We live in a world of information abundance. Articles, tweets, podcasts, YouTube videos, newsletters, voice messages — a constant stream of ideas, perspectives, and discoveries in every format imaginable. More content than any person could ever fully consume, arriving faster than we can process it.
Something catches your eye. For a moment, it sparks your curiosity, you feel like this could matter. Most of the time, that moment is lost – you save it, and forget it. Most saved content rots in the graveyard of good intentions. We just tend to swipe — the next video, the next article, the next thread. We consume, share, react, and move on. The flood keeps coming.
And somewhere in that flow, something gets lost. The deep connection — why did this hook you, how does it relate to your life? We're very good at collecting information. We're much worse at understanding what it actually means for us. That question — what does this mean for me? — is what a spark is built around.
Spark is not just a link or note — but a unit of interest. It's something that caught your attention and might deserve more of it.
AI can already process information at a scale no human can match — reading transcripts, extracting ideas, finding connections across everything you've ever saved. But human attention is still finite. What AI can't replace is judgment: what actually matters to you, how it fits your thinking, where you want it to lead. The real opportunity is in combining both — AI handles the volume, you build the meta-understanding. That combination needs a format.
Every capture becomes a structured spark — insights extracted, connections made, context preserved. Each spark can grow into a real hobby, project, or skill — or just stay as an atomic unit of interest. You choose how much time, curiosity and effort to invest in it.
The knowledge units people collect — notes, links, articles, highlights — are scattered by default. No consistent structure, no personal context. You know you saved something useful — but by the time you look for it, the reason you saved it is gone. The meaning evaporated with the moment.
The tools we have aren't solving the right problem. Storage apps save everything but preserve no meaning. LLMs tell you what something says — not why you care. Search helps you find things, but only if you remember they exist. It just makes no sense.
What we actually need from information is simpler: understand what the core idea is, why it matters to us personally, and what we can do with it. And spark.md was designed around it – it's a format built on three questions that mirror how we naturally think about anything worth remembering:
- What is it? — the essence, the core idea, in your own words.
- Why does it click? — insights and connections to your personal context.
- So What? — applicability, actionable items and intentions.
spark.md is an open format designed around meaning behind saved content. Built to be:
- Human-readable and machine-readable — structured for both you and your agents
- Personal by design — not what the content says, but why it sparked you
- Evolvable — each spark is a seed; it can grow into a project, habit, or goal
- Open and portable — your interests shouldn't be locked inside any one app
Drop a source — a link, a transcript, a voice note — and ask your LLM to process it as a spark. The format is simple enough that any LLM can follow it; consistent enough that you can search and connect across hundreds of sparks.
Use wikilinks [[spark-title]] to connect sparks to each other, to projects, to goals. As they accumulate, group related sparks into Maps of Content (MoC) — a living index for a topic, a skill, a decision you're working through.
You follow a topic across sources over weeks or months — podcasts, articles, tweets, conversations. spark.md gives each capture a permanent structure. When you revisit the topic, you have insights and connections — not just a pile of links you'll never open.
Each spark is a node. Over time, nodes form an organized knowledge graph — a personal map of everything that's caught your attention and what you made of it.
When you're researching a topic systematically — reading papers, watching talks, following debates — sparks become your research layer. Ask your LLM to process a new source as a spark and it doesn't just summarize: it connects to what you already have. Recurring ideas solidify. Contradictions surface. Gaps become visible. The more sparks you accumulate on a topic, the richer each new one arrives.
spark.md is a digestible unit of structured personal context:
- Essence gives agents the signal — what matters and why, without noise
- Connections give agents the graph — how this spark relates to other knowledge
- So What? gives agents the intent — what the user wants to do about it
An agent can easily digest spark to understand what was saved and why it matters. This is the structured personal context that agents need to be useful.
P.S.
This is an early version of the spark definition. It certainly will evolve over time, and I believe there definetely will be people coming up with their own ideas about spark structure, fields, and use cases. Will be happy to discuss any suggestions. The format itself is open and free to use with any tool, agent, or application.
spark.md was created as part of the Sparks project. It's a mobile-first app that lets you seamlessly capture your day-to-day interests and ideas into a structured personal inbox.