Step 1: Start with a balance instead of a plan
Here is the fast lane. SwiftRelay is a pay-per-token API relay, and the whole idea is simple: you top up a balance, and every call deducts what it actually used. There is no monthly fee and nothing to subscribe to, so a week when your team is quiet does not quietly cost you the same as a week when everyone is shipping.
That flexibility is the point. If you are used to paying for a fixed seat or a fixed allowance on someone else's account, SwiftRelay asks you to flip the switch: you are buying tokens as you go, and your balance follows your real use. Busy weeks draw more, slow weeks draw less, and there is no ceremony involved in either case. No contract to negotiate, no tier to justify, no renewal date to remember.
Signing up is where the stopwatch starts. Create your account, and SwiftRelay gives you a sign-up gift credit of $0.22 to put toward your first calls. That is enough to run something small, see the numbers land in your dashboard, and understand the rhythm before you commit a larger top-up. Think of it as a first lap around the track rather than a long contract.
One more thing to know up front: SwiftRelay is not a bundle of features you have to decode. You pick your models, you pick the amount you top up, and you watch the usage list show you tokens and cost per call. The dashboard tells you what happened; you decide what to do next.
Step 2: Choose a top-up that matches your real workload
Now the only decision: how much do you want in the tank? SwiftRelay offers top-up amounts of $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. That spread is deliberate. A small amount is fine for a slow week or a weekend experiment. A larger amount gives you room to sprint through a busy release cycle without topping up again mid-flight.
Because each call deducts only what it used, a bigger top-up is not a bigger commitment. It is simply more runway. If the week turns out to be quieter than you expected, the unused credit is still sitting there, moving with you into the next one. If the week turns out to be louder, you have already bought the runway you need.
Worked example: suppose you top up $27.78 on Monday. By Thursday, your usage list shows a handful of calls, each with its token count and its cost. Whatever those calls consumed has been deducted; the rest of the balance is untouched and ready for the burst on Friday. There is no penalty for topping up and using only part of it, because you are not buying a period of access. You are buying tokens, and the balance tracks them.
Support, by the way, lives in the same place you are already working. If a question comes up about a call, a key or a charge, you can reach the team through Messages in the dashboard once you are logged in. No hunting for a support address, no separate help centre to learn.
Step 3: Connect in minutes and make the first call
This is the part where most people hesitate, and SwiftRelay removes the hesitation: connected in minutes, copy three lines and run. For OpenAI-compatible clients, the base URL is https://api.fastapirelay.com/v1, and you use your swft_ key as the API key. Codex, Cursor, Cline, Roo Code, Cherry Studio, OpenCode, SDKs and cURL all use that OpenAI-compatible endpoint, and they pair with GPT, DeepSeek and the other remaining models.
For Codex, the three lines look like this:
01 export OPENAI_BASE_URL=https://api.fastapirelay.com/v1 02 export OPENAI_API_KEY=<your swft_ key> 03 export codex
For Gemini models, the route is different and worth getting right: Gemini uses its own native protocol through Gemini CLI, and Gemini CLI is where those models run. The three lines are:
01 export GOOGLE_GEMINI_BASE_URL=https://api.fastapirelay.com 02 export GEMINI_API_KEY=<your swft_ key> 03 export gemini
That is the whole setup. Paste, run, and your first call goes out through SwiftRelay, with the cost deducted from your balance as it happens. No integration project, no onboarding call, no long checklist.
You can run up to 20 live API keys per account, which means a busy week can involve several tools, several services or several environments without forcing you into separate relationships with different providers. Each key stays yours, and each call is accounted for in the usage list with tokens and cost per call.
Step 4: Ride a busy week with the numbers in plain sight
Here is where a pay-per-token relay earns its keep. On a busy week, your usage climbs, and because the pricing is per 1M tokens, you can see exactly what you are paying for. The current SwiftRelay pricing page lists, for example, GPT-6.1 Sol at $0.1736 input and $0.8542 output per 1M tokens, with cache read at $0.0347. DeepSeek V4 Flash sits at $0.2639 input, $1.06 output and $0.006944 cache read. Gemini 2.5 Flash is $0.1875 input, $1.50 output and $0.0194 cache read.
Worked example, and this is an example rather than a promise: a single call using 1,000,000 input tokens and 1,000,000 output tokens would cost $1.0278 with GPT-6.1 Sol, using the listed per-1M figures. The same shape of call would be $1.3239 with DeepSeek V4 Flash and $1.6875 with Gemini 2.5 Flash. Multiply that pattern across a busy afternoon and the balance moves in a way you can actually predict, because every line of the usage list shows you the tokens behind the deduction.
Worked example for a heavier model: with Claude Sonnet 5.5, a call using 1,000,000 input tokens, 1,000,000 output tokens and 1,000,000 cache read tokens would come to $0.9792 plus $4.89 plus $0.0861, which is $5.9553. Claude models answer through the official Claude Code or Claude Desktop client, version 2.1.285 or later.
The point of a busy week is not that it should hurt. The point is that it should be visible. With SwiftRelay, a heavy day shows up as a heavier deduction, and a light day shows up as a lighter one. You are never paying a flat rate for a week you did not fill.
Step 5: Coast through a slow week and keep the extras working
Slow weeks are where a balance model shines. If traffic drops off, your deductions drop with it. Nothing expires on a schedule you have to defend, and nothing is wasted on unused capacity you never touched.
There is also a refund path for anyone who tops up and changes their mind: unused credit can be refunded within 11 days of a top-up, minus what was used and the payment fee. That makes a larger top-up a low-drama decision on a busy week, because you still have a defined window if the plan changes.
And the balance can grow on its own. SwiftRelay referral rewards give you 14% of each top-up by someone you invited, plus 3% from the people they invite. Those rewards can be withdrawn, with a minimum of $4.86, or moved into the balance to fund more calls. Shared, they become part of the same smooth loop: invite a teammate, they top up, and the balance you use every day gets a little more fuel.
So the pattern is complete. Top up the amount that fits the week ahead, connect in minutes with three copied lines, let each call deduct what it used, and watch the usage list tell the story. Busy weeks burn brighter, slow weeks burn cooler, and SwiftRelay keeps matching the bill to what you actually did.
Lap it out: sign up for SwiftRelay today, claim your $0.22 sign-up gift credit, and put your first call on the track.