hermes - 💡(How to fix) Fix Visible Multi-Agent Conversations in Gateway Chats [1 participants]

Feature Description

Add a native way to start a visible multi-agent conversation from a gateway chat.

Example:

/converse Have two agents discuss the best way to implement OAuth support

Hermes would then post each agent’s turn back into the same chat/thread:

Agent A: I’d start by...
Agent B: The main risk with that is...
Agent A: Good point, revised approach...
Agent B: Final recommendation...

Motivation

Hermes already supports tool-using agents, skills, background tasks, webhooks, and gateway chats, but there is not currently a simple way to have multiple real agents discuss something visibly inside the current chat.

This would be useful for planning, debugging, architecture discussions, research, brainstorming, and having one agent propose while another critiques.

Proposed Behavior

When a user runs something like:

/converse <topic>

Hermes should:

• start a bounded conversation between two or more real agents
• give each agent a clear role, such as proposer/reviewer
• allow agents to use normal available tools and skills
• post each turn visibly into the same chat/thread where the command was invoked
• label each turn clearly by agent/round
• stop after a configured turn limit or when a final answer is reached
• end with a short summary or recommendation

Requirements

• Agents should be real Hermes agents, not simple text relays.
• The user should not need to manually create webhooks or move messages between agents.
• The conversation should preserve the current gateway chat/thread.
• The feature should work across gateway platforms where possible.
• There should be guardrails against infinite loops or runaway cost.

Example Use Cases

• one agent proposes an implementation and another critiques it
• two agents compare libraries or approaches
• one agent investigates a bug and another challenges the diagnosis
• agents brainstorm ideas and converge on a recommendation