
AI MODEL GATEWAY
LINGJING
One gateway for your AI work.
Use LINGJING for web AI chat and a unified model API. Compare models, estimate usage costs in USD, and configure WorkBuddy with your own API key.
Models & pricing
USD per million tokens. See the cost estimator for current rates.
ANTHROPIC
MiMo
DEEPSEEK
QWEN
KIMI
GLM
MiniMax
META
AMAZON
Your first request
Verify your email
Register, complete human verification and verify your email before signing in.
Integration guideBalance & API keys
Check prices, add prepaid balance and create a separate API key for each app.
Integration guideConnect an app
Use the settings below or run an API example with your own key.
Integration guideApps & agents
WorkBuddy
5.5.6: basic chat and test-file creation/readback verified. Advanced plugins and terminal workflows are not certified.
Cline
CLI 3.0.62: basic text verified; file-tool workflow not yet verified.
Cherry Studio
Validation status and configuration differ by client; see the individual guides below. OpenAI compatibility does not mean every parameter is supported.
Aider
Validation status and configuration differ by client; see the individual guides below. OpenAI compatibility does not mean every parameter is supported.
System prompts, tool definitions and history are billed. Our WorkBuddy test used about 50,000 input tokens per round; even a short instruction can cost several cents. Multi-round tasks accumulate charges.
Before you start
How do I start?
Register and verify your email. Check models and prices in the console, add balance and create a separate API key for each app. Use web chat or a compatible client. Start with a short text test: real model calls are billable; documentation and the estimator do not call a model.
Which capabilities are available?
Check the console and integration guide for current models, parameters and capabilities. Web chat defaults to chatgpt-6-astra. Streaming, tools and image support vary by route; model names do not guarantee every feature.
How do payments and pricing work?
Balance is recorded in USD. WeChat Pay and Alipay collect the CNY amount shown on the order; Stripe options depend on checkout. CNY order rates update daily and are locked when the order is created: source rate multiplied by 1.007, rounded to two decimals. Uncached input, cache-hit input and output have separate rates. Applicable reasoning tokens count as output, not twice. History, system prompts and repeated calls increase usage. Check the estimator and reconcile against the request quote and usage record.
What if payment is missing or funds are reserved?
Refresh the order and check the payment receipt before paying again. Calls may reserve funds before settlement; uncertain results can retain a reservation for review. A timeout does not prove the provider incurred no cost. Send the order or request ID, time and redacted error. Billing disputes and refunds require individual review, not an automatic refund promise. Later price changes do not reprice old request quotes.
How do I configure a key? Is it a ChatGPT login?
Use https://api.lingjing.win/v1 as Base URL, /chat/completions for text, and the exact catalog model ID. A key authorizes API calls; it is not a ChatGPT, Claude Code or Codex account or subscription. Haiku can omit Idempotency-Key, but retries without it are new billable calls; other models require distinct request IDs. Follow the developer guide for client-specific tools, attachments and parameters.
Does data stay on my device?
No. Request content goes to the relevant model service. The gateway filters unnecessary account metadata but does not automatically remove personal information written inside prompts or guarantee that providers cannot recognize usage patterns. Do not submit secrets or unauthorized personal data. See the privacy policy for processing, storage and cross-border handling.
What can AI support do, and who pays?
AI support is restricted to LINGJING usage and configuration. It cannot perform unrelated tasks, refunds or permission changes. LINGJING pays; customer balance is not deducted. Each account has a daily procurement-cost cap of USD20, resetting at midnight Beijing time. Free documentation remains available afterwards. Out-of-scope requests are screened before model calls; contact human support for disputes.
How do I troubleshoot?
401: check the full key. 403: model permissions. 400: parameters, history and context. 409: inspect the original request. 429: reduce concurrency and honor Retry-After. For 5xx or timeouts, inspect usage before retrying. Copy the full newly created key, not its abbreviated list prefix. Revoke suspected leaked keys. Never send passwords, verification codes or full keys to support.