Local-first
Your files stay on your computer. CodexVault is designed around local folders and local AI.
Private notes. Local AI. Source citations.
CodexVault turns folders of notes and documents into a searchable local AI knowledge base. It converts supported files into clean Markdown, indexes them with SQLite, and answers questions with source citations through Ollama.
Why CodexVault
CodexVault is built around ordinary files, local search, and explicit approval before writes.
Your files stay on your computer. CodexVault is designed around local folders and local AI.
Output stays portable. Use it with Obsidian, another editor, or a normal file browser.
Answers include source file references so you can inspect where the information came from.
CodexVault should never rewrite vault files automatically. You approve before writes happen.
Workflow
Start with a test folder, build Markdown output, search locally, then ask questions against the index.
Point CodexVault at a folder containing .md, .txt, .pdf, or .docx.
Supported files are converted into clean Markdown for a portable local vault.
SQLite FTS5 builds a rebuildable local search index from the Markdown output.
Ollama answers questions using retrieved local context and source citations.
Review proposed edits before CodexVault writes anything to disk.
Release status
CodexVault currently runs as a local Python, FastAPI, React, SQLite, and Ollama app. It is being hardened before broad public promotion.
.md, .txt, selectable-text .pdf, and .docx.
Windows-first. Mac and Linux can be explored later after the workflow is stable.
No installer yet. Setup currently uses Git, Python, Node.js, and Ollama.
SQLite search first. Semantic search can wait until users report search quality issues.