Integration docs

Same thing, different name. AI API relays, API proxies and LLM gateways, as the community calls them, all describe the same layer between your tools and the model vendors: you integrate against one address, and it forwards requests upstream to OpenAI, Anthropic, Google and others while smoothing over the differences between their protocols. XylemNode is that layer — we call it a gateway throughout these docs.

Check three things in order: whether the endpoint matches the tool's requirements, whether the key was entered correctly with no missing characters, and whether the model name matches the console. Most tools need a restart (or a fresh terminal) after config changes.

It depends on the tool's protocol. OpenAI-compatible ones (Codex, Cursor, OpenCode, Cline, Continue and such) usually need /v1; Claude Code (and work-buddy, which runs on it) and Gemini CLI use their own native protocols and skip /v1; Cherry Studio and CC Switch's Universal Provider take only the domain and complete the path themselves. Each tool guide notes the right form.

The model name isn't one your key can use: either it's mistyped, or it isn't in your key's group. Go by the console's available list, mind the casing and the full identifier (version numbers, prefixes and such), and don't use abbreviations.

It's almost always the key: mistyped, copied with a stray space, expired, or — when set via env vars — not applied because the terminal wasn't reopened. Recheck the key, and check in the console that it is still enabled and hasn't expired.

Usually the address is at the wrong depth, and the two mistakes look different: leave out /v1 where it's needed and the request lands on the website, so a whole web page comes back and the client can't parse it (Unexpected token '<' and the like); add an extra segment (such as /v1/v1) and you get 404 Invalid URL. Match the tool's guide to enter the address at the right level, and don't manually tack on suffixes like /chat/completions — most tools add them automatically.

The endpoint stays the same — just change the model name to the one you want, as long as it's in your key's group; if it isn't, use a key from the group that has it. A few tools also need the new model added to their own model list first: an entry under models in the config file for OpenCode and OpenClaw, another model added in Dify and Coze Studio; and keep the openai/ prefix in aider.

Go by the available-models list in the console, which shows the groups each model is in; a key's group decides which models it can call. Tools that support /v1/models can also fetch the models your key can use automatically. Entering the full identifier from the list is the safest.

Yes. The key isn't tied to any tool — you can share one key across multiple clients and machines, and credits are deducted from the same account.

Top-ups land 1:1 — top up 100 and you get 100 credits — and a plan's credits arrive in full when you activate it and can be used during its one-month term. Both sit in one account and credits are deducted by actual usage. If a call fails, first check whether the balance is sufficient and whether the model is in your key's group.

First confirm your machine can reach api.xylemnode.com normally; corporate networks, firewalls or proxies sometimes block or slow requests. If it's still slow after ruling out the network, try another model or retry later.