OpenKB tutorial builds searchable AI base

A new MarkTechPost tutorial walks through building a local, searchable AI knowledge base with OpenKB, OpenRouter, and Meta’s Llama models. (marktechpost.com) The basic idea is to turn a folder of raw files into a wiki the model can navigate later, instead of making the model rediscover facts from scratch on every question. OpenKB describes that as compiling knowledge into linked Markdown pages. (github.com) (pageindex.ai) In the tutorial, MarkTechPost installs `openkb`, captures an OpenRouter API key with Python’s `getpass`, and sets the model to `openrouter/meta-llama/llama-3.3-70b-instruct:free`. The article says that setup avoids hardcoding secrets while keeping the demo free to run. (marktechpost.com) From there, the workflow adds source documents, generates summaries and concept pages, inspects the wiki directory, runs queries, and saves explorations. MarkTechPost also shows programmatic analysis of cross-links and page relationships inside the resulting knowledge base. (marktechpost.com) OpenKB’s pitch is different from a standard Retrieval-Augmented Generation system, which usually pulls chunks from a search index at question time. Its repository says the knowledge base is persistent, cross-references are maintained, and contradictions can be flagged as the wiki grows. (github.com) The project also skips a vector database. OpenKB says it uses PageIndex for “vectorless” retrieval on long documents, building a tree index so the model can drill into sections of a large PDF instead of scanning the whole file at once. (github.com) (pageindex.ai) That matters for the kinds of files teams actually store: OpenKB says it can ingest PDF, Word, PowerPoint, Excel, HTML, Markdown, CSV, and plain text through Microsoft’s MarkItDown library. The repository also lists watch mode, health checks, interactive chat, and Obsidian-compatible Markdown output. (pageindex.ai) (github.com) OpenRouter’s model catalog shows it brokers multiple Meta Llama models through one API, and the MarkTechPost tutorial uses the free Llama 3.3 70B Instruct route for the demo. That makes the article read less like a research prototype and more like a reproducible developer build. (openrouter.ai) (marktechpost.com) The thread running through all of it is simple: store documents once, compile them into a browsable wiki, and ask questions against that structure later. The tutorial packages that pattern into a single end-to-end example a developer can run and inspect line by line. (marktechpost.com)