AI generation

Generate a list

Open Generate and describe what you need. Your prompt doesn't go to a single model — it goes to a council of expert gnomes: a steward frames the topic, guild experts draft variants, a critic checks them, an elder synthesizes the final list. You get several candidates — compare them, regenerate for more, and pick one.

While the council runs, you can see which gnome proposed what and why the list turned out the way it did. The chosen candidate becomes a private draft owned by you. Nothing is published automatically: you review, edit and publish when it's actually good.

The council draws on precedents — similar lists already published on SetFork — so the catalog gets better on its own: the more good lists exist, the sharper the next ones. On the international service, generation runs on OpenRouter models.

Improve with AI

Inside the editor, select Improve with AI and give an instruction — "add verification sub-tasks to every step", "make commands idempotent", "translate descriptions to English". The AI edits the blocks in place; undo works as usual.

Dig deeper

Every list item has a "Dig deeper" button — a mini-conversation about that specific item: the reasons and sources, the mechanism ("why it works and when it does NOT"), the fine print. A shoring gnome keeps each answer grounded in facts rather than invented; if there's no data, it says so. Every layer is on-click, so you only spend quota on what you actually want to dig up.

Living lists

SetFork watches the health of published lists: broken links are caught both at generation time and by a background sweep, and stale steps get flagged for update. A list isn't a one-off text — it's an artifact kept current.

AI change notes

When publishing a new version, press the ✨ button next to the change note field: the note is generated from the actual diff between versions, like a commit message assistant.

Moderation and usage

• Only public content passes through AI moderation (MLCommons taxonomy). Unsafe content isn't deleted — it just can't go public.
• Every AI call is metered with its real cost. Admins see usage per user in the admin panel; you can see your own spend in Settings.