mohammadmaso/kherad

Kherad is a self-hosted wiki tool for teams, built so that writers get a Notion-like editing experience while every change is quietly stored as a real git commit underneath. Authors never need to know what a branch or merge is: each person edits on their own long-lived branch automatically, autosave keeps a draft safe in case of a crash, and clicking Save is what actually creates a commit. When a page is ready, the author submits it for review, which opens a merge request that a reviewer can approve, comment on, or reject, with an editor for resolving the rare conflict by hand. Pages are organized into bundles, which group content by team, product, or department, and each bundle can have its own admins, reviewers, authors, and viewers, with public bundles readable by anyone without logging in. The editor itself supports headings, nested lists, tables, code blocks with syntax highlighting, and Mermaid diagrams with a live preview, and everything written round trips back to plain markdown. Several AI features are built in. A specialist agent can be assigned a company role, ask the user a few questions, and produce a wiki-ready draft the author can import directly. A chat panel lets anyone ask questions about the knowledge base and get answers pulled from the reviewed content, with sources cited rather than a generic answer made up from nothing. There is also document and voice ingest, so a PDF, Office file, HTML page, scanned image, or voice recording can be converted into an editable markdown draft, and a bundle can be pointed at a folder of raw source documents to have an AI pipeline compile them into a structured wiki automatically. A link graph shows how pages reference each other, and the tool supports right-to-left languages and multiple languages out of the box. Under the hood it is a TypeScript monorepo with a Fastify backend that is the only process allowed to touch git, a Next.js frontend for the editor and admin pages, a separate ingest service for document conversion, and Postgres for search, users, permissions, and everything besides the page content itself. It is meant to be run as a single Docker Compose deployment for one organization's internal docs rather than as a service for many separate customers. The project is licensed under the Apache License 2.0.