Other

Hermes — AI Agents, Getting Started

AI-native agents that work your ERPNext data, with a human always in the loop.

Updated 2026-08-05

Overview

Hermes is an AI-native agent layer that runs on top of your existing modules, powered by the Claude API. Instead of a chatbot bolted onto your ERP, Hermes runs autonomous agents directly against your live data — reconciling accounts, flagging findings, and handling routine operational work end to end.

Every agent run is bounded by a per-run cost cap so spend never runs away, and every action an agent takes is logged to a full audit trail. When an agent hits a decision it genuinely can't make on its own, it pauses in a "Needs Input" state and notifies a human instead of guessing — and agents carry episodic memory across runs, so they get better at a task the more they do it.

Hermes can run standalone or embedded inside another module (it currently also runs inside PayrollOS, for example), so agent automation shows up inside the workflow it's automating rather than as a separate tool staff have to switch to.

Frequently asked questions

What happens if an agent gets stuck?

It stops and moves to a "Needs Input" state, and the person who requested the run gets notified — it never guesses past a decision point it isn't confident about.

Can I control how much an agent spends?

Yes — every run has a real, enforced per-run cost cap, not just a soft budget suggestion.

Is every agent action auditable?

Yes — every action Hermes takes is logged and traceable back to the specific run and agent that made it.

Related articles