Skip to content
vavio

Can ChatGPT or Claude scheduled tasks replace Hermes for a monitoring job?

Often, yes. If you already pay for ChatGPT or Claude, the check runs daily, weekly or hourly, and reading the result inside their app is good enough, their scheduled tasks can do the job with no extra bill or setup. Hermes Agent is the better fit when the result has to reach you in Telegram, Discord, Slack or email, when you want to hear only about real changes without paying for a model run every time, or when you want a model that OpenAI or Anthropic doesn’t offer. Hermes runs its schedule only while its gateway is running, so it needs a computer that stays on.

These comparisons come from each vendor’s documentation, reread on 10 November 2026. They are not a test of how well each option does the job. Anything a page doesn’t mention, such as Telegram delivery from ChatGPT or Claude, is treated here as unverified, not as missing.

The options side by side

How often Where you see the result Runs with your laptop off Models
ChatGPT scheduled tasks Daily, weekly or custom schedules. Tasks inside a chat can also repeat every few minutes as a follow-up loop. Some plans can also start a task on a Gmail, Slack or GitHub event The Scheduled view in ChatGPT, with an unread mark when a run needs your attention Not stated for web tasks. Desktop project tasks need the machine on and the ChatGPT app running OpenAI’s
ChatGPT dots Always on. It decides when to pause and wake, and can do scheduled check-ins you ask for ChatGPT apps, phone calls, Slack and Microsoft Teams. Texting is in beta for US Pro users Yes. It runs in the cloud with its own computer OpenAI’s
Claude scheduled tasks in Cowork Hourly, daily, weekly or weekdays Each run is a Cowork session under Scheduled in the Claude app Yes, unless the task needs files or apps on your computer Anthropic’s
Claude Code scheduled tasks (research preview) Hourly, nightly, weekly or a custom cron expression. Anything more often than once an hour is rejected A Claude Code session per run. Connectors can act for you, for example posting to Slack Yes. It runs on Anthropic’s cloud Anthropic’s
Hermes Agent cron Checked every 60 seconds. Accepts intervals like every 2h, times like weekdays at 9am and standard cron expressions Telegram, Discord, Slack, WhatsApp, Signal, email and more, or saved locally Only if the computer running its gateway stays on Any provider you configure

Which plans include them

  • ChatGPT scheduled tasks: OpenAI’s task documentation says only that availability depends on your plan and workspace. We couldn’t load OpenAI’s help centre on 10 November, so check your own plan.
  • ChatGPT dots: Pro 100, Pro 200 and Pro 500 (for users 18 and over), Business Premium, and eligible Enterprise, Edu and Healthcare workspaces. The Pro plans exclude the European Economic Area, the UK and Switzerland. Rollout to eligible Pro and Business Premium accounts is gradual.
  • Claude scheduled tasks in Cowork: all paid Claude plans, which Anthropic lists as Pro, Max, Ultra, Team and Enterprise.
  • Claude Code scheduled tasks: Pro, Max, Ultra, Team and Enterprise. Runs count against your subscription usage, and each account can start up to 100 scheduled runs an hour.
  • Hermes Agent: free and open source. You pay for the model and for whatever computer runs it.

When the assistant you already pay for is enough

Use what you have if all of these are true:

  • Daily, weekly or hourly is often enough. A morning summary of an inbox, a weekly look at a competitor’s pricing page or an hourly check of a status page all fit.
  • You’re happy to read the result in the app. ChatGPT and Claude Cowork both collect runs in a Scheduled list, and ChatGPT marks runs that need your attention. Claude Code lists each task’s runs on its own page.
  • Their models suit the job. If GPT or Claude models already do this work well for you, there’s nothing to switch.
  • You’d rather not run anything. There is no server, process or model account to look after.

If you use Slack or Teams at work, a dot or a Claude Code task with a Slack connector can reach you there. Check each vendor’s current limits and data handling before connecting work accounts.

When Hermes is the better fit

  • The result has to reach you somewhere else. Hermes can deliver a job’s output to Telegram, Discord, Slack, WhatsApp, Signal, email and other platforms, or to several of them at once. None of the ChatGPT or Claude pages above mentions Telegram.
  • You only want to hear about real changes. If Hermes’s reply is [SILENT], starts with it, or ends with it on its own line, it saves the output but sends nothing. Failed jobs still notify you. A job can also run a script first. If the script prints {"wakeAgent": false}, Hermes skips the model for that run, so a quiet check costs no model usage.
  • You want a particular model. Hermes works with any provider you configure, including OpenRouter, hosted open-weight models and any OpenAI-compatible endpoint. ChatGPT uses OpenAI’s models and Claude uses Anthropic’s.
  • You need checks more often than hourly from a cloud task. Claude Code rejects schedules more frequent than once an hour. Hermes checks for due jobs every minute. ChatGPT describes minute-based intervals for follow-up loops inside a chat; its page doesn’t say how often a standalone scheduled task can run.
  • You want the job to use your own Hermes setup. A job can attach Hermes skills and pick its own model and provider. Each run starts a fresh session with no memory of earlier runs, unless you turn on the option that passes the previous run’s output forward.

You can set up a Hermes job by asking it in plain language. Nous Research’s example is: “Every morning at 9am, check Hacker News for AI news and send me a summary on Telegram.” From the command line, hermes cron create "every 2h" "Check server status" does the same kind of thing.

The catch: Hermes needs a computer that stays on

ChatGPT and Claude run their tasks on their own infrastructure. Hermes runs its schedule inside its gateway, the background process that also connects your chat platforms, so the job only fires while that computer is running. If the gateway is down when a job is due, Hermes runs the job once when it comes back, but only within a catch-up window of half the job’s period, between 2 minutes and 2 hours. After that it waits for the next slot. Should Hermes run on your own computer or in the cloud? explains what that means for a laptop.

That leaves three ways to run it: a computer at home that never sleeps, a server you set up and maintain yourself, or a managed host.

On Vavio, $14.99 a month gives your Hermes its own computer in Europe with 8 GB memory and 100 GB SSD storage, running independently of your laptop. Vavio sets Hermes up and looks after that server. You choose the model, and your provider bills its usage separately. You can manage your Hermes from the AI you already use, through Vavio’s MCP server or CLI.

The delivery channels above come from Nous Research’s documentation. We haven’t yet tested a scheduled job delivering to Telegram or another chat app from a Vavio computer, so connect your channel and confirm the first delivery yourself.

A computer that stays on doesn’t guarantee that every run finishes. A model limit, a missing credential, an approval or an error can still stop a job. Before you rely on a monitor, check where its first two runs actually delivered their output.

How to decide

Write down the one check you want to run, then answer three questions:

  1. Where do you want to see the result? If the ChatGPT or Claude app is fine, start there.
  2. How often, and how often will nothing have changed? For frequent checks that are usually quiet, Hermes’s script gate saves model runs.
  3. Which model do you want doing it? If the answer isn’t an OpenAI or Anthropic model, use Hermes.

If the first answer is “in the app” and the rest don’t matter to you, you don’t need Hermes for this job.

Next step

Sources

Written by Vavio with AI assistance and checked against the sources listed at the end of the article.

More from the Vavio blog: Hermes Agent guides