English
Sign in
Back to Home

About

What We Do

Admesh.ai is a large-model aggregation gateway. We address a specific problem: when you need to use multiple models—such as Claude, GPT, and DeepSeek—at the same time, you don’t have to manage multiple sets of accounts, billing systems, or troubleshooting procedures.

In a nutshell

One Base URL, one API Key, one invoice—access the world’s leading large language models. The protocol remains OpenAI-compatible; migrating existing code requires only updating the Base URL, Key, and model name.

Limits: What We Don't Do

  • Do not train or fine-tune the model.We are not a model provider; our modeling capabilities are provided by upstream partners.
  • Do not perform cross-model automatic replacement.When you request a model from Anthropic, the response will always be an Anthropic model. Disaster recovery only occurs between multiple channels of the same model.
  • Do not save the body of your prompt.The usage log records only metadata, the number of tokens, the amount, error codes, and latency.
  • We don't hide our prices.The unit prices of the models are publicly available and can be viewed without registration.

Engineering Principles

  • The balance is derived from the ledger.Every deduction, top-up, and refund is recorded as a separate transaction, allowing for transaction-by-transaction reconciliation; there are no "black-box" operations that directly alter the balance.
  • Price Snapshots.Each usage reading captures a snapshot of the price at that time; price changes do not affect historical bills.
  • Every request is traceable.Each request has a unique `request_id`, which can be traced back to the actual model and channel used.
  • Upstream voucher isolation.The upstream key is stored in encrypted form and is not included in logs, browser code, or external interfaces.

Contact