Overview
Ever Works is an open-source AI directory builder platform from Ever Co. It helps users turn an idea into a content-rich directory or website, then keep it updated over time.
I worked on this project while working at Ever Co. I was the lead developer for the platform, including the web app, API, plugin system, AI generation flow, and background jobs.
What It Does
Ever Works lets a user create a directory from a prompt, build one manually, or import data from an existing repository. The platform stores the data in Git, generates content with AI, and can deploy the result as a website.
The product is not only a one-time generator. It also supports scheduled updates, knowledge base memory, tasks, agents, missions, usage budgets, and collaboration.
My Role
As lead developer, I worked across the main product layers:
- Next.js dashboard screens for directories, plugins, agents, missions, tasks, settings, and onboarding
- NestJS API modules for directory management, generation, plugins, budgets, tasks, and integrations
- Plugin architecture for AI providers, search, content extraction, deployment, storage, notifications, and job runtimes
- Pipeline system for AI generation, including step execution, checkpoints, progress events, and cancellation
- Background jobs with Trigger.dev for long-running generation and scheduled updates
Plugin System
A major part of the platform is the plugin system. Ever Works uses plugins so providers are not hardcoded into the core app.
Plugins can provide AI models, search, screenshots, content extraction, deployment, Git operations, storage, notification channels, job runtimes, secret stores, vector stores, task tracking, and pipeline engines.
Each plugin declares its own metadata, settings schema, capabilities, and lifecycle hooks. The platform loads them through a registry and exposes them through facades, so the rest of the app can call a capability without knowing which provider is behind it.
Pipeline System
The generation pipeline turns a prompt into directory content.
The standard pipeline runs through steps like prompt processing, domain detection, AI item generation, web search, content retrieval, item extraction, deduplication, categories and tags, source validation, image capture, and markdown generation.
The pipeline engine supports checkpoints, retries, cancellation, progress events, and parallel steps where it makes sense. It also supports full pipeline plugins, so tools like Claude Code, Codex, Gemini, OpenCode, Make, Zapier, or SIM AI can own the whole generation flow when needed.
Product Features
Some of the main features I worked around include:
- Directory creation from AI, manual input, or import
- Git-backed data and generated website repositories
- Scheduled updates for keeping generated content fresh
- Knowledge Base and memory per directory
- Agents with instructions, budgets, schedules, tasks, and inboxes
- Missions and Ideas for long-running goals
- Task tracking with assignees, reviewers, blockers, chat, and agent dispatch
- Plugin settings, provider selection, OAuth, and device auth flows
- Usage tracking and budget controls
- MCP server and CLI access for automation
Technical Approach
The frontend is built with Next.js and React. The API is built with NestJS and TypeORM. Long-running work runs through Trigger.dev and other job runtime plugins.
The architecture uses a Turborepo monorepo with shared packages for the agent logic, plugin SDK, contracts, monitoring, tasks, and CLI utilities. This kept the API, web app, CLI, MCP server, and plugin packages working from the same shared types and business logic.
The harder part was keeping the system extensible without making every feature provider-specific. The plugin and facade layers helped keep the core product stable while still supporting many providers and runtime options.
Outcome
Ever Works became a full open-source platform for building and maintaining AI-generated directories. My work helped shape the product from a generator into a broader system with plugins, pipelines, agents, missions, tasks, budgets, and Git-owned output.
Tech Stack
- Next.js
- React
- NestJS
- TypeScript
- TypeORM
- Trigger.dev
- Turborepo
- GitHub
- Docker
- Kubernetes