One endpoint. Every model. A bill you can predict.
OpenAI-compatible. Point base_url at api.aireadify.ai/v1 and call models with one key. DeepSeek is live today; Groq, Mistral and xAI follow as corporate accounts are approved. Routing, retries, metering, budgets and logs live in this layer.
from openai import OpenAI
client = OpenAI(
base_url="https://api.aireadify.ai/v1", # the only line that changes
api_key=AIREADIFY_API_KEY,
)
r = client.chat.completions.create(
model="deepseek-v4-flash",
messages=[{"role": "user", "content": "Hello"}],
)
print(r.choices[0].message.content)import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.aireadify.ai/v1", // the only line that changes
apiKey: process.env.AIREADIFY_API_KEY,
});
const r = await client.chat.completions.create({
model: "deepseek-v4-flash",
messages: [{ role: "user", content: "Hello" }],
});
console.log(r.choices[0].message.content);curl https://api.aireadify.ai/v1/chat/completions \ -H "Authorization: Bearer $AIREADIFY_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "deepseek-v4-flash", "messages": [{"role": "user", "content": "Hello"}] }'
You are buying this layer, not cheap tokens.
Unified API
Works with the OpenAI SDKs, streaming and non-streaming. Model ids are ours; switching providers is a one-string change.
Routing (failover in progress)
Unified model names, and a response header tells you which provider served the call. Automatic failover to an equivalent model is being built and is not promised until it ships. No single provider carries more than half our traffic.
Per-key metering
Every call records tokens, upstream cost, billed amount and latency. The console breaks it down by key, model and day.
Hard budget ceilings
Set a monthly budget per key. When credits run out, requests stop with a 402 instead of a frightening invoice.
Logs and request ids
Every response carries an x-request-id you can look up for 7 to 30 days. Prompt bodies are not stored unless you turn debugging on.
Multiple keys
Many keys per account, each with its own name, budget and rate limit. Revocation takes effect immediately.
Peak 20% below list, off-peak at list. Written here, not hidden in a calculator.
Data region and zero-retention status are stated honestly. DeepSeek processes data in China with no zero-retention option; pick another provider for sensitive workloads.
| Model | Provider | Context | Data region | Zero retention | Peak · in / cached / out | Off-peak · in / cached / out |
|---|---|---|---|---|---|---|
| deepseek-v4-flash | DeepSeek | 1M | CN | No | $0.33 / $0.011 / $0.99 | $0.21 / $0.007 / $0.62 |
| deepseek-v4-pro | DeepSeek | 1M | CN | No | $0.99 / $0.033 / $2.96 | $0.62 / $0.021 / $1.85 |
| deepseek-v4-flash-vision-exp | DeepSeek | 1M | CN | No | $0.33 / $0.011 / $0.99 | $0.21 / $0.007 / $0.62 |
| gpt-6-astra | OpenAI · via vetted reseller | 400K | US | Unknown | $1.95 / $0.292 / $9.75 | $1.95 / $0.292 / $9.75 |
| gpt-5.6 | OpenAI · via vetted reseller | 400K | US | Unknown | $1.03 / $0.155 / $8.28 | $1.03 / $0.155 / $8.28 |
| gpt-5.6-sol | OpenAI · via vetted reseller | 400K | US | Unknown | $0.78 / $0.117 / $3.90 | $0.78 / $0.117 / $3.90 |
| gpt-5.6-terra | OpenAI · via vetted reseller | 400K | US | Unknown | $0.39 / $0.059 / $2.34 | $0.39 / $0.059 / $2.34 |
| gpt-5.6-luna | OpenAI · via vetted reseller | 400K | US | Unknown | $0.10 / $0.016 / $0.62 | $0.10 / $0.016 / $0.62 |
USD per million tokens. Peak hours are Beijing time Monday to Friday 09:00–12:00 and 14:00–18:00, priced 20% below DeepSeek's own peak list price; all other hours are off-peak at DeepSeek's off-peak price. The OpenAI models (gpt-6-astra and the gpt-5.6 family) are served through a vetted reseller at a flat price with no peak/off-peak schedule; cached prompt tokens are billed at 15% of the input price. Billed per token from each response's usage block, no per-request minimum, balance shown in USD. DeepSeek requests are processed in China with no zero-retention option.
Every cent, explained before you spend it.
Charged per token. Peak hours (Beijing time Monday–Friday 09:00–12:00 and 14:00–18:00) are priced 20% below DeepSeek's peak list price; all other hours equal DeepSeek's off-peak price. Settled exactly from the usage block of each response, no per-request minimum.
Upstream errors are not billed. Streaming requests are billed from the usage in the final chunk.
Prepaid credits, $10 minimum, never expire. The Pro plan includes $20 each month.
Each request first reserves credits up to its ceiling (input length plus max_tokens). If the balance or the key's monthly budget can't cover it, you get a 402 before anything is sent upstream; after the response, the actual usage is settled and the difference returned.
Credits are not refundable for cash; see the refund policy. Unused subscription time is refunded pro rata.
Straight answers
How is this different from calling providers directly?
One key, one bill, one set of rate limits and budgets; switching models needs no code change and retries are automatic. You pay no more than DeepSeek direct and skip the integration, reconciliation and on-call.
How is this different from OpenRouter?
Feature for feature we can't build what it lacks, and we don't pretend to. The difference is who we serve: teams building agent products, Chinese-language docs and support, finer budget control, and one account shared with the scanner and the upgrade service.
Is my data used for training?
We don't train models and we don't keep prompt bodies. Upstream policies differ; the model table marks data region and zero-retention status, and DeepSeek offers no zero-retention option.
Why no OpenAI or Anthropic models?
We only connect providers where Kestro Labs, LLC can open a business account under their terms and where those terms allow integration into our own product. Providers that give written confirmation will be added.
Can I resell this to my users?
You can integrate the Gateway into your own product for your end users; that is what it is for. You can't resell bare keys or run a functionally identical relay; see the acceptable use policy.