Hermes Guide
Hermes Agent Interactive Answers: Live Charts, Calculators and Forms in Chat
Switch on Interactive in Hermes Quick Chat and your bot can answer with a small tool instead of a wall of text: a tip calculator with a slider, a chart of your spending, a side-by-side comparison of three plans. You use it right in the chat, and moving a slider updates the result instantly without sending a new message. This guide covers how to turn it on, what it costs, what it cannot do, and how it compares with ChatGPT’s new Intelligent UI.
What interactive answers are
Normally a chat bot answers in text. With Interactive on, your Hermes bot can also reply with a live view: a calculator, a chart, a form, a comparison table or another small tool. The view sits in the conversation like any other message, but you can use it. Type a number, move a slider or flip a toggle, and the result changes on the spot. That interaction happens in your browser, so it sends no new message and spends no tokens.
The bot still decides when a view helps. Ask a plain question and you get a plain answer. Ask for something you would want to adjust or explore, and you get a view.
Interactive is available in the Hermes Agent Quick Chat on the OpenClaw Launch dashboard, for every Hermes bot. It is not available in Telegram, Discord or WhatsApp, because those apps cannot render a live view, and it is not yet in OpenClaw’s dashboard chat.
How to turn it on
- Open your dashboard and click Quick Chat on a Hermes bot.
- The first time, a small card above the message box explains the feature. Click Turn on, or Not now to leave it off.
- After that, click the ✦ Interactive chip in the message box, to the right of / commands, to switch it on or off at any time.
- When it is on, the message box gets a blue edge and the hint “Ask for a calculator, chart or form…”.
Interactive is off by default. The setting is saved per bot on our servers, so it follows you to another browser or device. Click the chip again to turn it off.
Prompts to try
- “Make me a tip calculator.”
- “Compare these 3 phone plans”, followed by the plan details.
- “Chart my weekly spending”, with the numbers pasted in.
- “Build a loan payment calculator with sliders for amount, rate and term.”
- “Make a packing checklist form for a 4-day trip.”
- “Search for the current prices of these laptops and put them in a comparison table.” The bot searches first with its own tools, then builds the table from the results.
How it works
When a reply should be a view, the bot writes it in a small markup called DIL inside a ```dil code block. DIL looks like JSX tags with a little JavaScript: if and each blocks, state, and event handlers. The chat turns that block into a live view.
The building blocks are:
- Layout: rows, columns and cards.
- Text, buttons, inputs, sliders, toggles and selects.
- Tables, and line, bar, area and pie charts.
- 246 icons and a copy-to-clipboard action.
Every view has a View code toggle that shows the source, so you can see exactly what the bot wrote.
Safety
The view’s code runs in a locked-down sandbox: a sandboxed frame plus isolated background workers. It cannot reach your page, cookies, storage, network or account, and it can only draw from the fixed set of components above. If a view is broken, too large, or hangs, the chat shows its source code instead, and the rest of the conversation carries on normally.
Limits
- No network calls or file access from inside a view. A view cannot fetch live prices by itself. The bot gathers data with its own tools, such as web search, and then builds the view around it.
- Views live in the web chat only. Messaging apps get plain text.
- Very old browsers without nested-worker support show the code instead of the view.
What it costs
We measured this on a real Hermes bot:
| Situation | Extra cost |
|---|---|
| Interactive off | No format note on new messages. In a chat where it was on before, earlier notes stay in the history until you start a new chat. |
| Interactive on, plain text reply | Roughly 3 to 8 percent more input tokens, mostly prompt-cached. No extra output. |
| Interactive on, reply builds a view | About 500 to 1,500 extra output tokens, the size of a medium to long text answer. |
| Using a view (sliders, inputs, toggles) | None. It runs in your browser. |
It works with whatever model the bot uses, from your plan or your own key, so the price per token is whatever that model charges. Turning Interactive off does not break views already in the chat history.
Hermes Interactive vs ChatGPT Intelligent UI
On October 7, 2026, OpenAI began rolling out GPT-6 in ChatGPT with a feature it calls Intelligent UI: answers that mix text with charts, tappable buttons, forms and small interactive tools. The rollout started with Plus, Pro, Business and Enterprise, with Free and Go following on October 8, according to OpenAI’s announcement. The idea is close to Interactive in Hermes Quick Chat. The products around it are different.
| Hermes Interactive (OpenClaw Launch) | ChatGPT Intelligent UI | |
|---|---|---|
| Where it runs | Hermes Quick Chat in the OpenClaw Launch dashboard | The Chat tab in ChatGPT |
| Model | Any model your bot uses, from the plan or your own key | GPT-6: Sol on paid plans, Luna on Free and Go |
| On or off | Off by default. You switch it on per bot. | Automatic. The model decides how each answer is laid out, and you can also ask for a UI. |
| Cost | Tokens from your plan or your own key; extra tokens only while on (see above) | Included in ChatGPT plans, including Free |
| Messaging apps | Web chat only; Telegram, Discord and WhatsApp get plain text | Inside ChatGPT |
| The agent behind it | Your own always-on Hermes agent with its own tools, memory, skills and connected platforms | ChatGPT, OpenAI’s assistant |
| How views are drawn | DIL markup run in a sandboxed frame and isolated workers, limited to a fixed component set | A library of native, streamable components plus a compiler |
| Format | DIL in a ```dil block | Not publicly documented |
Where ChatGPT is ahead
- It needs no setup. Intelligent UI is on in the Chat tab and the model chooses when to use it. OpenAI says it trained GPT-6 to decide how each answer is laid out.
- It is a polished, first-party feature built together with the model that uses it.
Where a Hermes bot is different
- It is your own agent. A Hermes bot runs around the clock with its own tools, memory and skills, so a view can be built from data the agent just gathered, files in its workspace, or what it remembers about your work.
- Any model. Interactive is not tied to one vendor. Use a model from your plan or bring your own key.
- You control the switch and the cost. Off means no format note on your messages. On costs a measured, small amount, and only view-building replies cost more output.
- The same agent stays on Telegram, Discord and other platforms in plain text, while the dashboard chat gets the views.
If you mainly want a general assistant with rich answers, ChatGPT is the simpler choice. If you want interactive answers from an agent that works on your own tasks with your own tools and models, that is what Interactive in Hermes Quick Chat is for. The comparison above reflects OpenAI’s announcement and press coverage as of October 8, 2026.
Getting started
New to Hermes? Read What Is Hermes Agent?, then deploy a bot from Hermes hosting. Open Quick Chat, click ✦ Interactive, and ask for a calculator, chart or form. For other ways to talk to the same agent, see Hermes Agent GUI options.
Interactive Answers FAQ
Does Interactive cost more?
A little, and only while it is on. Each message carries a short format note, which adds roughly 3 to 8 percent to input tokens, and most of that is prompt-cached. Replies in plain text cost no extra output. Only a reply that actually builds a view costs more: about 500 to 1,500 extra output tokens, similar to a medium or long text answer. With Interactive off, no format note is added to your new messages. Earlier notes in a chat where it was on stay in that chat’s history, so start a new chat for a clean slate.
Is it safe to run code the bot wrote?
The view’s code runs in a locked-down sandbox: a sandboxed frame plus isolated background workers. It has no access to your page, cookies, storage, network or account, and it can only draw from a fixed set of safe components. If a view is broken, too large, or hangs, the chat shows its source code instead and the rest of the conversation is unaffected.
Can I use interactive answers in Telegram, Discord or WhatsApp?
No. Those apps cannot render a live view, so Interactive works only in the web chat: Hermes Quick Chat in the OpenClaw Launch dashboard. Your bot keeps answering on Telegram and other platforms in plain text as usual.
Can I turn it off?
Yes. Click the ✦ Interactive chip again. It is off by default, and the setting is saved per bot on our servers, so it follows you across devices. When it is off, no format note is added to your new messages. In a chat where it was already on, the earlier notes stay in the history and the bot may still build views there, so start a new chat for plain answers. Views already in your chat history keep working.
Which models does it work with?
Whatever model your Hermes bot uses: any model on your plan, or your own key through BYOK. The bot writes the view as markup in its reply, so it does not depend on one vendor.
Can a view show live prices or fetch data?
Not by itself. A view cannot make network calls or read files. The bot fetches data first with its own tools, such as web search, and then builds the view around what it found.
Does OpenClaw have this too?
Not yet. Interactive is in the Hermes Agent Quick Chat for every Hermes bot. OpenClaw’s dashboard chat does not have it at the moment.