The app store
Three more apps, one click each — and every one of them is taught in this course
1Overview
At the bottom of the dashboard is a short shelf: Flowise, Langflow and Open Notebook. It is short on purpose. An earlier version offered nine apps, and the seven with no chapter behind them were removed — a tile that teaches nothing is a tile a student installs, does not understand, and pays for in memory. This chapter installs one, watches what it actually does to the machine, and follows the link to the chapter that teaches it properly. → `bench-dashboard` is the page it lives on; `bench-under-the-hood` is what an install writes to disk.
The store is one JSON file the bench ships — the entry IS the install. The folder written into stacks/ and the container that gets started are both generated from the tile, so a tile can never offer something the bench cannot actually run. What a tile promises is worth naming: one sentence you can act on, a rough size in gigabytes, the memory free on your machine right now, and the chapter that teaches it. The size is checked against what is free BEFORE the install rather than after, which is the difference between a warning and a post-mortem.
When a chapter asks for a tool the bench does not ship — Flowise, Langflow or Open Notebook today.
Every entry names the chapter that teaches it. That is the rule the shelf is built on: the store and the curriculum are one list.
Each installed app holds memory the whole time it runs. On 8 GB, install what a lesson asks for and stop it afterwards.
2Lessons 2
2.1 Read a tile before you press Install
Each tile is one app the course teaches. The entry in the bench's store file is also the install recipe, so what you see is what will run.
Know what a tile is telling you, and pick one on purpose.
- One sentence you can act on. "Drag boxes into a chatbot or agent, then call it from anything as an API" — not a category like "LLM orchestration".
- Two numbers. What it needs, and what is free on this machine right now. 0.9 GB needed against 12.7 GB free is a yes; against 0.4 GB free it is a no, and the tile says so before you click.
- A chapter link. Every app on this shelf has a chapter. Seven apps that did not were removed from the store in August for exactly that reason.
- You'll see Three tiles, each priced against the memory your machine has free at that moment.
- Takeaway A store tile is a promise to teach the thing it installs. That is why the shelf is three apps and not thirty.
2.2 Install Flowise and watch what it costs
Do this first Read a tile before you press Install
Add one app to your bench, see it appear as a tile, and see the memory it takes.
- Press Install on Flowise. The bench writes
stacks/flowise/compose.yamland starts the container. The image is about 900 MB, so the first install is a few minutes. - Watch the installed-apps total. Flowise joins the grid with its own memory figure. That number is the rent you now pay while it runs.
- It already knows your model. The tile inherits the gateway, so Flowise answers with whatever "Which AI answers" says — no key to paste.
- Open it, then follow the chapter link. The bench got it running;
flowiseteaches what to do with it.
- You'll see A new tile in Installed apps, green, with an Open button pointing at localhost:3100.
- Takeaway Installing from the shelf is one click and one download; everything after that is the same as any other app on the bench.
- Check You install Flowise, use it for one seminar, and want the memory back tomorrow — what do you press, and what happens to the flows you built?
- Cost Disk: about 0.9 GB. Memory: a few hundred megabytes while it runs, zero when stopped.
💬 Discuss this chapter
Ask, share, or report — over on the Heidelberg AI community forum.