Get a Second Opinion from Hermes and OpenClaw in One Chat

Compare both bots' replies in one conversation, without copying the first answer into another chat.

In a live test, a group chat was asked for two opinions on a simple question: should a new startup use Postgres or MongoDB?

OpenClaw answered first. Then it brought Hermes into the conversation to give its own recommendation. The user had not typed an @mention or copied the answer into another chat.

Both picked Postgres, but explained the choice differently. OpenClaw focused on database features. Hermes focused on how uncertain a startup's future queries can be. The useful part was being able to read both replies, with each one attributed to the bot that produced it.

I build OpenClaw Launch, and this is now possible in the website's group chat: you can put Hermes and OpenClaw bots in one conversation and let them pass a question to each other.

How the conversation works

A group has a default bot: the first one you select when creating it. Send a message without addressing anyone, and that bot gets the first turn.

It can answer directly. When another perspective is useful, it can mention another member of the group. That sends the receiving bot your original question together with the first bot's reply. The receiving bot then answers under its own name in the same conversation.

You can also address a specific bot yourself with its @mention. Asking for a second opinion gives the default bot a clear reason to involve someone else, without requiring you to route every message manually.

For someone who already asks two agents to review the same decision, this puts the exchange in one place. You can follow the question, the handoff, and the response without moving the first answer between separate chats.

Try it with your own bots

Start with a running Hermes bot and a running OpenClaw bot in your OpenClaw Launch account.

1. Open the dashboard's chat sidebar and click the + button.
2. Choose “Create group chat.”
3. Give the group a name and select the bots to include. Pick your preferred default responder first.
4. Create the group and send a question that would benefit from a second perspective.

A useful first prompt is:

Give me your recommendation on this decision, explain your assumptions, and ask another bot in this group for a second opinion.

Add the decision and the relevant facts underneath it. For a writing task, include the draft. For a technical choice, describe the constraints. The more self-contained the question, the less the receiving bot has to guess.

Look for a reply attributed to the second bot. That is how you can tell it actually joined the exchange.

Give each bot a clear job

A second opinion is the easiest place to start because the request has a clear endpoint. One bot gives its recommendation, another examines it, and you decide what to do with the answers.

You can make the request more specific as well. Here are two examples to try:

• Review a draft: ask one bot to propose a version, then ask another to identify unclear wording or unsupported claims.
• Review a plan: ask one bot to outline the steps, then ask another to look for missing prerequisites and assumptions.

These are suggested workflows, not measured results. What the bots can contribute depends on their models, instructions, and available tools. Naming a bot “researcher” does not give it access to sources it cannot already reach.

For a first group, I would start with two bots and one concrete question. It is easier to judge whether the extra reply helped when you know what you wanted it to check.

What is shared, and what stays separate

The group gives you one visible conversation. Each bot remains a separate agent with its own setup. Creating a group does not merge their private chat histories, files, or tools.

The handoff carries your question and the sending bot's reply. If a task depends on a document or detail from elsewhere, include the necessary information or make it accessible to the bot that needs it.

The second bot also sees the first bot's answer. This is a second perspective in a conversation; it is not a blind evaluation. Agreement does not prove a conclusion is correct, and disagreement still needs your judgment.

Each additional bot response involves another agent run, so it adds model usage. Group chat bounds the number of handoffs and prevents the same member from being called repeatedly within one user turn. That keeps the exchange finite; it does not make additional replies free.

Start with a question you already ask twice

If you already take a draft, a plan, or a technical decision to more than one agent, try that same task in a group. Read what each bot adds and whether its assumptions hold up.

Hermes and OpenClaw group chat is available on OpenClaw Launch. Create a group with your bots and ask for a second opinion.

The question I want to explore next is which pairings add useful information, and when a single bot would have been enough.