How do I know my hosted Hermes is actually working?

Your hosted Hermes is working once it has replied to a message using the model you chose. A computer that’s switched on proves the machine is up, and a model that’s been set proves the setting saved. Neither shows that Hermes can actually answer you. On Vavio, ask the AI you’ve connected to check your Hermes, or run vavio instance <id> status. Either one reads four facts from the machine: whether it’s reachable, whether the gateway is running, which model is set and whether that model has replied yet. If it hasn’t replied, send Hermes one message. That reply is the proof.

The four checks, in order

Each check depends on the one before it. A failure early in the list explains every failure after it, so work from the top.

Check	What it proves	What it doesn’t prove	If it fails
1. The computer is reachable	Hermes’s computer is on and answers Vavio.	That Hermes has a model or can reply.	If the computer has stopped, Vavio starts it again, and the status says it’s restarting. Check again in a few minutes.
2. The gateway is running	The background process Hermes uses for chat channels and scheduled jobs is up.	That a model will answer.	Ask your AI to restart the gateway. Channels such as Telegram reconnect by themselves.
3. A model is set, with access	Hermes knows which model to use and has a saved key or a finished sign-in for that provider.	That the provider accepts the request, or that your plan or credits cover it.	Save a key on the Vavio page your AI gives you, or sign in to your provider, then choose a model.
4. Hermes has replied with that model	A conversation reached the model and came back. This is the setup working.	That a later task will succeed.	Send one message. If no reply comes back, read the recent errors (below).

Status on Vavio reports check 4 as replied, not yet or not checked. “Not yet” also covers a model you’ve just changed: Hermes’s last reply came from the previous model, so the new one still has to answer once. “Not checked” means the machine couldn’t say, which happens on older installations. Send a message and see what comes back.

Why choosing a model isn’t proof

Setting a model saves Hermes’s choice for new conversations. It doesn’t show that the provider will answer. The provider can still refuse the first real message: the key may have expired, a free model may be rate-limited or the account may have run out of credit. Vavio’s own tools say this outright. After a model change, the next step they give your AI is to ask you to send Hermes one message and report whether it replied.

So a dashboard showing the model you picked tells you what Hermes will try to use. It doesn’t tell you whether that worked.

Check it from the AI you already use

Connect your AI to Vavio’s MCP server, then ask in plain words:

Check my Vavio Hermes. Is the computer reachable, is the gateway running, which model is set, and has Hermes replied with it yet? If something’s wrong, show me the recent errors.

Your AI reads the instance live from its machine. It gets the Hermes version, gateway state, model and provider, which provider keys are saved, each connected channel and its state, the latest errors and the next setup step. Ask it to run Hermes’s doctor, too. That’s Hermes’s own diagnostic, and Vavio returns its verdict as healthy or as a list of problems. Optional tools that stay off until you add their keys are listed separately and aren’t problems.

Vavio has tested this end to end with Claude Code and Codex. In Vavio’s production test on 7 November 2026, run with a test discount, Codex bought the hosting. The computer was ready about 15 minutes after checkout. A model was connected on a separate Vavio page, Hermes replied to a message, and Claude Code then checked the machine’s status and health.

Check it from a terminal

The Vavio CLI reads the same facts without an AI in between. It needs Node.js 20 or later.

npm i -g @vavio/cli
vavio login                      # opens your browser to approve access
vavio instances list             # shows each Hermes and its id
vavio instance <id> status       # reachable, gateway, model, reply, errors, next step

To run Hermes’s doctor from the terminal, pass the instance id and a short label of your choosing:

vavio call run_check --args '{"instance_id":"<id>","check":"doctor","request_key":"doctor-check-1"}'

Use a new label each time you run a fresh check. The doctor usually answers within half a minute. If it takes longer, the result gives you an operation id to read later with vavio operations get <operation-id>.

Working once isn’t the same as a job working

A reply proves the setup works. It doesn’t prove that every task will finish. A computer that stays online can still have a job stopped by a missing credential, an approval Hermes is waiting for, a model limit or a tool that isn’t connected.

Vavio’s own early testing hit exactly this. Hermes was online and replying when it was asked to connect Gmail. It looked up the available connectors, correctly reported that Gmail wasn’t one of them, and then asked for a technical connector name. The model and the machine were both fine. The job was blocked on a missing connection.

For any job you care about, check its result, not just that Hermes is running. Ask Hermes what it did, look at the output it produced and read the recent errors if the answer looks incomplete. A job that Hermes says it finished and that produced what you asked for is working. Anything short of that is still unconfirmed.

If you run Hermes yourself

The same order applies on your own computer or VPS. Nous Research’s CLI covers each check:

• hermes status shows agent, auth and platform status.
• hermes doctor diagnoses configuration and dependency problems.
• hermes logs errors shows recent errors.

Then send one message and wait for the reply. On a server, the messaging gateway must also be running as a service for channels and scheduled jobs to keep working after you log out. The VPS or managed hosting guide walks through that setup.

Next step

• Already have a Vavio Hermes? Ask your AI to check it with the prompt above, or run vavio instance <id> status. If it says Hermes hasn’t replied yet, send it one message.
• Still deciding? Read what Hermes Agent costs each month, check the current terms on the pricing page and the FAQ, or create your Vavio account. No card is needed until you choose hosting.