Klion

Klion is my main work tool — a management system I built to organize all my freelance projects, and I use it every day. It combines a Kanban board with artificial intelligence, semantic code indexing (RAG), a knowledge base, and a deep Git integration. What started as a task board ended up becoming the tool that connects my code editors with my management workflow.

AI Integration with RAG and pgvector

The system indexes entire repositories by generating embeddings stored in PostgreSQL with the pgvector extension. When I need to understand how something works in a codebase — mine or a client's — I run a semantic search and Klion finds the relevant code without needing to remember exact files or paths. Gemini 2.0 Flash is the primary provider for its low latency; OpenAI serves as fallback. The AI also generates automatic commit messages from Git diffs, technical documentation, and per-project context summaries.

MCP Server and CLI

Klion works as an MCP server (Model Context Protocol), meaning it integrates directly with Claude, GitHub Copilot, and Cursor. From any editor I can create tasks, check the board, search the knowledge base, or index a project without leaving the chat. It also has a complete CLI: klion board to view the Kanban in the terminal, klion task create for quick task creation, and klion search for semantic search from the command line.

Jira-style Project Management

Each project has a unique code (e.g., BES) and tasks are numbered sequentially (BES-22), making references easy without relying on UUIDs. The board supports drag & drop with dnd-kit, subtasks with visual progress, time worklogs per task, and filters by status, project, and client. Everything oriented to how I actually work: multiple clients with multiple active projects simultaneously.

Semantic Knowledge Base

Beyond tasks, Klion stores architectural decisions, snippets, technical notes, and documentation in a knowledge base with semantic search. When I need to recall why a certain technology was chosen or how we solved a specific problem months ago, the embedding search finds it. This became a real advantage: I don't just manage tasks, I preserve the technical context of each project.

Stack and Architecture

Modular NestJS backend with TypeORM and PostgreSQL + pgvector. Next.js 14 with App Router, TanStack Query for server state, and TailwindCSS for the UI. The MCP server is an independent package with Commander.js for the CLI. Everything runs on Docker with Nginx as reverse proxy.