[Bug] chat.z.ai Discord Connector Agent Fails with Chinese Capacity Error "当前模型使用人数较多" and Lacks Model Configuration

Problem Statement

When integrating an Agent on chat.z.ai with a Discord Bot via the platform Connectors (e.g. @Diana bot), mentioning the bot in a Discord channel repeatedly fails with an untranslated Chinese capacity error:

当前模型使用人数较多，请稍后再试或切换到其他模型。
(Translation: "The current model is experiencing high usage. Please try again later or switch to another model.")

Steps to Reproduce

1. In chat.z.ai, configure a custom Agent and connect it to a Discord Bot using a Discord Bot Token under Connectors.
2. Invite the bot to a Discord server / channel.
3. In any channel, mention the bot (e.g. @Diana in english).
4. Observed Result: The bot replies with:
   当前模型使用人数较多，请稍后再试或切换到其他模型。
5. Re-pairing the bot or regenerating the Discord bot token does not resolve the issue.

Root Causes & Architectural Gaps

1. Raw Gateway Error Leakage: The backend connector service does not catch or handle inference gateway 429/503 capacity errors, leaking internal untranslated Chinese server error messages directly to end-users in Discord.
2. Missing Model Configuration for Connector Agents: The error instructs the user to "switch to another model", but the Discord agent connector interface provides no option to configure or switch the underlying model (e.g., fallback to GLM-5.3-Flash or GLM-4). The bot is locked to a default backend model that repeatedly triggers capacity limits.
3. No Language Localization: Error messages ignore the user's language preferences and conversation language.

Expected Behavior

• Connector bots should have configurable primary and fallback models (e.g., allow binding to GLM-5.3-Flash).
• Capacity errors should automatically retry with fallback or return localized error messages.

Reference

• Discord Report Thread: Discord Connector Issue
• Reported by: Community member @Jessie Ablene

Regards,
Roman M.

• Contributor / Ambassador at Z.ai
• The Vibe Coder's Blog on Telegram: https://t.me/VibeCodePrompterSystem

Start with the Connectors flow for configuring an Agent with a Discord Bot Token and trace a mention from the Discord channel to the model request. Compare the reported capacity error with the expected primary and fallback model behavior and localized response requirements; done means the connector no longer leaks the raw error and supports the requested model handling.