Claude Code setup with the Relay365 API relay for indie apps
The morning version: four vendor tabs and a feature that will not ship
It starts the way most indie app weeks start. You promised a small AI feature — a summarizer, a rewrite button, a chat sidebar — and by mid-morning you are moving between four vendor tabs, copying keys into four config files, and re-reading four pricing pages to see which one takes dollars and which one keeps the account in another currency. Every vendor has its own request shape, its own dashboard, its own idea of what a token costs. The feature itself is the easy part. The plumbing is what eats the day.
By lunchtime the honest question appears: could I ship this from one place instead? One table of models and rates. One endpoint shape. One log where I can see what each call actually consumed. That is not a wish for a magic service; it is a request for less coordination. Relay365 is an independent third-party relay, run like a round-the-clock duty desk, and its whole idea is simple: every lane sits on one table.
So here is the same working day, told twice. The first telling is the scattered one, with tabs open and keys half-configured. The second telling is the version where the desk has your back, the pricing is visible before you write the integration, and the feature ships while you still have the afternoon left.
The afternoon version: one table, one log, four token types
Relay365 exposes an OpenAI-compatible API base, so the mainstream lanes drop into the request format your code already expects. You create an API key, tag it, and bind it to a lane. That binding is the point: instead of one shared key that quietly routes to a surprise model, the dashboard tells you which lane the key belongs to, and the quick-wiring snippets show you how to point your app at the relay.
Then there is the usage log, which keeps the latest 200 requests and shows the time, the model, the input, the output, the cache recall, the charge and the result of each call. One caveat worth knowing before you rely on it: the log does not show which key made the call. It shows you what happened, not which teammate pressed the button.
Around that log sits the ledger of wallet, bounty and pass accounts, plus in-site messages to the desk, and six interface languages. And the pricing itself is itemised rather than blended: input, output, cache write and cache read are priced separately. For an indie builder, that separation is not bookkeeping trivia. It is the difference between guessing at a bill and reading one. The pricing page shows the full table; the desk keeps the same logic behind the scenes.
What the day actually costs on the Relay365 lanes
Take one lane you might genuinely build on. GPT-6 Astra, wire name gpt-6-astra, is priced at $0.7444 per 1 million input tokens and $3.72 per 1 million output tokens, with cache read at $0.075 and no cache write rate listed. Its output rate is about 7% of the official list price of $50 per 1 million output tokens. For example, if you send 100,000 input tokens and 10,000 output tokens a day, that is $0.07444 plus $0.0372, or $0.11164 a day and about $3.35 over 30 days. Small change for a shipped feature, and the number comes from the published rates rather than a guess.
Compare the same volume on DeepSeek V4 Flash, wire name deepseek-v4-flash: $0.2889 per 1 million input tokens and $1.15 per 1 million output tokens, with cache read at $0.006944. No official price is quoted for this family, so the desk makes no comparison — only the rates. For example, if you send 1,000,000 input tokens and 1,000,000 output tokens once, that is $0.2889 plus $1.15, or $1.44 for that pair of million-token runs.
On the Claude side of the table, Claude Fable 5 (wire name claude-fable-5) is $3.79 per 1 million input tokens and $18.94 per 1 million output tokens, with cache write at $4.66 and cache read at $0.2917. Its output rate is about 38% of the official list price of $50. The same 1,000,000 input plus 1,000,000 output would be $3.79 plus $18.94, or $22.73. Seeing both figures on one table lets you choose a lane for the job instead of choosing a vendor for the quarter.
Duty passes for the Claude and GPT lanes
Some indie apps go from demo to real traffic quickly, and a prepaid wallet alone can feel like the wrong shape for that week. Relay365 offers optional duty passes for the Claude and GPT lanes, and each pass comes with a dedicated key that never touches the wallet, so pass spending and wallet spending stay separated on the ledger.
The Claude pass priced at $720 of quota for 30 days costs $94.03, with at most $230 of quota per 7 days. Quota is drawn at the duty-pass rate x2.5, which is at most about $288 of usage at list prices, and less on models whose pass rate sits above list. The GPT pass priced at $2350 of quota for 30 days costs $81.53, with at most $700 of quota per 7 days, drawn at the duty-pass rate x2.5 for at most about $940 of usage at list prices, and less on models whose pass rate sits above list.
Two details the desk would rather you read now than discover at 2 a.m. A call already let through on a pass is drawn in full from the pass even if it runs past the quota or the 7-day ceiling, and that overrun counts against the earliest-expiring card. When a pass is nearly used up, each running call holds $3 x the draw rate, and an extra parallel call gets HTTP 429 until the running ones finish. That is the desk telling you to wait, rather than silently overspending.
Client rules, limits and the billing behaviour behind the endpoint
Rules matter more than slogans here. 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, SDKs and 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: :generateContent for single replies and :streamGenerateContent?alt=sse for streams, and it cannot go through /v1/chat/completions. The desk will happily point you to the right lane, but it will not pretend the protocols are interchangeable.
Limits are plain as well. You will see HTTP 429 if one key sends more than 120 calls in 60 seconds, or if one account has more than 8 requests in flight. Billing is 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. A disconnected stream is billed for what was actually used, which is the fair reading of a half-finished answer.
For Gemini lanes, the native protocol also means its own rate shape on the table — Gemini 2.5 Flash, for example, lists $0.2389 per 1 million input tokens, $1.91 per 1 million output tokens, cache write $0.2389 and cache read $0.0236, with no official price quoted for that family. The desk keeps it listed anyway, so you can plan the budget before you choose the client.
Opening the account and keeping the wallet honest
A new Relay365 account starts with $0.25 in signup credit, which is enough to send a first test request and see the usage log light up. Top-ups go into the prepaid wallet, and refunds are available within 14 days of a top-up: unused balance can be refunded minus what was consumed and the payment channel's fees. Bounty and promotional credit are not refundable, which is worth keeping in mind when you top up for a launch week.
There is also a referral bounty of 18% at level 1 and 6% at level 2 of what referred members pay. It can be withdrawn from $11.11 or moved into the wallet, so if your launch has a community around it, the bounty can simply fund the next round of usage.
The difference between the two versions of this day is not the feature. It is the number of places you have to look. With Relay365, the lanes share one table, the four token types carry their own rates, and the usage log keeps the latest 200 requests with time, model, input, output, cache recall, charge and result. Create a Relay365 account, put your key on a lane, and ship the AI feature before the day gets away from you.