hermes - 💡(How to fix) Fix [Feature]: Persistent usage/cost footer on gateway (Discord, Telegram, Slack) messages [1 participants]

Problem or Use Case

When using Hermes via Discord or Telegram (messaging gateway), there is no way to see token usage or cost information without manually running /usage each time. The /usage command returns a one-shot stats block as a separate message, but there is no persistent display that shows cost/usage information alongside every response.

In the CLI, show_cost in config.yaml was intended to show cost in the status bar, but:

1. display.show_cost is a dead config key — defined as a default in hermes_cli/config.py (line 525) but never read anywhere in the codebase. It has zero effect.
2. /usage is one-shot only — it returns a single message with a detailed stats block, but does not toggle persistent display.
3. No gateway mechanism exists to append usage info to outbound messages.

This matters for users running multiple agents (different profiles) across Discord/Telegram who need to monitor spending in real-time without remembering to type /usage after every interaction.

Proposed Solution

Add a display.usage_footer config option (default: false) that, when enabled, appends a compact usage line to every gateway message response.

Config example:

display:
  usage_footer: true    # Append usage/cost footer to every gateway message

Behavior when enabled: After each agent response on messaging platforms (Discord, Telegram, Slack, WhatsApp), append a footer line like:

_📊 12.3k tokens · ~$0.0234 · ctx 45%_

Or, for a more detailed variant:

_📊 Input: 10,200 · Output: 2,100 · Total: 12,300 · Cost: ~$0.0234 · Context: 45%_

Implementation approach: The footer would be appended in gateway/platforms/base.py _process_message_background(), after text_content is prepared (around line 1745) but before self._send_with_retry(). The token counts and cost data are already available in the agent_result dict returned by _handle_message_with_agent() (which includes input_tokens, output_tokens, last_prompt_tokens, model).

A new _format_usage_footer() method on GatewayRunner (or a shared utility) could format the compact line, using the existing agent.usage_pricing.estimate_usage_cost() for cost estimation and the agent context compressor for context percentage.

The config resolution should follow the same pattern as tool_progress — check display.usage_footer first, then display.platforms.<platform>.usage_footer for per-platform overrides.

Alternatives Considered

1. Hook system — The gateway/hooks.py event system fires agent:end with context data, but hooks are fire-and-forget and cannot modify response content. A hook cannot inject a footer into the response text.
2. Agent instructions — Adding "always include your token usage" to SOUL.md. This is unreliable — models forget, format inconsistently, and counts may be inaccurate (the model does not know its actual token counts).
3. Separate follow-up message — Auto-sending /usage after every response. This would double the message volume and clutter the chat.
4. Discord embed footer — Using Discord embed footer fields instead of text. This would be platform-specific and does not work for Telegram/Slack.

Feature Type

Gateway / messaging improvement

Scope

Medium (few files, < 300 lines)

Additional Context

I investigated the full message delivery pipeline:

• GatewayRunner._handle_message_with_agent() returns dict with final_response, input_tokens, output_tokens, last_prompt_tokens, model
• _process_message_background() in base.py receives response string, sends via _send_with_retry()
• Token/cost data is already available in the agent result dict but is currently only surfaced through the /usage command handler
• display.show_cost is defined but never consumed — it should either be wired up or replaced with usage_footer

The display.platforms override mechanism already exists for tool_progress, so this would follow the same pattern.