Bybit Open-Sources KaaS: LLM-Powered Knowledge Wiki Compiler
Key Takeaways
- ▸Bybit open-sourced KaaS, a tool that compiles unstructured knowledge (documents, emails, transcripts) into searchable, structured wiki articles using LLMs
- ▸Uses a 4-phase pipeline (Extract → Classify → Write → Index) to generate human-readable, git-manageable Markdown instead of opaque vector stores
- ▸Supports any OpenAI-compatible LLM API and offers multiple deployment options: Docker, CLI, or MCP integration with coding agents
Summary
Bybit has open-sourced KaaS (Knowledge as a Service), a tool designed to solve a common organizational problem: knowledge loss when employees leave. Originally built as an internal solution, KaaS transforms scattered documents, notes, emails, and transcripts into searchable, role-based wiki articles using a 4-phase LLM compilation pipeline.
The tool processes raw content through four stages: extraction (identifying concepts and entities), classification (organizing into topics), writing (generating structured Markdown), and indexing (enabling search). Unlike typical RAG systems that store opaque vector embeddings, KaaS produces human-readable, git-manageable Markdown files that teams can read, edit, and version control.
KaaS supports any OpenAI-compatible API (OpenAI, DeepSeek, Ollama, Azure OpenAI, etc.) and offers flexible deployment: Docker containers for full backend setup, CLI installation on Linux/macOS, or direct integration with AI agents like Claude Code through Model Context Protocol (MCP). This makes it accessible whether deployed as a standalone service or embedded into existing AI workflows.
- Solves organizational knowledge loss by creating role-based knowledge repositories that persist after employee transitions
