sage-wiki

sage-wiki turns documents into linked articles and a knowledge graph that both people and AI agents can query. It's for personal research collections, shared team knowledge, and agents that need a lasting memory of what they've learned. The software is open source under the MIT license and runs as a Go binary or in Docker on your own machine or server. It supports local models through Ollama.

The wiki stays readable as Obsidian-compatible Markdown, with a browser interface and terminal dashboard over the same collection. It can work with an existing Obsidian vault. Sources can include PDFs, Office documents, email, code, and transcripts; image descriptions require a vision model such as Gemini, Claude, or GPT-4o. Sources, articles, and indexes live in your workspace, while using a cloud model sends content to that provider for processing.

Search combines text matches, semantic similarity, and connections between concepts. Cited Q&A handles questions about the collection, while graph queries follow relationships across multiple linked facts. Optional graph extraction attaches supporting text, source documents, and confidence to relations. Alias resolution connects names such as K8s and Kubernetes, and historical queries can retrieve what the graph recorded at an earlier point.

MCP gives agents access to search, graph queries, and knowledge capture, including writing insights back into the wiki. Generated skills support assistants such as Claude Code, Cursor, and Codex. Teams can share a wiki through Git or a self-hosted server, search across multiple wikis, and use PostgreSQL with pgvector for storage.