Platform Overview
The Ever Works Platform provides the backend infrastructure for building, generating, and deploying AI-powered work websites.
What Ever Works is
Ever Works is an open-source, agentic runtime that researches, ships, and maintains content-rich websites and Git repositories. It brings together two things most tools keep apart:
- The builder experience — describe what you want and get a shipped website, blog, directory, landing page (and, soon, stores) generated from a template.
- The autonomous workforce — an army of AI Agents acting as real employees that keep the work going long after the first build: writing content, finding and adding items, improving code, researching, and proposing what to do next — 24/7, on a schedule.
One-shot AI builders generate code and stop. Ever Works keeps building, maintaining, and growing — and because code and content both live in your own Git, you own everything and nothing is locked in. The platform is open source under AGPLv3.
The mental model
- A Mission is an ambitious, ongoing goal the system keeps pursuing.
- An Idea is a one-shot proposal — one Idea becomes one Work. (Unsure what to build? Capture an Idea and let it be researched first, like a lightweight plan.)
- A Work is the buildable, self-maintaining unit. The kinds you pick from the chip catalog at creation time are website, landing page, blog, directory, awesome repo and repository — the last registers an existing code repository as a Work so Tasks, Goals and fleet runs can attach to it, instead of generating a site (it is not the Repositories connection registry in settings). Two further kinds exist but are minted by dedicated flows instead of the create screen: company (the Register-Company flow) and campaign (go-to-market template activation). A store kind has not shipped yet — see Store Builder. Full matrix: Work Kinds & Capabilities.
- Agents are named AI workers (CEO, CTO, Researcher, …) that run the Missions, Ideas, and Works for you.
For the full step-by-step, see the Founder Journey guide.
How It Works
- Create a Work — A user creates a work project through the web dashboard (+ New in the sidebar →
/new, or/works/newfor the full form), the platform chat, the REST API, the CLI, or an MCP client, providing a topic and description. - AI Generation Pipeline — The platform's AI agents generate work items by researching the web, extracting relevant listings, validating sources, and organizing content into categories. The run is carried out by a pipeline plugin: the 15-step
standard-pipelineby default, or one of the coding-agent and automation pipelines you can select per Work. - Repository Management — Generated content is committed to three Git repositories per Work that the user owns:
{slug}-data(structured item data),{slug}(rendered markdown), and{slug}-website(the deployable site). They are created in your own GitHub account or organization, or in the managed Ever Works Git org if you keep the onboarding default. Thecompanyandcampaignkinds produce no site, so they get the data and work repositories only. - Website Deployment — The website repository is deployed to Vercel, to your own Kubernetes cluster, or to the managed Ever Works cluster, which allocates a
*.ever.workssubdomain and provisions a database for the Work. You choose the target in the onboarding wizard or per Work, and can attach your own domain afterwards. - Ongoing Updates — Works can be regenerated, updated on a schedule, or enriched through AI conversations — and Agents with a heartbeat keep proposing and doing the next thing without being asked.
Surfaces
Everything above is one API with several front doors. Nothing is dashboard-only: the same operations are reachable from chat, the CLI, an MCP client, and the REST API.
| Surface | What it is | Start here |
|---|---|---|
Dashboard (apps/web) | The Next.js App Router cockpit — home at /, then Works, Missions, Ideas, Tasks, Teams, Knowledge Base and Settings. | The Dashboard · Platform Tour |
| Platform chat & canvas | A chat rail on every dashboard page that calls the same API the buttons do — roughly 400 tools, acting as you, one entity at a time, with a confirmation card before anything destructive and rich answers rendered on a side Canvas. | Platform Chat · Do Everything From Chat |
REST API (apps/api) | NestJS 11 under the /api prefix, authenticated by session JWT or an ew_live_… API key. Outside production it also serves Swagger UI at /api/swagger, the Scalar reference at /api/docs, and the raw spec at /api/openapi.json. | API Reference · API Keys |
MCP server (apps/mcp) | 66 whitelisted API operations plus 6 Knowledge Base tools, register_work and ping (74 tools in all), generated from the OpenAPI spec and served over stdio or streamable-http, so Claude Desktop, Claude Code, or any MCP client can drive the platform. | MCP Server · MCP Client Setup |
CLI (ever-works-cli) | npm install -g ever-works-cli, then the ever-works command with its auth, work, plugins, and kb command groups. | CLI Overview · CLI Quickstart |
Desktop app (apps/desktop) | A native shell that either runs the whole stack on your own machine (local-stack mode) or acts as a client for an instance running elsewhere (remote-client mode). | Desktop App · Desktop guide |
Fleet nodes (apps/node, apps/desktop-node) | ever-works-node enroll, then start, pause, resume, status, capabilities, clear-quarantine, and unenroll — your own machines lease and run platform jobs through the job-runtime-node runtime. | Fleet |
The admin interface (apps/admin) and the internal NestJS CLI (apps/internal-cli) are operator-only surfaces and are not part of the product UI.
Technology Stack
| Layer | Technology | Version |
|---|---|---|
| Runtime | Node.js, TypeScript | Node >= 22, TypeScript 5.9 |
| API Framework | NestJS | 11 |
| Web Dashboard | Next.js (App Router), React, Tailwind CSS | Next 16.2, React 19.2, Tailwind 4 |
| Database ORM | TypeORM | ^0.3.31 |
| AI / LLM | Vercel AI SDK (ai, @ai-sdk/openai-compatible) | ai ^6, openai-compatible ^2 |
| Monorepo | Turborepo | 2.x |
| Package Manager | pnpm | 10.33.x (engines require >= 9.9.0) |
| Background Jobs | Six pluggable job runtimes — BullMQ, pg-boss, Temporal, Trigger.dev, Inngest, Fleet nodes | — |
| Monitoring | Sentry, PostHog | — |
| Git Operations | isomorphic-git, Octokit | — |
| Search | Nine search plugins — Tavily (default), Brave, Exa, SerpAPI, Perplexity, Bright Data, Firecrawl, Linkup, Valyu | — |
The background runtime is selected per instance under Settings → Job Runtime; see Job Runtimes.
Key Repositories
| Repository | Description |
|---|---|
ever-works | Platform monorepo — API, web dashboard, MCP server, CLI, desktop app, fleet node, AI agents, shared packages, the plugin catalog, and this documentation site (apps/docs renders docs/) |
directory-web-template | The Classic template (Next.js) — the default website repository for directory and awesome-repo Works |
directory-web-minimal-template | The Minimal template (Astro, static) — opt-in for directory Works, registered through WEBSITE_TEMPLATE_MINIMAL_REPO |
web-template | The Website template (Next.js) — the default for website, landing-page, and blog Works |
web-minimal-template | The Website (Minimal) template (Astro, static) — the opt-in general-purpose variant |
agents | Ready-made Agent definitions (CEO, CTO, and more) used as Agent templates |
ever-works/apps/docs | This documentation site, built from the platform monorepo |
The template registry lives in packages/agent/src/generators/website-generator/config/website-template.config.ts; see Website Templates for the full catalogue.
AI Providers
The platform supports 11 LLM providers, all accessed through an OpenAI-compatible interface:
- OpenAI — GPT-5.1, GPT-5-nano, GPT-4o-mini
- Google — Gemini 2.5 Flash, Gemini 2.5 Pro
- Anthropic — Claude Sonnet 4.5, Claude Haiku 4.5
- Groq — Fast inference with open-weight models (
qwen/qwen3-32bby default) - Grok (xAI) —
grok-2-latestand the rest of the xAI catalog - Mistral — Mistral Small, Medium, Large
- OpenRouter — Multi-provider gateway (400+ models)
- Ollama — Local model inference
- LM Studio — Models served by LM Studio's local server (
http://localhost:1234/v1by default) - vLLM — Your own self-hosted vLLM OpenAI-compatible server (
http://localhost:8000/v1by default) - Vercel AI Gateway — Multi-provider routing via Vercel
See AI & Generation for details, and Bring Your Own AI Provider for the setup walkthrough.
Plugin System
The platform uses a capability-driven plugin architecture. All external integrations — AI providers, search engines, deployment, screenshots, and more — are implemented as plugins. At the time of writing the repository holds 102 plugin packages under packages/plugins/, spread across 19 categories (AI providers, pipelines, search, content extractors, connectors, job runtimes, secret stores, storage, vector stores, notification channels, email providers, metrics, and more), and new plugins can be added without modifying core code. See Plugin System for details.