model](/docs/concepts/security): What a signed-in app can and cannot do, which guardrails ship by default, and where your obligations start.
- Guides
  - [Build a chat page](/docs/guides/chat-app): A working chat UI with sign-in, a model picker, and streamed responses, in one component.
  - [Custom sign-in UI](/docs/guides/custom-ui): Build your own login UI with useLoginWithChatGPT() while keeping the consent obligation.
  - [Cross-origin setup](/docs/guides/cross-origin): Run the frontend and the handler on different origins with allowedOrigins, SameSite=None, and credentialed fetches.
  - [Fast tier & thinking effort](/docs/guides/fast-and-reasoning): Offer Codex Fast mode and a reasoning-effort control through validated request headers.
  - [Headless & CLI logins](/docs/guides/headless): Run the device-code flow without a browser using the core primitives, then stream with the direct provider.
  - [Production checklist](/docs/guides/production): Everything to set, review, and verify before real users sign in.
- Reference
  - [Server handler](/docs/reference/server): createChatGPTHandler() options, defaults, and the getSession / proxyFetch / getModels helpers.
  - [HTTP routes](/docs/reference/routes): Every route the handler serves, with request/response shapes, status codes, and CSRF rules.
  - [React](/docs/reference/react): The LoginWithChatGPT widget, the useLoginWithChatGPT hook, and the consent popup helper.
  - [AI SDK providers](/docs/reference/ai): createChatGPTProxyProvider for browser/server proxy mode, createChatGPT for direct token mode, and the shared provider API.
  - [Core primitives](/docs/reference/core): ChatGPTConfig, exported defaults, and the device-flow, token, and transport primitives in @opencoredev/loginwithchatgpt-core.
  - [Error codes](/docs/reference/errors): Every ChatGPTAuthError code, the HTTP error shapes, and ChatGPTProxyError.
  - [Troubleshooting](/docs/reference/troubleshooting): Diagnose login, session, model, and streaming failures by symptom.