Files.md: Open-Source Local-First Markdown Note-Taking App with LLM-Friendly Design

Files.md is an open-source alternative to Obsidian that treats every piece of your life as plain .md files. The project has been in development for five years and currently has 886 GitHub stars. It is designed to be private, local-first, and LLM-friendly — the entire codebase is simple enough for one person or an LLM to hold in their head.
Key Features
- Local-first by default: Your data never leaves your device unless you choose to sync. No telemetry or server contact.
- Multiple sync options: iCloud, Dropbox, Google Drive (cloud-folder sync), self-hosted Go binary, or hosted beta at app.files.md.
- Works offline: Open
web/index.htmldirectly — no build system needed. Portable, just a browser. - Telegram chatbot: for on-the-go quick entry of notes, tasks, journal entries, and checklists.
- LLM-optimized: The code is simple and modular, making it easy for AI agents to extend the app to custom needs.
- One idea per note: Encourages atomic notes (one concept per file), linking via
[, and revisiting to connect ideas.
Quick Start
To try it without any setup, open app.files.md in Chrome, click Install files.md in the address bar, and open a local folder. Changes persist to disk. Force refresh (Cmd+Shift+R) occasionally for updates.
To sync across devices without a server, place your markdown folder in iCloud/Dropbox/Google Drive. For full control, run the self-hosted server (single Go binary) and optionally enable the Telegram bot.
Philosophy
Files.md intentionally omits complex plugins, graph views, and advanced templates. It restricts features to force deep thinking: start with zero folders, one note per idea, link related notes, and apply knowledge immediately. The project is built around the idea that software should stay out of the way and that LLMs can help grow the software as you grow your knowledge.
📖 Read the full source: HN LLM Tools
👀 See Also

Claude Code Used to Simulate 4,000+ Blind Werewolf Games with LLMs
A developer used Claude Code to build a simulator where LLMs play blind one-night Werewolf, running ~4,600 games across OpenAI and xAI models. The experiment revealed consistent name-based voting patterns despite minimal game signals.

Claude Code skill combines DeepMind Aletheia and Anthropic harness approaches
A Claude Code skill implements a Planner→Generator→Evaluator→Reviser pipeline that synthesizes DeepMind's Aletheia math research agent with Anthropic's multi-agent coding architecture, adding blind pre-analysis where the evaluator reasons about correct approaches before seeing candidate code.

Understudy: A Teachable Desktop Agent That Learns Tasks by Demonstration
Understudy is a local-first desktop agent runtime that can operate GUI apps, browsers, shell tools, files, and messaging in one session. You demonstrate a task once, it records screen video and semantic events, extracts intent rather than coordinates, and turns it into a reusable skill.

Open-Foundry: A Framework for Multi-Agent Debates with Claude Code
Open-foundry is a Python framework that assembles multiple Claude Code agents into a panel to debate complex questions, producing fully inspectable reasoning trails with transcripts, orchestrator logs, and per-agent working notes.