Skip to content
ToolscaledToolscaled docs

Hermes Agent on your computer

Hermes runs on your computer and handles the agent steps of your runs.

  • Every minute, a small Toolscaled script checks for a run that needs an agent.
  • No run → Hermes doesn’t start and uses no tokens.
  • A run → Hermes starts with that run’s instructions and works it with your Toolscaled tools.
  • Nothing of the workflow stays on your computer.

Hermes only takes runs while your computer is on. For an agent that works with your computer off, use a Claude task or a ChatGPT task.

Each run uses tokens on your own model plan or API account. Toolscaled sets no limit. To set one → Pause, limits and stop.

Setup

Section titled “Setup”

  1. In Toolscaled, open Get started > Set up an agent > Connect Hermes, or Settings > Agents.

  2. Paste this into Terminal (Mac, Linux):

    Terminal window
    curl -fsSL https://app.toolscaled.com/agents/hermes/setup.sh | bash

    Or into PowerShell (Windows):

    Terminal window
    iex (irm https://app.toolscaled.com/agents/hermes/setup.ps1)

  3. The setup shows the cost and asks for your yes.

  4. It asks to connect your ChatGPT account. Press Enter to connect it, or type n to pick another: a Claude API key (not your Claude plan), a Nous Portal plan, or the free Reflection Beam 1.0 (for a test only).

  5. Approve Toolscaled in the browser when it opens.

What the setup does:

  • Installs Hermes Agent with its own profile, toolscaled. Your other Hermes use doesn’t change.
  • Saves the Toolscaled script and makes one job that takes one run at a time.
  • Installs the Hermes gateway as a service, so the job starts at login and again after a crash.

If the setup stops, run the same line again. It only does what’s missing.

Once Hermes checks in, Set up an agent in Get started shows Done, and Settings > Agents > Your agents lists the job.

Setup from a chat

Section titled “Setup from a chat”

Paste this into a chat that has Toolscaled. The chat takes you through the setup step by step, or runs it for you if it can run commands on your computer:

Set up my agents for Toolscaled: read skill_guide('agent-runner') and take me through it, one step at a time.

When it doesn’t check in

Section titled “When it doesn’t check in”

A job with no check-in for 10 minutes shows up in Action items: “Your agent has not checked in for N minutes”. Usual causes: the computer is off, the gateway stopped, or a Hermes update broke the job. In Terminal, hermes -p toolscaled cron status tells you which.

When a run fails

Section titled “When a run fails”

If Hermes stops with an error before it finishes a run (for example, your model provider removed the model and answers “not found”), the job’s next tick gives the run back. The run comes back to you with Hermes’s error on it.

Logs are in the profile’s folder (hermes profile show toolscaled names it):

  • logs/agent.log and logs/errors.log → the agent.
  • toolscaled/log.jsonl → each claim and release.

For a model error, pick another model with the setup line and --model:

Terminal window
curl -fsSL https://app.toolscaled.com/agents/hermes/setup.sh | bash -s -- --model

On Windows, run $env:TOOLSCALED_MODEL = 'pick' in PowerShell, then the setup line. Each run of the setup also checks that your model answers.

Stop Hermes

Section titled “Stop Hermes”
  • Pause one job: hermes -p toolscaled cron pause <job id>. hermes -p toolscaled cron list shows the ids, and cron resume starts a job again. A run Hermes already holds, it finishes.
  • Remove it all: hermes -p toolscaled gateway uninstall, then hermes profile delete toolscaled. Then press Remove on the job in Settings > Agents > Your agents.