LLMBase
A personal knowledge base where an LLM restructures raw notes and sources into a linked, structured wiki of Markdown articles rather than storing every word for later semantic search.
LLMBase ingests URLs, PDFs, local files or corpus plugins into a raw folder with provenance metadata, then has an LLM compile that material into composed Markdown articles with wiki-links, backlinks and a taxonomy that emerges from the content, in English, Chinese and Japanese. A query stage runs a research loop that pulls context from the compiled articles and answers in a chosen voice, such as scholar, classical Chinese, ELI5 or a plainer register, and can file the answer back into the wiki; agents that need the original wording can call a separate tool to search the raw sources directly.
A lint pipeline detects broken links, low-quality stubs, inconsistent tags and duplicate concepts and repairs them, run on a schedule by a background worker, so duplicate concepts merge and broken links generate placeholder stubs automatically. A single registry of operations is shared by the MCP server, CLI and HTTP API, so registering a new knowledge-base operation once makes it reachable from an MCP-enabled agent immediately.
The project's own live deployments include a trilingual Buddhist knowledge base and a classical-Chinese knowledge base of Confucian, Daoist and Buddhist texts. It is written in Python and released under the MIT license.