obsidian-wiki is a framework for AI agents to build and maintain a persistent knowledge base as an interconnected markdown vault in Obsidian.
The problem it solves is knowledge loss: solutions discovered in one project disappear into chat logs and are rediscovered from scratch months later. Rather than repeatedly querying an LLM or running RAG pipelines for the same questions, the tool compiles learned knowledge once into interconnected markdown files that stay current. An agent ingests text from any source—docs, PDFs, transcripts, URLs—and the tool structures it as wikilinked markdown that any agent can read and execute. The approach treats the vault as a growing second brain that agents maintain through discrete skills, each implemented as a markdown file.
Anyone adopting this should know it works with multiple AI agents including Claude Code, Cursor, Codex, Windsurf, and Gemini CLI, with no runtime, API keys, or vendor lock-in. It suits projects where knowledge accumulates over time and needs to be discoverable and reusable across different contexts. The tool includes skills for feeding the vault with new information, querying it with citations, and maintaining it as the vault grows. Installation is straightforward via standard Python package managers. The vault can be visualized in Obsidian's graph view with node coloring by tag or category, and the knowledge graph can be exported to multiple formats including interactive HTML, GraphML for Gephi, Neo4j Cypher, and Postgres SQL.
Development shows consistent engagement with pull requests welcomed and active community channels. The project maintains a documented list of compatible agents and a comprehensive skills reference, indicating ongoing curation of the tool's capabilities. Documentation covers installation paths, agent compatibility, and skill usage, suggesting the maintainer is responsive to helping users get started and understand what the tool can do.
