Your Crab or Hermes Stays Silent After Setup: Connect an AI Model First
A freshly built server runs the agent, and the agent waits for an AI model. Until a model answers one test request, every chat stays quiet. This guide shows where that one step lives for the crab (OpenClaw) and for Hermes, where an API key comes from, and what to check when the silence stays.
You built the server, you opened the chat, you typed hello. Nothing came back. The server is running, the dashboard shows it green, and the agent says nothing at all.
In nearly every ticket we have seen with this picture, one thing was missing: an AI model. The crab (OpenClaw) and Hermes are the hands and the memory. The model is the part that thinks. Until a model is connected and has answered one real test request, conversations stay off. This guide shows where that step lives, for both agents, and where the key comes from.
Why the agent is silent
OneClickClaw gives you the server and installs the agent. The AI model comes from a provider you choose: OpenAI, Anthropic, Google, xAI or one of twenty more. You bring your own key or sign in with a plan you already pay for. That is what keeps the AI bill in your hands.
Right after setup, the agent has a home and a voice and nobody to think with. It waits. The dashboard shows the server as running because the server is running. The silence lives one level up, inside the agent, and that is where the fix goes.
A second common cause looks the same from outside: a key was saved, and the default model was never picked. The crab then answers a chat with the words "OpenClaw requires working inference". The steps below cover both.
OpenClaw: connect a model in three steps
Every OpenClaw server is configured inside the agent itself. Keys and models live there, so the steps start from the dashboard and end inside the crab.
- Open the crab. On the server page of the OneClickClaw dashboard press Open agent. The crab's own interface opens in a new tab.
- Go to Settings, then Models, then Model Setup. Pick one of three ways in: sign in with a paid ChatGPT plan (open More sign-in options and press Pair next to ChatGPT Device Pairing, then approve the device code on chatgpt.com), pair an xAI account, or press Add provider and paste an API key. Press Connect & verify. The crab sends one real request and switches conversations on only after a model answered.
- Pick the default model. A provider added through Add provider stores the key and waits for your choice. On the same Models page open the Default model dropdown, choose the model, save. This is the step most people skip.
One detail about the ChatGPT route: use the Device Pairing option. The plain ChatGPT Login button expects a browser on the same computer as the agent, and on a hosted server that round trip has nowhere to land. Device Pairing needs a paid plan from Plus up.
Now go to the crab's Home page. It is a full chat and works with no channel connected. Type hello. A reply means the model is live, and every channel you add later (Telegram, WhatsApp, Discord) uses the same model. The whole thing takes about three minutes once the key is in your hand, and the crab remembers the choice across restarts and updates.
Hermes: connect a model from the AI Model tab
Hermes keeps its own dashboard on your server, and OneClickClaw gives you a shortcut to the one card that matters.
- On the server page of the OneClickClaw dashboard open the AI Model tab.
- The AI Provider card shows the current model. Sign in with a plan you already have (ChatGPT, Grok, Nous Portal or MiniMax) using the buttons on the card.
- For an API key press Manage keys in Hermes. The Hermes dashboard opens on your server. Paste the key on its Keys page and pick the model.
A Hermes server with no provider stays quiet on every channel. Our own fleet checks flag exactly this state several times a month, always on a server a few minutes old, and the fix is always this card.
Where the key comes from, and how to keep it safe
OneClickClaw issues no AI keys. The provider does, and the provider bills you for what the agent uses. The three most common sources:
- OpenAI: platform.openai.com, section API keys.
- Anthropic (Claude): console.anthropic.com, section API Keys. Keys start with sk-ant.
- Google (Gemini): aistudio.google.com, Get API key.
A new key is shown once. Copy it straight into the agent's key field. Most providers also ask for a small prepaid balance before the first request; a key with zero balance connects and then answers nothing, which looks like the silence from the top of this page.
Keep the key inside the agent. Paste it in no ticket, no email and no screenshot. Our support deletes a key the moment it appears in a ticket and asks you to create a fresh one, because a key in a message is a key someone else can read.
Still silent? Four quick checks
- Default model chosen? On OpenClaw, Settings, Models: the Default model dropdown must name a model, and the page must be saved.
- Does the key have balance? Open the provider's billing page. A zero balance means the model accepts the key and refuses the work.
- Does the Home chat answer? Test inside the agent first. A silent Home chat is a model problem; a Home chat that answers while Telegram stays quiet is a channel problem, covered in our Telegram troubleshooting page.
- Still stuck? Every server page has a Stuck? Get help link. It opens a ticket for that exact server, and support can see the agent's state from our side.
Frequently asked questions
- The server is running. Why does the agent say nothing?
- The server and the agent are ready; the AI model is missing. Connect a provider inside the agent (OpenClaw: Settings, Models, Model Setup; Hermes: the AI Model tab on the server page). The agent answers once a model has replied to one test request.
- I added my API key and the chat still says OpenClaw requires working inference.
- The key is stored and no default model is chosen yet. Open Settings, Models inside the crab, pick a model in the Default model dropdown and save.
- Where do I get an API key?
- From the AI provider's own site: OpenAI (platform.openai.com, API keys), Anthropic (console.anthropic.com, API Keys) or Google AI Studio (aistudio.google.com). The key is shown once; copy it straight into the agent. Most providers need a small prepaid balance before the first request.
- Can I use my ChatGPT subscription without an API key?
- Yes, with a paid plan from Plus up. On OpenClaw open Model Setup, More sign-in options, and press Pair next to ChatGPT Device Pairing. On Hermes use the ChatGPT button on the AI Provider card of the AI Model tab.
Related guides
- How to Install Hermes Agent and Start Your First ChatInstall Hermes Agent, connect a model, verify your first chat and add Telegram when you want to message it away from your computer.
- Best AI Models for OpenClaw and Hermes Agent in 2026OpenClaw and Hermes Agent both need an AI model to think with. This guide lists twelve ways to connect one, by API key, by subscription sign-in or on your own machine with Ollama, with what each costs in a month.
- OpenClaw update failed: what to check and how to roll backWhen an OpenClaw update stops halfway, the program has usually put the old version back on its own before you even read the message. Your job is to read what it wrote down, fix the one thing it named and run the update again. This guide walks through that in order, using the official OpenClaw documentation and the project's public issue tracker, both read on the 6th of October 2026.
- Hermes Agent broke after an update: what to check firstThree checks in order when Hermes Agent stops working after an update: the version, the built-in doctor and the gateway, plus how to read the common warnings and get back to a working version.