Relay365 API relay setup: Claude Code 401, 402 and 429

Start at the duty desk: when 401 says the key and the lane do not line up

Welcome to the Relay365 duty desk. Relay365 is an independent third-party relay, run like a round-the-clock duty desk, and the first hour after signing up is usually spent wiring a client and reading what comes back. Three codes deserve a place on your checklist early: 401, 402 and 429. None of them need to be mysterious. At Relay365 every lane sits on one table, so the same dashboard shows your keys, the lanes, the prices and the history behind each request.

When a request comes back as 401, the desk would check the wiring before changing anything else. Open the dashboard and look at the API keys you created: each key can carry tags and can be bound to a lane, and quick-wiring snippets are there to copy the base address and the model wire name into your client. Then confirm that the key your client is actually sending is the key you think it is sending, that it is bound to the lane you intend to call, and that the client is allowed to speak to that lane at all.

That last point matters more than it looks. Claude models work only from the official Claude Code and Claude Desktop clients, with Claude Code 2.1.285 or newer. Cursor, Cherry Studio, Cline, OpenCode, any SDK or cURL pair only with the GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax and Hunyuan lanes. Gemini runs on its native protocol only, through Gemini CLI, using :generateContent for single replies and :streamGenerateContent?alt=sse for streams; it cannot go through /v1/chat/completions. If a 401 shows up while you are wiring a new client, the lane-to-client pairing is the first thing to verify.

Wire names are worth copying exactly too. Claude Fable 5 is claude-fable-5, and GPT-6 Astra is gpt-6-astra. A near-miss in a wire name is a quick fix in the first hour and a long chase later.

402 and the wallet: check the balance, then check the token mix

Relay365 bills from a prepaid wallet, with no monthly fee, and every call is settled on the token counts the upstream reports. A call that fails upstream with a 502 or 503 is not billed, while a disconnected stream is billed for what was actually used. So when a 402 shows up, treat it as a balance-and-billing question: look at the wallet account in the ledger, confirm there is room for the call, and remember that the charge is assembled from four separately priced token types: input, output, cache write and cache read.

The four-way split is why two calls with a similar visible length can cost different amounts. Claude Fable 5 is listed per 1 million tokens in approximate USD at input $3.79, output $18.94, cache write $4.66 and cache read $0.2917, with an output rate of about 38% of the official list price. GPT-6 Astra is listed at input $0.7444 and output $3.72, with an output rate of about 7% of the official list price. The account itself is kept in CNY.

For example, if you send 1,000,000 input tokens, 1,000,000 output tokens and 1,000,000 cache write tokens a day on Claude Fable 5, that is $3.79 + $18.94 + $4.66 = $27.39 a day, before any cache read. On GPT-6 Astra, 1,000,000 input tokens plus 1,000,000 output tokens a day works out to $0.7444 + $3.72 = $4.4644 a day. The pricing page shows the full table for every lane.

Optional duty passes are worth a look in the same hour. A pass comes with a dedicated key that never touches the wallet, and it is available for Claude and GPT only; other models always use the wallet. The Claude 14-day pass offers $700 of quota with a $380 per week limit for $50.97, drawn at the duty-pass rate x2.5, which is at most about $280 of usage at list prices and less on models whose pass rate sits above list. The GPT 30-day pass offers $3000 of quota with a $900 per week limit for $81.53, at most about $1200 of usage at list prices.

A new account also carries $0.25 of signup credit. If you top up, unused balance can be refunded within 14 days of the top-up, minus what was consumed and the payment channel's fees; bounty and promotional credit are not refundable. The referral bounty pays 18% at level 1 and 6% at level 2 of what referred members pay, and it can be withdrawn from $11.11 or moved into the wallet.

429 and the traffic desk: know the two ceilings before you scale

Relay365 returns 429 in two situations, and both are worth checking while your client is still small. The first is more than 120 calls in 60 seconds on one key. The second is more than 8 requests in flight for one account. Almost every 429 hunt at the desk comes back to those two numbers.

Start with the usage log: the latest 200 requests are listed with the time, model, input, output, cache recall, charge and result of each call, so you can see whether the traffic that hit the limit came from one loop you did not notice. Then spread the load across keys with clear tags, and add a small queue or backoff so your client waits instead of firing.

One caveat: the usage log does not show which key made the call. If you run several clients at once, give each one its own tagged key and its own lane binding. That keeps the desk's picture tidy and turns both the 401 and the 429 checks into quick work.

Every request on one table: the usage log, the ledger and the desk

Relay365 keeps the operational picture in one place. The usage log of the latest 200 requests shows the time, model, input, output, cache recall, charge and result of each call, which means the four token types are visible individually rather than collapsed into a single line. The ledger lists the wallet, bounty and pass accounts side by side, so you can tell which balance a pass is drawing from and how much the wallet still holds.

If something still looks off, the desk is reachable in-site. Log in first, then send a message to the desk with the time, model and wire name from the log, and the desk can pick the request up from there. The interface is available in six interface languages, and the quick-wiring snippets sit next to the keys, so a fresh client can be pointed at the right lane without guesswork.

Because Relay365 is an independent third-party relay, it is worth keeping your own notes on which lane and wire name each client uses. The desk can help you read the log, but a tidy key naming scheme is what turns a 401 into a two-minute fix.

The first-hour checklist at the Relay365 desk

• Confirm your key is bound to the lane you intend to call, and give it a tag you will recognise later.
• Match the client to the lane: Claude models only in the official Claude Code and Claude Desktop clients, with Claude Code 2.1.285 or newer, Gemini only through Gemini CLI on its native protocol, and the other lanes in Cursor, Cherry Studio, Cline, OpenCode, any SDK or cURL.
• Copy the wire name exactly, such as claude-fable-5 or gpt-6-astra, and start from the quick-wiring snippet.
• Check the prepaid wallet balance and the ledger before you send a long prompt, and remember that input, output, cache write and cache read are priced separately.
• Keep traffic under 120 calls in 60 seconds on one key and under 8 requests in flight for one account, and add backoff before you scale.
• Read the usage log after your first calls and confirm the time, model, input, output, cache recall, charge and result match what you expected.
• Note that a failed upstream call returning 502 or 503 is not billed, while a disconnected stream is billed for what was actually used.
• If a duty pass fits your workload, compare it with wallet pricing: the Claude 14-day pass is $50.97 for $700 of quota, and the GPT 30-day pass is $81.53 for $3000 of quota.

That is the whole first hour: keys and lanes checked, wallet and token mix checked, traffic ceilings checked, and the usage log read once end to end. When you are ready, create your Relay365 account and start with one lane, one client and one small prompt.