# 08 — Hermes as an employee

The fun part: **Hermes can be a managed employee inside a Paperclip company.** Then
Paperclip assigns it issues, and it works using its full toolkit (memory, skills,
browser, files — everything) and reports back.

## The idea

Normally **you** talk to Hermes directly in a chat. In Paperclip, an **adapter**
plugs Hermes in as a worker: Paperclip hands it an issue, Hermes does the work
automatically, and Paperclip tracks the run and result. You supervise via
approvals and heartbeats instead of typing prompts one at a time.

## How it locks together

- **Adapters** are the plumbing. The **Hermes adapter** (`hermes_local`) runs
  Hermes Agent as the employee's engine.
- Paperclip spawns Hermes in single-query mode (`hermes chat -q ...`): it gets
  the task, uses its 30+ tools / 80+ skills, does the work, reports back, exits.
- Because it's plain Hermes under the hood, it gets **persistent memory** and
  **skills** — not just a fresh model each time.

## Do you have this already?

Your 3050 has Paperclip and Hermes on the same box, and the workgroup project
already registered/demonstrated the Hermes adapter — the "Wind Sculpture Agent"
was built that way. So the pieces exist. **This project doesn't rebuild them** —
it teaches you to understand and drive that connection yourself.

Useful to know:

- Official docs: `docs.paperclip.ing/reference/adapters/hermes/`
- The adapter repo: `github.com/NousResearch/hermes-paperclip-adapter`

## The mental model

```
Paperclip  company
    │   creates issues, tracks runs, handles approvals & budget
    ▼
Adapter (hermes_local)
    │   spawns Hermes per task
    ▼
Hermes Agent  (you!)
    │   uses tools, memory, skills
    ▼
reports result back to Paperclip
```

## Why this is powerful for you

It's the difference between **you driving one AI manually** and **an org that
runs agents on a schedule with checks and budgets**. Same brain, but Paperclip
gives it a job, a budget, a check-in cadence, and a paper trail.

## Try it

On your live instance, list the agents and note which one is your Hermes-powered
worker. Then read `docs/09-practice-runs.md` for the safe hands-on part.
