Setup
Choose an application and follow the short setup. Create the key first under Dashboard -> API keys - it is shown once, at creation.
Automatic setup for Codex, Claude Code and OpenCode
One command connects Codex (CLI, desktop app and IDE extension) to the GPT models your plan includes, Claude Code (CLI and IDE extensions) to the Claude models, and OpenCode (CLI and desktop app) to every chat model. Install the ones you use first, then run it.
Open PowerShell (press the Windows key, type PowerShell, press Enter) and run:
irm https://gensite.tech/install.ps1 | iex
- It asks which of the tools it finds to set up: all of them, or any you pick. The ones you leave out are not touched.
- It asks for an API key from Dashboard -> API keys and stores it encrypted for your Windows account. The key is never written into the apps' settings.
- It lets you pick a default model for each, backs up your current settings first, and sends one short test message to confirm each connection.
- It offers to add a web browser the AI can drive: it opens pages, clicks, types and takes screenshots in a browser window of its own. This needs Node.js; say no to leave it out.
- Run it again at any time to change the key or refresh the model list.
Needs: Nothing extra: Windows PowerShell is built in. Read the script before you run it.
Undo the setup
Press the Windows key, type "Undo sitegen" and press Enter. Or run this in PowerShell:
irm https://gensite.tech/uninstall.ps1 | iex
It puts your previous Codex, Claude Code and OpenCode settings back exactly as they were and deletes the saved key.
The ChatGPT and Claude chat apps (desktop, web and phone) cannot be pointed at another server, so they keep using your own OpenAI and Anthropic accounts. For chat through sitegen, use a client from the list below that accepts a custom API address.
Claude Code (CLI / IDE) - Windows
Speaks the Anthropic Messages format, so its base URL is https://gensite.tech.
- 1
Quicker: the automatic setup at the top of this page does all of this for you, IDE extensions included. The steps below are the manual route.
- 2
Make sure Claude Code is already installed on this computer.
- 3
Point it at this gateway with three environment variables. Run these in PowerShell, then open a new terminal - setx only reaches shells started afterwards.
setx ANTHROPIC_BASE_URL "https://gensite.tech" setx ANTHROPIC_AUTH_TOKEN "sk_live_YOUR_API_KEY_HERE" setx ANTHROPIC_MODEL "claude-fable-5"
- 4
ANTHROPIC_AUTH_TOKEN is the variable to set, not ANTHROPIC_API_KEY: the gateway reads it as a bearer token.
- 5
Open a new terminal, run claude, and send a test prompt.
Download a local Markdown copy of this selection. It includes endpoint values and a placeholder only; it never includes a live API key.
Gateway values
The key is created under Dashboard -> API keys.https://gensite.tech/v1https://gensite.tech/v1/chat/completionshttps://gensite.tech/v1/responseshttps://gensite.techhttps://gensite.tech/v1/messagesgemini-3.1-proclaude-fable-5sk_live_YOUR_API_KEY_HEREUse the base URL when the client asks for a provider endpoint. Use a full path only when a client explicitly asks for a request URL. Model names and availability can change, so check the prices page before saving a model.
Connection problems
ECONNRESET, repeated timeouts, and streaming connections that reset are network-level problems rather than a bad key: a key the gateway rejects comes back as a clean 401 carrying a request id, not a dropped socket.
One application may work while another fails because they use different network protocols - a client that streams over HTTP/2 can reset where a plain request succeeds. Retrying with streaming turned off is the quickest way to tell the two apart.
API call examples
Use the OpenAI-compatible endpoint for GPT apps and the Anthropic-compatible endpoint for Claude Code or Claude Desktop. Both formats route over the same channels and bill identically, so the choice is about what your client speaks, not what the model runs on.
curl https://gensite.tech/v1/chat/completions \
-H "Authorization: Bearer sk_live_YOUR_API_KEY_HERE" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-3.1-pro",
"messages": [{"role": "user", "content": "Reply with OK"}]
}'Model names come from Model prices, and only models your plan permits will answer. The full request and response shapes, error codes, idempotency, and billing are in the API reference.