How Ever Works Works
Building and managing your directory is easy with Ever Works. Follow these simple steps to create a beautiful, functional directory powered by modern tools.
1. Choose Your Template
Start with our free Next.js template or select a premium theme that matches your vision.
2. Connect with GitHub
Link your GitHub account to store your directory data in a new or existing repository.
3. Configure Your Directory
Set up your directory structure, categories, and customize the appearance.
4. Add Content
Add items manually or use our AI tools to generate and enrich content automatically.
5. Deploy & Share
Deploy to Vercel with one click and share your directory with the world.
Key Features
GitHub-Native Storage
All your directory data is stored in your GitHub repository, giving you full control and version history.
AI-Powered Tools
Use AI to generate descriptions, categorize items, and enrich your directory content.
Automatic Updates
Your directory stays up-to-date with automated content updates and synchronization.
Analytics & Insights
Track visitor engagement, popular items, and directory performance.
Ready to Get Started?
Create your first directory in minutes. No credit card required.
How Ever Works actually works
Ever Works is an open agentic runtime: you set a Mission, it proposes Ideas, builds the ones you pick into Works — real websites and Git repositories — and an AI workforce of Agents keeps researching, shipping and maintaining them for as long as you let it.
1
Mission
A Mission is a standing goal rather than a one-off brief — "run the best cats business worldwide", "build out our presence online". It runs once by default, or on a cron you set, and every tick is bounded by a spend cap and a limit on how many unbuilt Ideas may be outstanding.
Learn more →2Ideas
Each tick proposes Ideas: a title, a description and suggested categories for something worth building. Ideas also arrive from "Suggest more", from discovery feeds, from a scheduled refresh, or from you typing one yourself. You hit Build or Dismiss.
Learn more →3Works
Building an Idea produces a Work — a website, landing page, blog, directory or awesome-list repo, generated from a production template rather than an empty scaffold. Each Work lands as three Git repositories in your own account: your content data, the readable list, and the deployable site. Store and Company are coming.
Learn more →4Tasks
Concrete work is tracked as Tasks — T-1, T-2, … — on a kanban board with status, priority and labels. Turn on task isolation and every agent-run Task gets its own branch and a private checkout, has to pass its acceptance checks, and comes back as a pull request instead of a direct commit.
Learn more →5Agents & Teams
Agents are named AI employees — a CEO, a Researcher, a PR Reviewer — each scoped to your account, a Mission, an Idea or one Work. Give one a heartbeat cron and it wakes on schedule to pick its next action; every Agent carries its own spend cap and per-tool permissions that all start switched off. Group them into nested Teams and read the structure off the Org Chart.
Learn more →6Knowledge & Memory
Every Work has a typed Knowledge Base — brand, legal, SEO, glossary, personas, style, research — stored in that Work’s Git repository and read by every run, so the tenth generation knows what the first one learned. Org-wide Memory adds shared documents, a review queue for agent-written notes, and first-class decision records.
Learn more →7It keeps running
Nothing stops after the first build. Scheduled updates re-run the pipeline on your cadence, Missions tick, agents heartbeat, item source URLs get re-validated, community pull requests are triaged into your data, and the template layer under your site is kept current. Budgets, pause and resume, and auto-pause on repeated failure are the brakes.
Learn more →8You stay informed
When an agent needs a decision only you can make, it writes to your Inbox — a blocking question, an approval request, an escalation or a notice — and your reply unparks the run waiting on it. Daily and weekly digests summarise runs, tasks, pull requests and goals, and the Activity log keeps every generation, deployment and schedule run on the record.
Learn more →What happens after you type your idea
Both halves are generated, and both are committed to repositories you own — the content into your data repository, the code into a website repository cloned from a maintained template rather than assembled out of blocks. Of the nine steps below, exactly one needs your hands: number seven.
- 1
You type a sentence
One line of intent is the whole input. Send it from this site and you land on onboarding with a temporary guest session already minted, so the first Work can be generated before you have created anything; claim it later with an email and a password and it upgrades the same account rather than starting a new one.
Learn more → - 2
Research runs first
Before a word is written, agents read the topic: they gather and score sources, then produce a brief per page the Work will have — angle, audience, sources to cite, points to cover. For directory and awesome-repo Works the same pass produces the entity map: the actual catalogue of items the Work will list, with tags and short descriptions.
Learn more → - 3
A template is cloned into your Git
A Work does not start from an empty scaffold. A maintained base template is cloned into a website repository in your own account and kept in sync with upstream afterwards. Directory and awesome-repo Works default to the classic Next.js base; website, landing-page and blog Works default to the general-purpose web base. Pin a different one, fork one, or bring your own repository.
Learn more → - 4
Content is written into your data repo
The writing pass fills a second repository you own: items and their markdown bodies, categories, tags, collections, cited references and head-to-head comparisons. One pipeline serves all five Work kinds today, which is why directory and awesome-repo Works are the deepest of them — dedicated blog-post and landing-copy writers are rolling out.
Learn more → - 5
The site is generated
What lands in the website repository is a real application, not an export: sign-in and OAuth, a database layer, payments, search, an admin panel, 21 languages, SEO metadata, a sitemap and feeds. All of it is in the repository on day one, whether you switch each part on or not.
Learn more → - 6
Quality gates, then a pull request
Where a Work sets its gate to required, the declared checks — build, test, lint, type-check — have to come back green. A red gate withholds the pull request and hands the failing check, its exit code and its log tail back to the agent to fix, inside an attempt budget. Merge policy decides who lands the pull request once it is open.
Learn more → - 7
You deploy it — once
This is the step that needs your hands. No code path publishes a freshly generated Work by itself: the first deploy is a human action, from the Work’s Deploy tab or the CLI, to Vercel, Docker, Kubernetes or your own cloud. Everything after it can run without you.
Learn more → - 8
A schedule takes over
Pick one of seven cadences, from hourly to monthly, and the pipeline re-runs itself on that clock — re-researching, rewriting stale pages, re-validating item URLs, regenerating and redeploying. A failed run is retried, and repeated failures auto-pause the schedule and notify you instead of burning budget.
Learn more → - 9
And it proposes the next one
A Mission left on a cron keeps ticking out new Ideas, and with auto-build switched on those become Works without a click — bounded by the spend cap and the limit on how many unbuilt Ideas may be outstanding. Two switches have to be on first: the account-level Work agent gate, which is off on every new account, and the Mission’s own auto-build toggle. The machine that built one Work is the machine that proposes the next.
Learn more →
Three ways to drive it
The dashboard is one client, not the only one. The same operations are available from chat, from your terminal, from an AI assistant, and from a copy of the platform you run yourself.
Dashboard & chat
Click through the dashboard, or say it to the chat rail that sits on every page: every platform operation is also a chat tool. Anything destructive stops on a Confirm card before it runs, and answers render on a canvas — charts, tables, kanban — instead of a wall of prose.
Learn more →Slack, MCP, CLI, API
Drive the same operations from outside the browser. The MCP server exposes 66 whitelisted API operations plus Knowledge Base tools to any MCP-compatible assistant, the CLI runs Works, plugins and the KB from a terminal, the REST API takes an API key, and Slack carries notifications and approvals back to your team.
Learn more →Self-host or desktop
Ever Works is open source under AGPLv3 and runs where you want it: our cloud, your own Docker Compose stack, your Kubernetes cluster, or the desktop app (early access) running the whole platform on your machine. Either way, the code and content stay in your Git.
Learn more →