hermes - 💡(How to fix) Fix [Feature Request]: Configurable gateway response metadata/footer hooks [1 comments, 2 participants]

Summary

Please consider adding a configurable response metadata/footer hook for gateway platforms.

Local deployments often need lightweight observability directly in delivered gateway messages: model name, elapsed time, token usage, cache read/write tokens, context window, provider route, and similar metadata. Today, adding this to platform messages can require patching platform-specific gateway code and the agent/gateway handoff path.

Proposed capability

Expose response metadata to gateway renderers and allow deployments to configure a footer/template without patching source code.

For example:

gateway:
  response_metadata:
    enabled: true
    platforms:
      feishu:
        footer_template: "{model} · {elapsed}s · ctx {context_window} · cache {cache_read_tokens}/{cache_write_tokens}"

Or a more generic hook/template mechanism:

gateway:
  message_hooks:
    after_response:
      include_metadata:
        - model
        - provider
        - elapsed_seconds
        - input_tokens
        - output_tokens
        - cache_read_tokens
        - cache_write_tokens
        - context_window

The exact design can differ. The important part is that the gateway pipeline exposes the relevant metadata and lets platform renderers append or update a footer in a supported way.

Use cases

• Operational visibility for Feishu/Lark, Slack, Telegram, etc.
• Debugging cache and context-window behavior from mobile/chat clients
• Cost-awareness without opening logs or dashboards
• Enterprise/local deployments that want message-level provenance

Expected outcome

A deployment should be able to add or remove gateway response metadata using config/templates. Upgrades should not require patches to platform adapters or gateway/run.py just to pass token/cache/context metadata through.

Why this matters

Gateway messages are often the primary user interface for Hermes. A small, configurable observability layer would reduce source patches while making production deployments easier to operate.