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.
Resources
Every guide here walks the real product: the screens, buttons and routes you will actually see, with an honest note wherever a part is still rolling out. Start with the pages on this site, then pick the guide for what you are building — the full set is published in the Ever Works documentation at docs.ever.works, under Guides, by the titles below.
Four pages on this site, in the order most people need them.
The three ways in — sign up, type a prompt as a guest, or self-host with Docker — the ten-step setup wizard, and the four moves that take a first Work from generated to live and self-maintaining. The docs’ Getting Started page adds the local path: a clone of the open-source monorepo and the plugin keys it needs.
Every sidebar entry from Dashboard to Settings, what lives inside a Work and inside an Agent, the settings pages you will actually open, and the keyboard shortcuts. The docs carry a longer, screen-by-screen version.
The model behind the product — Missions, Ideas, Works, Tasks, Agents and Teams, Knowledge and Memory — and the path from one typed sentence to a deployed site that a schedule keeps current.
Start, Build, Sell, Scale: one playbook in which every step maps to something the platform does for you — Missions and Ideas to plan, Works and templates to build, Agents and autonomous operation to run it.
One quickstart per Work kind, each taking a single Work from an empty prompt to a deployed site and saying plainly what that kind does not do yet.
A multi-page site as a Website Work: the chip, the template, seeding the Knowledge Base, a Work-scoped Agent, quality gates, the Pull requests tab, the deploy and a custom domain.
Create a Blog Work, seed its Memory with your brand voice, generate, set a refresh cadence, deploy it on your own domain and hand it to a Work-scoped Agent — with an honest account of what the blog kind does not do yet.
Pick a template and restyle it with AI, deploy it to a live address, then give the page a Knowledge Base and its own Agent. The guide says up front that the platform does not write landing-page copy yet, and shows where the copy lives.
The most complete Work kind, from an empty prompt to a scheduled, deployed site: generation, items, comparisons, the deploy, the schedule and community contributions.
From a prompt, or from a README that already exists, to a generated index, structured data you own, community pull-request intake and a weekly source sync.
Three guides that sit beside every quickstart. The templates guide puts every template system side by side — fork, add by URL, customize with AI, sync from base. Import an Existing Repository brings a repo you already own under management. Custom Domains and Deploy Targets covers the managed ever.works address, Vercel and your own Kubernetes cluster.
For when one Work is running and you want the platform to keep going without you driving every step.
Stand up a standing AI organization — a Mission and its Works, Goals wired to real metrics, Agents hired from the catalog, Teams, heartbeats, budgets and guardrails — then operate it from the Inbox, the approval queue and the digest.
Wire a template, a Mission with auto-build, the Ideas queue, budgets and schedules into one standing loop, then watch it from Sessions, Schedules and the Inbox instead of driving it by hand.
Every ceiling the platform gives you, money limits and action limits side by side — spend caps from account to Agent, the dispatch gate, approvals, tool grants, merge policy and quality gates — ending in a safe-defaults checklist for a new team.
Create an Organization, invite people, build nested Teams with a manager Agent, read the Org Chart, import a prebuilt company, and hand out the per-Work roles that actually restrict access.
Seed a Work’s Knowledge Base from real files, review what the Agents wrote, ask it questions and get cited answers, and keep the organization-wide Memory and the Agent definition files honest.
For developers and operators: the other ways to drive the platform, and the ways to run it yourself.
Connect Claude Desktop, Claude Code or any MCP-compatible client: create an API key, run the stdio or streamable-HTTP transport, and drive Works, Missions, Ideas and the Knowledge Base as tools.
Install the ever-works-cli package, sign in, and drive Works, plugins and the Knowledge Base from your terminal — with an honest map of what the CLI does today and what still belongs to the API or chat.
Ten worked prompts that drive the real platform from the dashboard chat rail, each mapped to the tool it calls — plus the confirmation cards, the canvas, History, Slack and the OpenAI-compatible endpoint.
Point Ever Works at your own model API — a provider plugin, a local Ollama, LM Studio or vLLM server, or a gateway — set the model tiers, override it per Work, and see what your own keys do to your credits.
The setup path for every integration: the Slack app and its slash command, the GitHub App versus the pull-request review webhook, the connector plugins, and notification channels with a test send.
Self-host with Docker Compose and Kubernetes covers the layered Compose files, the published images and the Kubernetes manifests — there is no Helm chart. The Desktop App guide runs the whole platform locally or as a client of another instance; it is built in CI, with no public installer yet.