n8n on your bench
Automation that is already running, already pointed at your model
1Overview
The automation tool, already installed and already wired to your model.
The bench starts n8n for you and hands it your provider, so the first workflow you build has a working AI node with no key to paste. This is the short road: open it, build one flow that reads a file from the shared drive and writes an answer back, and you have automated something real. → the full chapter, `n8n`, is the same tool on a server you administer, with credentials, queues and a public URL.
n8n is the automation tool the rest of this course leans on hardest. On the bench it is already running at localhost:5678, on your machine, with no account and no sign-up — and its AI node already answers with whatever "Which AI answers" says. Everything you build here is portable: the same workflow JSON imports into an n8n on a server. The bench is where you learn it; nothing about it is a toy version.
Any repeating job: renaming files, summarising a folder, watching a feed, calling an API on a schedule.
Your workflows live in the n8n-data volume. The shared drive is how files get in and out.
The encryption key that protects saved credentials lives in your bench folder — the setup keeps it across re-runs, which is why a re-install does not lock you out.
2Lessons 3
2.1 Open it and make the AI node answer
n8n is a canvas of nodes: something triggers, then each node does one thing to the data flowing through.
Get an AI answer out of n8n without configuring a single credential.
- Open n8n from its tile, or localhost:5678. The first run asks you to make a local owner account — it never leaves your machine.
- New workflow, add a Basic LLM Chain. Its model credential is already there: the bench seeded it against the gateway when it installed.
- Type a prompt and press Test step. The answer comes from whatever your dashboard says is answering.
- You'll see A green node with a real answer in its output panel.
- Takeaway On the bench, "wire up an AI" is a step that already happened.
2.2 A workflow that reads a file and writes one back
Do this first Open it and make the AI node answer
Automate something with a result you can open: drop a text file in the shared drive, get a summary next to it.
- Put a file in the shared drive. "Open the shared drive" on the dashboard, or the
shared/folder in HeidelbergBench — the two are the same place. - Read Binary File node, path
/shared/. Inside the container the shared drive is.txt /shared. - Basic LLM Chain with a prompt like "Summarise this in five bullet points:" and the file text.
- Write Binary File back to
/shared/. Execute the workflow.-summary.md
- You'll see A new file appears in the shared drive, on your own disk, written by the workflow.
- Takeaway The shared drive is the door in and out. Everything else stays sealed in its container.
- Check Why does the node path say /shared and not the folder name you see in your file manager?
2.3 When to move it to a server
A bench workflow runs while your laptop is on. That is the one limit that matters, and it is the reason the server chapter exists.
Do this first A workflow that reads a file and writes one back
Know exactly when the easy version stops being enough.
- A schedule needs a machine that is awake. Nightly jobs belong on a server, not a laptop that sleeps.
- A webhook needs a public address. localhost is not reachable from the internet, by design.
- Several people need one instance. The bench is single-user by construction.
- The move is an export. Download the workflow JSON, import it into the server n8n — the nodes are identical.
- You'll see The same workflow running somewhere that does not sleep.
- Takeaway Learn it here, move it when a schedule, a webhook or a second person appears. →
n8nis that chapter.
💬 Discuss this chapter
Ask, share, or report — over on the Heidelberg AI community forum.