The AI-PoweredCompany Builder
An open agentic runtime that autonomously researches, ships, and maintains entire businesses, 24/7.
Simple. Transparent. Real Impact.
Everything you need to know about building and managing directories with Ever Works.
View TemplateGitHub Native
Content and code both live in your own GitHub repos. Open, version-controlled, nothing locked in.
AI-Powered
Agents research your topic, write the content, generate the code, deploy, and keep the site improving on a schedule — not just the first draft.
OSS & SaaS
Run the free open-source runtime yourself, or use our managed SaaS for hosted agents, observability, and team features.
More than a builder
It builds your website — then keeps running your business
Missions
Goals
Ideas
Works
Tasks
Agents
Memory
Teams
Skills
Missions
Long-running goals that keep generating ideas — and Works — for you. Tune the cadence, the cap, and whether it builds on its own.
Goals
Measurable targets checked automatically against your metrics providers. Each one tracks a live number toward a target, so progress is observed, not claimed.
Ideas
Everything the platform has drafted for you, plus anything you've typed in yourself. Build the ones worth building; the rest just sit there.
Works
Your content hubs. Each is a live website plus the GitHub repositories behind it — organise, configure, and publish them from one place.
Tasks
Trackable work items assigned to people or Agents, so nothing in flight falls through the cracks.
Agents
Your AI teammates. Each carries a soul, its own instructions, and a budget — and acts on a Mission, a single Work, or your whole workspace.
Memory
Everything your organisation knows, in one place — aggregated across every Work's knowledge base. What one agent learns, the next one starts with.
Teams
Organise your Agents and members into teams inside the organisation, and grow the roster as the work does.
Skills
Reusable, versioned instructions Agents pull into a run when they are relevant — your house style and rules, applied without repeating yourself.
AI Agents
Named AI employees — a CEO, a CTO, a Researcher — that run your Missions, Ideas, and Works around the clock.
Agent Sessions
Run many agents at once from one cockpit — live status, current activity, tokens and cost — and steer, interrupt, or resume any of them.
Task Workspaces
Every task gets its own branch and its own checkout. Agents open pull requests, and conflicts surface before the PR — with the conflicting files named.
Live Terminals
Watch an agent work in real time. Attach, detach, reconnect — and read back the whole session afterwards.
Quality Gates
Declare the checks a change must pass. The platform runs them for real — and a red check opens no pull request.
Integrations
Drive the platform from Slack, get pull requests reviewed against your knowledge base, and turn meetings into memory.
Memory
Agents recall what your team already decided. Recall is injected into every run, decisions are recorded and reviewed, and superseded knowledge retires itself.
24/7 Autonomy
One-shot builders generate a site and stop. Ever Works keeps researching, writing, building, and improving — on a schedule.
Agent Email
Real inbound and outbound mailboxes for your Agents, Missions, Ideas, and Works — your AI workforce on email.
Knowledge Base
A per-Work, Git-backed store of brand, voice, research, and rules that every generation reads from — your built-in wiki and memory.
Templates
Every Work is generated from a template — production-grade Next.js and Astro bases — so you never start from a blank page.
Git Ownership
Code and content both live in your own Git, and the platform is open source under AGPLv3 — nothing is locked in.
How it works
Describe a goal. Get a business that keeps running.
Four nouns, one flow. A Mission is the goal you set, Ideas are the proposals it generates, Works are the live sites those become, and Agents are the AI team that runs all of it — long after the first build.
1
Mission
Hand Ever Works a standing goal instead of 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 stays inside a spend cap and a limit on how many unbuilt Ideas may be waiting. One more switch on the same Mission — auto-build, off until you turn it on — decides whether the Ideas it spawns wait for you or become Works on their own.
Learn more →2Ideas
Each tick proposes specific things worth building — a title, a description, suggested categories. Ideas also arrive from "Suggest more", from discovery, or from you typing one in. You press Build or Dismiss — or switch the Mission to auto-build and every Idea it spawns is queued into its own Work without you clicking anything. One Idea becomes one Work either way.
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. The researched content and the deployable code both land in Git repositories in your own account. One content pipeline writes for all five kinds today, so directory and awesome-list Works are the deepest of them and blog- and landing-copy writers are rolling out. A Company Work exists today too, minted by its own Register-Company flow rather than by building an Idea; a Store Work is still to come.
Learn more →4Agents
Named AI employees — a CEO, a Researcher, a Copywriter, or whatever roles you name yourself — scoped to your account, a Mission, an Idea or a single Work. Give one a heartbeat cron and it wakes on schedule to take the next useful action, bounded by its own spend cap and per-tool permissions that all start switched off.
Learn more →…and it keeps going
Nothing stops after the first build. Scheduled updates re-run a Work’s generation pipeline on the cadence you pick, Agents wake on their heartbeats, and directory and awesome-repo Works get their item source links re-validated and incoming community pull requests triaged into your data. Daily and weekly digests tell you what happened while you were away — and budgets, pause and resume, and auto-pause on repeated failure are the brakes.
Scheduled updatesAgent heartbeatsMission auto-buildSource validationCommunity pull requestsDaily & weekly digests
See what runs 24/7 →The deliverable
What you get with every Work
A Work is not a preview or a page export. It is repositories you own, a deployed application, the knowledge behind it, and the record of every change made to it.
Three repositories in your Git
Every Work is provisioned as repositories on your own Git account: one for the structured data, one for the Work’s Markdown, and one for the website code. Campaign Works skip the website repo — they organize work rather than publish a site — and Company Works, which are still coming, will do the same.
A production app, not an export
For a directory or awesome-repo Work the website repo is a real Next.js 16 App Router application cloned from a maintained template: sign-in, search, SEO metadata, sitemap and robots routes, RSS, Atom and JSON feeds, and 21 languages already wired up. Those Works can pick a lighter static Astro base instead; website, landing-page and blog Works are generated from the general web template, and their kind-specific content writers are still rolling out.
A deploy target you choose
Publish to Ever Works managed hosting — a subdomain, its DNS record and a PostgreSQL database provisioned for that Work, on a shared cluster still being brought up environment by environment — or to Vercel with your own API token, or to your own Kubernetes cluster from a pasted kubeconfig. Add a custom domain on the Deploy tab and press Verify DNS to promote it.
A Knowledge Base and Memory
Each Work carries a typed, Git-backed Knowledge Base — brand, legal, SEO, glossary, personas, research — that Agents read on every run and write findings back into. Uploads are extracted and embedded for cited search, and organization-wide Memory consolidates what is learned across Works.
A schedule and the Agents to run it
Pick one of seven cadences, from hourly to monthly, and hit Run now whenever you do not want to wait; after three consecutive failures by default — you can set anywhere from 1 to 10 — a schedule pauses itself instead of burning budget. Agents wake on their own cron heartbeats and choose the most useful next action when nothing is assigned.
An audit trail for everything
Every generation, deployment, import, plugin change and sign-in is a row in the Activity log you can filter by type, status or text and export to CSV. Every content and code change is also a commit in your repository — so you can diff it, review it, or revert it.
Control & safety
Bounded autonomy, 24/7
Nothing here waits for you to type. Agents wake on their own heartbeat cron and take the single most useful next action, scheduled updates re-run a Work’s generation pipeline on the cadence you pick, and a Mission on a schedule keeps ticking out fresh Ideas as the world changes — and once that Mission’s auto-build is switched on, those Ideas are queued straight into Works of their own, inside the same spend cap, where a cap already reached skips the build rather than saving it for later. The work continues overnight, at weekends, and while you are in meetings.
And none of it runs unbounded. Every AI call is billed against a budget you set and refused before it runs when the cap is hard. Every side-effectful action either clears a guardrail you wrote or queues for a human. Every run is recorded as a session with its tokens, cost and changed files — and every change an agent ships is a commit in a Git repository you own, so reviewing or reverting a night of work is always an option.
Any Agent, Mission or schedule can be paused instantly, and resumed when you are ready.
- ✓Budgets per Agent, Work and MissionCap spend on your account, a Mission, an Idea and a Work — and give each Agent its own cap for the hour, day, week or month. A hard cap refuses the next AI call before it runs; a soft one alerts you and lets the work carry on.
- ✓Approval queue and InboxSide-effectful actions queue as proposals carrying their risk flags — destructive, cross-scope, high fan-out, budget override — and you approve or reject per row. Questions and escalations land in the same Inbox, and on to Slack, Discord, Telegram, WhatsApp or Novu once you connect a channel.
- ✓Quality gates and merge policyTurn the gate on and build, test, lint, type check or a command of your own must pass before a pull request is opened — set it to required and a red check opens nothing, handing the failure back to the agent for a set number of attempts. Merge policy then decides who may land it, resolved at tenant, organization, Work or Agent level.
- ✓Auto-pause on repeated failureAn Agent that fails three runs in a row pauses itself instead of retrying forever — adjustable from one to twenty. A run caught cycling on the same failure is stopped and escalated with the evidence rather than paying for another identical attempt.
- ✓Live terminals and sessionsEvery run is a session showing status, current activity, tokens, cost, duration and its gate verdict, refreshed every ten seconds while anything is running. Attach a terminal to a live run, steer it mid-flight, interrupt it without losing committed work, or resume it later.
- ✓Daily and weekly digestsA short briefing of what shipped, what moved to review and what still needs a decision from you — every number counted from the actual run history, on the cadence you pick. A night of autonomous work becomes something you read in a minute.
- ✓Autonomy is a switch you throwA brand-new account has the Work agent switched off, and that one setting gates both the Build button and a Mission’s auto-build — while it is off, a Mission tick queues Ideas that never become Works, and the API refuses the build outright. The first deploy of a Work is a human action too: nothing publishes a freshly generated Work by itself, and scheduled updates only keep it fresh after that one click.
Every surface
Drive it from wherever you work
The dashboard is one way in, not the only one. Talk to the platform in chat, from a Slack channel, from your terminal, from your own agent client, or straight over the API — each one acting as you, against the same platform. How far each surface reaches differs, and every card below says where its limit is.
The dashboard
Sign in and you land on a cockpit rather than a launcher: stat tiles, the actions your Agents are asking you to approve, anything errored, blocked or over budget, and the next three scheduled runs — then your Missions, Ideas, Works, Tasks and Agents. The blocks with nothing to report stay hidden instead of showing empty shells.
Take the tour →Platform chat
Every dashboard page carries a chat rail that does what the buttons on that page do — create a Mission, pause an Agent, invite an editor, enable a connector, chart last month’s spend. It runs roughly 400 tools as you, one entity at a time, and asks before anything irreversible; charts, tables and boards open in a side Canvas instead of a wall of text.
See what chat can do →Slack
Mention @works in a channel and your message is routed to the same platform chat the dashboard uses, with the answer posted back into the thread — grounded in the same Work Knowledge Base. The /works slash command takes the same path: a private acknowledgement inside Slack’s three-second window, then the reply in the channel. Slack answers questions rather than operating the platform: it does not run the dashboard chat’s confirmation-gated tools, so “delete the staging webhook” from a channel is a question, not an action. Every delivery is signature-verified, and anything unverified is refused rather than guessed at.
See the integrations →MCP server
Point any MCP-compatible client at Ever Works and it gets 74 tools from four sources: 66 verbs derived from the API’s own OpenAPI spec — Works, Missions, Ideas, generation, deployment, plugins, schedules, usage — six Knowledge Base tools that accept a document path like brand/voice, register_work, and a ping health check that proves the client is connected without touching the API. Run it locally over stdio with your own API key, or connect to the hosted endpoint at mcp.ever.works, which signs you in as yourself rather than with a shared key.
Connect a client →The CLI
Install ever-works-cli from npm and the ever-works binary gives you four command groups: auth, work, plugins and kb. Create a Work, generate its content, poll its status and deploy it without leaving the terminal — or upload a brand-voice document straight into a Work’s Knowledge Base and lock it so no agent run can rewrite it. Credentials are stored 0600 in your home directory, and login, work register and the shared HTTP client all refuse to attach your token when the URL is plain http and the host is not loopback.
Read the commands →REST API & webhooks
Everything the dashboard does is a REST endpoint. Mint an API key under Settings → API Keys — up to ten, shown once, sent as an x-api-key or bearer header — and script the platform from CI. Signed inbound triggers let an outside system start platform work over HMAC, and webhook subscriptions push platform events back out with signed, retryable deliveries.
Read the API surface →Self-host it, or run the work on your own machines
Ever Works is open source, so the whole platform can run on your own infrastructure from the official container images or a Kubernetes manifest. The work itself does not have to run in our cluster either: enroll a laptop, a server, or the nodes of a cluster you already operate as execution nodes, and your Agents run there — over an outbound-only channel, with no port to open.
Counted, not claimed
Every figure below is counted from the open-source repository this product is built in — the plugin packages each install carries, the capability categories they cover, the tools the MCP server exposes, and the locales a generated site is built in.
102
Plugins in every install
19
Capability categories
74
MCP tools for your assistant
21
Languages a generated site ships in