We’re building Ever Works in public. Some things may be incomplete, missing, or broken while we continue improving the platform.We’re building Ever Works in public — expect a few rough edges.
Use Case
Directories are some of the best SEO assets on the web — and the hardest to maintain. Ever Works builds a directory in any niche, researches the entries, writes the descriptions, validates the sources, and keeps the catalog current by adding new items and pruning dead ones, on a schedule.
The platform finds and vets items for your niche.
On-brand copy for every listing, generated at scale.
Links are checked for reachability and quality.
New items added, dead ones pruned, on a schedule.
Directories are a first-class Work type. Add filters, categories, comparisons, and collections — and let the runtime keep the whole catalog accurate as the space changes.
Four moves, all of them in the dashboard, from an empty prompt box to a live catalogue that refreshes itself. Nothing below needs a config file.
Press + New in the sidebar, type what the directory should cover, and choose the Directory chip — you land on the create form with the kind already set and your brief already in the chat. Open Advanced Settings if you want to choose the machinery: the pipeline that orchestrates the run, the AI provider behind its simple, medium and complex model tiers, the search provider that powers discovery, and optional screenshot capture and content extraction. Submitting creates three repositories — data, markdown and website — and starts the run, which streams on the Work’s Worker tab where you can cancel mid-flight and fix the brief.
Items opens on five sub-tabs: Browse Items, Categories, Tags, Collections and Source Health. Every item takes exactly one category and may sit in at most one collection, and the standard pipeline keeps tagging tight — one to three tags per item — so the catalogue stays navigable instead of turning into tag soup. All three lists live as categories.yml, tags.yml and collections.yml in your data repository, so a rename is a commit. To bring in a list you already keep in a spreadsheet, switch on Item Import & Export in the Work’s Settings (both directions are off by default) and run the wizard: upload the CSV or XLSX, map the columns, preview the per-row errors, pick a duplicate strategy.
Enable comparisons in the generator settings and Worker → Comparisons writes head-to-head pages: Generate Next takes the best pair nobody has compared, Compare Items lets you name both sides, Generate All walks the backlog behind a progress bar you can stop — and a category needs at least three items by default before a pair is produced, a minimum you can configure. On Source Health, turn on scheduled checks: reachability separates a real 404 from an ambiguous timeout, an AI pass judges whether the link is genuinely specific to that item, and Apply suggestion swaps in the replacement URL it proposes. Turn on community pull requests and an hourly job reads contributor PRs on your repository, extracts structured items from the diff, commits them with a link back to the PR and comments with what it added — with no human in the loop, so enable it where that is what you want.
On the Deploy tab pick a target: Ever Works managed hosting, which needs no infrastructure of your own; your own Vercel token; or a kubeconfig for your own Kubernetes cluster. The shared customer cluster behind managed hosting is still being brought up environment by environment — where it is not yet provisioned for your installation a deploy says so plainly instead of landing somewhere unexpected, and you can point the Work at your own Vercel token or kubeconfig in the meantime. Add a custom domain in the same place — type the hostname, copy the exact CNAME or A record it shows you, press Verify DNS. Then set a cadence on Worker → Schedule, one of seven: hourly, every 3, 8 or 12 hours, daily, weekly or monthly — and press Run now whenever you do not want to wait. Comparisons keep generating on their own six-hourly cycle, source checks run on the frequency you picked in Source Health Checks and after every successful generation run, and a schedule that keeps failing pauses itself instead of burning credits.
The deployed directory is a real application, not a page-builder export. The runtime clones the classic template — a Next.js 16 App Router monorepo with TypeScript, Drizzle and its own Playwright suite — fills its Git-based CMS with the items it researched, and ships it. All of this is in the repository from the first build, whether you switch it on or not; the full inventory is on the What every site includes page linked below.