Swift API

Chat Completions for the Swift models, compatible with the OpenAI SDKs. Point an SDK at this base URL and use a key from the dashboard.

1. Create a key

2. Add credit

Add at least $5 on the Billing page. Usage is taken from your balance per token, and paid credit doesn't expire.

3. Make a request

Models and pricing

Prices are in US dollars per million tokens. Reasoning tokens count as output. GET /v1/models returns the same list.

Chat completions

POST /v1/chat/completions takes the OpenAI request format. These fields are used:

model and messages, with the roles system, developer, user, assistant and tool
max_tokens or max_completion_tokens: 8,192 by default, up to the model's max output
temperature, top_p, top_k, min_p, presence_penalty, frequency_penalty, repetition_penalty, seed, stop
tools, tool_choice, parallel_tool_calls, response_format
logprobs and top_logprobs (up to 5)
stream and stream_options
chat_template_kwargs.enable_thinking, see Reasoning

Other fields are ignored, and n must be 1. If your balance can't cover max_tokens, it's lowered to what the balance covers and the response carries an x-max-tokens-applied header.

GET /v1/credits returns your balance, tier and limits for the key you send.

Streaming

With "stream": true the response is a stream of server-sent events that ends with data: [DONE]. Set stream_options.include_usage to get a last chunk with the token counts. If you stop reading early, you pay for the tokens generated up to that point.

Tool calls

Tools use the OpenAI function format. The model replies with tool_calls; run the tool and send its result back as a tool message with the matching tool_call_id.

Images

Send images as image_url content parts with a base64 data URL in PNG, JPEG, WebP or GIF. Links to images on the web are refused. Images are billed as input tokens.

Reasoning

Swift thinks before it answers. The thinking arrives in reasoning_content, next to content, and is billed as output tokens. To answer without thinking, set enable_thinking to false:

Prompt caching

When a request starts with the same tokens as a recent one, that shared start is billed at the cached input price. It happens on its own; usage.prompt_tokens_details.cached_tokens shows how many tokens were cached. Keep the parts that don't change, such as the system prompt, tools and documents, at the start.

Rate limits

Limits apply to your account across all its keys. Tiers rise with the credit you've paid in and the age of your first payment. Tokens per minute count uncached input plus output; requests at once are counted per model.

Every response carries x-ratelimit-limit-requests, x-ratelimit-remaining-requests, x-ratelimit-reset-requests and the same three for tokens. A 429 response carries Retry-After.

Errors

Errors use the OpenAI shape, {"error": {"message", "type", "code"}}, and every response has an x-request-id.

Status	Code	What to do
400	`invalid_request`	Fix the request; the message says what's wrong.
401	`invalid_api_key`	Check the key. Revoked keys stop working at once.
402	`insufficient_quota`	Add credit. Retrying won't help.
403	`account_suspended`	Write to hello@ukisai.com.
404	`model_not_found`	Use a model from `GET /v1/models`.
429	`rate_limit_exceeded`	Retry after `Retry-After` seconds.
429	`concurrency_limit_exceeded`	Wait for a request to finish, or move up a tier.
429	`model_at_capacity`	The model is busy. Retry after `Retry-After` seconds.
429	`spend_limit_exceeded`	Your tier's monthly limit is reached.
503	`model_unavailable`	The model is down. Retry later.

Data and privacy

We don't store or log prompts or outputs, and we don't train on them. For billing we keep the token counts, cost and timing of each request.

The API runs on a best-effort basis with no uptime guarantee. Models marked Preview can change or go away. See the Terms.