Add the operating cost to the open-source shortlist

TL;DR: A repository shortlist becomes useful when each entry explains what you must operate. Start with one tool and account for its model calls, stored data and exit path.

Evolving AI's nine-project carousel is a discovery list spanning research, writing, search and media. Its Open Notebook recommendation offers a concrete place to go deeper. Open Notebook is a self-hosted research application for working with sources, notes and model-assisted conversations.

The project's setup describes both the application and a database, followed by model-provider configuration. It supports hosted providers and local options. Choosing a local application therefore does not, by itself, settle where every inference request goes. That depends on the provider configuration you actually use.

For an engineer comparing tools, this is the column I would add beside the feature list: what must remain running for one useful output to arrive? The answer should include human attention as well as infrastructure.

My proposed trial would use a few public documents and one specific research question. I would record the time to a usable answer, inspect the cited passages and note any corrections required. Then I would account for model usage and the effort needed to preserve the notes and sources outside the application.

That last step belongs in the first trial. An export you can reopen elsewhere makes experimentation cheaper because abandoning the tool does not mean abandoning the work. If the available exit is a database backup, record the restoration effort honestly instead of calling it a portable notebook.

This is an evaluation design, not a benchmark I ran. The repository establishes documented capabilities and configuration choices; it does not establish my costs, the reliability of my deployment or the quality of answers on my documents. The other projects in the carousel need their own checks.

Open-source access is valuable precisely because it lets us inspect these choices. We should use that access to ask a better adoption question than whether the download costs money.

Shortlist the software by its useful output, and choose it with the running and leaving costs visible.