Production integration docs · examples use router-api.mengbi-ai.net and a personal API key placeholder.
Why does one task create multiple billing entries?
Agent tasks may involve several model and tool requests.
Multiple requests can be expected
An agent may read context, call tools, and generate a final answer. Each stage can create its own usage and token charge.
Models may switch
If the client switches models by task stage, usage is recorded against the model used for each request.
Where should I keep my API key?
Personal keys should only appear in server-side environment variables or protected local client configuration.
If a key is exposed
- Delete the old key in your personal console.
- Create a replacement key and update your environment variables.
- Check logs, screenshots, and repository history for the full key.
Why do the model list and prices change?
Availability and prices can be updated with the current catalog cycle.
Use the current catalog and API
Read /v1/models before making a request and use the current price snapshot shown in the model catalog. Do not cache models or prices permanently.
502 and upstream errors
A 502 means Mengbi received a non-success response from the upstream provider.
Use the recorded status
- Send support the requestId, request time, and model ID.
- upstream_status_code=402 means the upstream account is out of balance; 401/403 means an upstream key or permission issue.
- upstream_status_code=404 means you should check the protocol endpoint, duplicate /v1 path, and model ID.