A 24/7 research agent that deposits structured, actionable notes directly into my vault — and why the files matter more than the tools.
Every week I have the same experience. I ask an AI about something I'm researching, get a genuinely useful answer, think I should save this, and don't. Three weeks later I'm starting from scratch, because the conversation is buried in history, disconnected from everything else I know.
That's the dirty secret of AI research: brilliant at generating insight, terrible at preserving it. And even if you did save the conversation — copy-pasted it into Notion, bookmarked it — it would already be going stale. The AI landscape moves fast enough that best practices from six months ago are often wrong.
What you need isn't just a way to save research. You need a way to keep it alive.
The Philosophy First
Obsidian CEO Steph Ango (@kepano) wrote an essay called File over app with a deceptively simple argument: the apps you use to create things will eventually die, pivot, or lock you out. Plain text files, stored locally, owned by you, have a chance to last forever.
"Apps are ephemeral, but your files have a chance to last."
Apply this to AI outputs and the implication is uncomfortable. Every insight you generate in ChatGPT lives in ChatGPT's infrastructure. Every Claude research session exists only as long as Anthropic maintains that history. When the tool changes — and it will — your accumulated research evaporates with it.
What if every AI research output was automatically saved as a plain .md file in a local vault you control? That's what this stack delivers.
The Three Tools
OpenClaw is an open-source personal AI assistant with 215K GitHub stars — the fastest-growing OSS project in history. It runs on a server (I use an AWS EC2 instance) and connects to Telegram, WhatsApp, or wherever you prefer to work. Think of it as the always-on brain of the operation: it receives your messages, decides which tools to invoke, and orchestrates the work while you get on with your day.
The last30days (@slashlast30days) skill is a Claude Code skill by @mvanhorn (2,900+ GitHub stars) that does something specific and genuinely useful: given any topic, it sweeps Reddit, X, YouTube, and the broader web for the freshest signal from the past thirty days, then synthesizes the findings into a structured, actionable report. When you ask OpenClaw to research a topic using last30days, it knows to open Claude Code, run it with the last30days skill loaded, and hand you back a clean synthesis — not a raw dump of links, but a structured breakdown of what's trending, which tools matter, and what you should pay attention to. The skill itself just produces output; where that output goes is up to you. I've configured OpenClaw with a standing instruction: whenever I say "save to Obsidian," it takes the results and writes them as a .md file into a dedicated openclaw-notes folder, named with a timestamp and a descriptive title — something like 20260220-obsidian-openclaw-agentic-pkm-research.md. That naming convention matters: it makes notes sortable by date, grep-able by topic, and immediately readable by any other tool in the chain. The reason this skill exists at all captures something true about this moment: the AI landscape moves fast enough that the best practices and tools from six months ago are often already obsolete. You don't need a static knowledge base — you need continuous, fresh signal.
Obsidian (@obsdmd): is where all of that research lands. If you haven't used it, here's the quick version: Obsidian is a local-first note-taking app that stores everything as plain Markdown files on your machine. It built its reputation with the productivity and PKM (Personal Knowledge Management) crowd — people who take knowledge organisation seriously and want full control over their notes. But it's having a second renaissance right now, because it turns out that the same qualities that made it beloved by productivity nerds — plain text, local files, no cloud lock-in — also make it the ideal destination for AI-generated content. Every note your AI produces can live alongside your own thinking, link to other notes via `[[wikilinks]]`, and be read by any other AI tool in your workflow. Obsidian is less a notes app and more a vault: a durable, queryable, growing record of everything you know.
The Workflow
I run OpenClaw 24/7 on an AWS EC2 instance, with Telegram as the interface — always on, always available, no need to keep a local machine awake. A typical research session starts on my phone. I text my assistant — I call it Mouja:
"Research Obsidian + OpenClaw + agentic PKM workflows"
That's the whole input. I put my phone away. OpenClaw picks it up, runs /last30days asynchronously, and I don't think about it again. The skill sweeps the last thirty days of signal on that topic, filters the noise, and hands the synthesis to the AI. What comes back isn't a brain dump — it's a structured note: what's trending and why, key tools with GitHub links, emerging patterns, and specific action items tied to my projects.
When I'm ready to keep it, I just say: "save to Obsidian." OpenClaw takes the output, writes it as a .md file into my openclaw-notes folder with a timestamped, descriptive filename, commits it, and pushes it to GitHub. By the time I open Obsidian on my Mac, the note is already there, named, dated, and ready to link.
The sync setup is simple but worth explaining. GitHub sits in the middle as the source of truth: both the EC2 instance and my Mac push and pull from the same private repo, so whichever machine writes a note first, the other picks it up automatically. No manual file management, no conflicts. From the Mac to my phone, I use Obsidian Sync — it's the one paid piece of the stack, but it handles mobile sync seamlessly with no extra configuration.
Plain text. Owned by me. Searchable. Linkable. Forever.
Why This Compounds
Most AI workflows are one-shot transactions. You get an answer, the exchange is over, and the output has no relationship to anything else you know.
This one is structurally different. Every research session deposits a new, self-contained note into your vault. And because .md is the native language of modern AI tools — it's the format Claude Code, OpenClaw, and virtually every agent reads and writes natively — those notes don't just sit there waiting for a human to read them. They become inputs to other workflows.
Ask OpenClaw to research a topic and save it to your vault, then ask Claude Code to read that file and generate a landing page prototype, a feature spec, or a competitive analysis. The research flows directly into the build. No copy-paste, no context-switching — just a file that one tool wrote and another tool reads.
The Obsidian [[wikilink]] system takes this further over time: notes can link to previous research, project fragments, and idea threads, so the vault gradually builds a topological map of your knowledge — which topics connect, where the gaps are, which ideas are stable versus which are being actively disrupted.
Six months from now you'll have a searchable, interconnected map of everything you've ever researched. The AI was the instrument. The vault is the asset.
A Note on the Tools Changing
The AI model powering this pipeline will change. It's Claude today. It might be something better tomorrow, from a completely different lab. OpenClaw itself recently transitioned to an open-source foundation after its creator joined OpenAI.
None of that affects the vault.
The .md files created by last30days will be readable in fifty years on hardware that doesn't exist yet, because plain text is the most durable format humans have ever produced. And right now, in 2026, it also happens to be the format every serious AI tool prefers to work with. That's not a coincidence — it's why this entire stack converges on Markdown. The tools will change. The vault grows.
That's the deepest version of what this workflow is: a system for turning ephemeral AI-generated insight into durable, owned, compounding knowledge.
The Full Stack
OpenClaw → github.com/openclaw/openclaw — 215K ⭐, AWS EC2, Telegram interface
@slashlast30days skill by @mvanhorn → 2,900+ ⭐
Obsidian (@obsdmd) → obsidian.md — local-first Markdown vault, free for personal use
GitHub → syncs EC2 ↔ Mac via git push/pull
Obsidian Sync → syncs Mac ↔ phone
What does your research workflow look like? Especially curious if anyone has pushed this further with scheduled watchlists or automated daily briefings.