From a first API test to a bill you can inspect
Getting an API workflow running should not require a long setup process, a fixed subscription, or a billing model that hides what happened inside each request. SwiftRelay takes a direct approach: top up a balance, connect with three lines, and run. There is no monthly fee and nothing to subscribe to. Each call deducts only what it used.
That makes SwiftRelay a practical API proxy for a first experiment. You can begin with a small top-up, send a short request, and inspect the usage list, which shows tokens and cost per call. Before moving a workflow into regular production use, you can check the recorded usage against the request you expected to send.
The available top-up amounts are $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. A new sign-up also includes $0.22 in gift credit. That gives you room to test the connection without committing to a large balance, while keeping the cost of the trial visible.
SwiftRelay supports up to 20 live API keys per account, so an experiment and a later production workflow can remain part of the same simple setup. Support is available through Messages in the dashboard after you log in. There is no support email to hunt for, and there is no separate support plan to purchase.
What happens when a call is interrupted halfway
An interrupted call is exactly where a pay-per-token API proxy should make its billing rules clear. SwiftRelay does not charge a flat amount for every request. It deducts only what the call used, and the usage list provides the tokens and cost per call so you can review what was recorded.
Suppose a request stops after producing only part of the expected output. You do not need to assume that the full intended answer was delivered and charged as if it had been. The amount taken from your balance follows the usage recorded for that call. You can then compare the tokens and cost shown in the usage list with the stage at which the call ended.
For example, using the listed GPT-6.1 Sol price, an input of 1,000 tokens at $0.1736 per 1M tokens is $0.0001736. An output of 500 tokens at $0.8542 per 1M tokens is $0.0004271. The total for those two token counts is $0.0006007. This is a worked example based on the listed prices, not a fixed charge: if an interruption produces different token counts, the recorded cost will reflect what was used.
The same principle applies when a call is interrupted during a longer agent run. Instead of estimating the finished cost in advance, let the run happen, then check the usage entry. If the expected call is missing or the recorded figures look wrong, open Messages in the logged-in dashboard and ask about that specific entry. That is the fastest route to a clearer answer.
For a production workflow, this inspection step is worth repeating whenever a request behaves differently from the usual pattern. SwiftRelay gives you the usage list as the source for the tokens and cost per call, while the balance shows the effect of those calls over time.
Choose the model and client for the job
SwiftRelay makes it easy to start with the model that fits the task and the client that can run it. The current prices per 1M tokens include GPT-6.1 Sol at $0.1736 for input, $0.8542 for output and $0.0347 for cache read. GPT-5.6 Sol is listed at $0.3056 for input, $1.72 for output and $0.0264 for cache read.
For a quick production-minded test with GPT-6.1 Sol, the Codex setup is three lines:
01 export OPENAI_BASE_URL=https://api.fastapirelay.com/v1 02 export OPENAI_API_KEY=<your swft_ key>
codex
Other OpenAI-compatible clients use the base URL https://api.fastapirelay.com/v1 with a swft_ key as the API key. This route pairs with GPT, DeepSeek and the other remaining models. It gives you a straightforward way to move from a manual test into a script or an established toolchain without changing the billing model.
Claude models answer through the official Claude Code or Claude Desktop client, version 2.1.285 or later. Their three lines are:
01 export ANTHROPIC_BASE_URL=https://api.fastapirelay.com 02 export ANTHROPIC_AUTH_TOKEN=<your swft_ key>
claude
Gemini models use their native protocol through Gemini CLI, with these three lines:
01 export GOOGLE_GEMINI_BASE_URL=https://api.fastapirelay.com 02 export GEMINI_API_KEY=<your swft_ key>
gemini
Keeping the client and model pairing correct avoids a confusing failed request. More importantly, it lets you focus on the experiment itself: send a small request, review the usage, and confirm that the connection behaves as expected before expanding the workflow.
Build a small test before the production run
A sensible path starts with one measurable task. Top up the smallest amount that suits your test, or choose a larger option when you already know how much traffic the workflow will generate. Then run a short request and open the usage list. Record the tokens, the cost per call and the balance change.
Next, test an interruption deliberately. Start a call, stop it partway through, and inspect the entry that SwiftRelay recorded. You will be looking for two things: the tokens actually used and the resulting cost. Because the deduction follows what the call used, this check gives you a practical basis for estimating production usage without assuming that every request will consume the same amount.
Repeat the test with a longer prompt and a longer answer. Compare the entries, and check whether the balance change matches the sum of the recorded call costs. If you need help, use Messages in the dashboard. Keeping this loop short is useful: it turns billing from a mystery into a set of figures you can review before the workload grows.
You can also try more than one model against the same narrow task. DeepSeek V4 Flash is listed at $0.2639 for input, $1.06 for output and $0.006944 for cache read per 1M tokens. GPT-6.1 Sol has the listed input price of $0.1736 and output price of $0.8542. Run each model through the appropriate client, inspect the usage list, and compare the actual results rather than relying on a guess.
When the small tests look right, add the next production step. Create or assign a live key for that workflow, keep it within the limit of 20 live API keys per account, and run a short monitored batch before increasing traffic. SwiftRelay still uses the same pay-per-token approach throughout, so there is no plan to redesign around as usage changes.
Keep the balance flexible and the rewards working
The balance model stays flexible. If you have unused credit from a top-up, you can request a refund within 11 days, minus what was used and the payment fee. That rule is designed for a straightforward top-up rather than a recurring commitment, which is useful when an early experiment does not go as planned.
Inviting others can add value as well. You receive 14% of each top-up made by someone you invited, plus 3% from the people they invite. Rewards can be withdrawn once the minimum is $4.86, or moved into your SwiftRelay balance. Moving them into the balance is the quickest way to turn referral activity into calling credit for another test or a production run.
By the time you reach regular use, the picture should be clear: which client you configured, which model answered, how many tokens each call used, what it cost, and how the balance changed. That is the full path from the first experiment to production use, and SwiftRelay keeps every step connected to a visible per-token charge.
Sign up for SwiftRelay, top up your balance, copy three lines, and put your next API call on the start line.