ideia · in construction view source →

Pokomos: An Observatory for the Thoughts You Chose to Keep

A personal knowledge management observatory. Not 10,000× productivity. Just a simpler human, better organized.

The problem is not information scarcity

We are not starving for content. The opposite is true: there is too much of it, pushed at us relentlessly, optimized for attention that we cannot sustain. Feeds, notifications, shorts, reels, threads. The infrastructure of modern consumption was built so the firehose never stops. You scroll, you swipe, you consume. And you forget.

Byung-Chul Han calls ours a society of tiredness — the exhaustion that comes from constant optimization, from self-programming, from the dopamine circuits of infinite content. The diet is poor. The feeds are fast food for the mind.

When people talk about being a 10,000× developer with AI, I wince a little. That framing misses something important. What would a 10,000× human even look like? More prolific, perhaps. More output. But also: more tired, more fragmented, more lost in their own exhaust. The question was never "how do I consume more." It was "how do I think well with what I have."

Pokomos is not about productivity

Pokomos stands for Personal Knowledge Management Observatory. The name is a little ridiculous, which I like. It describes the thing I needed and couldn't find: a place to observe my own thoughts — the ones I deliberately chose to keep, to reflect on, to come back to.

It is not a feed aggregator. It does not summarize the internet for me. It is not an oracle. Pokomos does one thing: it watches the vaults where I store my own stuff and helps me see patterns in what I captured, what I said, what I read, what I wrote. The control of ingestion is mine. What I eat — cognitively speaking — is my responsibility. Pokomos just helps me see what I've been eating.

This distinction matters. The difference between a newsfeed and a personal library is the difference between being fed and feeding yourself. Pokomos is a library. You curate it. It curates nothing for you.

The old form: index cards

People have been doing this for centuries. Niklas Luhmann famously kept a Zettelkasten of index cards, each one a single idea, cross-referenced to others, slowly accumulating into a body of thought that produced dozens of books. Montaigne kept commonplace books. Dostoevsky, under deadline pressure, dictated The Gambler to Anna Grigorievna — his stenographer, later his wife. The technique is ancient. You capture, you cross-reference, you return.

What changed is not the impulse. What changed are the tools. Index cards are slow. Bliki and wiki software made it faster. Obsidian and Roam made it graph-shaped. Now large language models can do the tedious bookkeeping: the cross-referencing, the tagging, the summarizing, the linking. The human still curates. The machine handles the linkage.

Pokomos sits on top of that. It's one layer higher. It does not produce the cards. It watches the multiple decks I keep and shows me what's in them, what connects what, what's been neglected, what's growing.

What Pokomos observes

My vaults are not scraped from the internet. They are the residue of my own cognition. A few examples of what Pokomos watches for me:

  • Voice memos I record while walking. I talk into a Telegram bot when I have a thought. Whisper transcribes it. A classifier tags it. A synthesizer turns it into a concept in my knowledge base. The whole pipeline runs without me opening a laptop.
  • Messages I send to a community chat. I am in a Telegram group with about 950 developers (Hipsters Builders). When I post a link or a thought, a bot captures it into a separate vault. A mini-classifier tags it. Later, these become material for a weekly newsletter.
  • Book quotes I copied out by hand in 2009. I kept two Google Docs with passages from Dostoevsky, Hesse, Tolstói, Wilde, Ionesco, Nietzsche, Voltaire, Shakespeare. They sat in Drive for fifteen years. Now they live in /quotes, structured, cross-referenced, queryable.
  • Published blog posts. Everything I've written on paulo.com.br becomes part of the library. Pokomos can ask, across everything you've written and read, what patterns emerge?

There are four vaults right now. None of them contains content pushed at me by an algorithm. They contain things I said, wrote, or deliberately chose to keep. That is the whole idea.

Ingestion, not consumption

There is a word I keep coming back to: digestion. Ingest something, sit with it, turn it over, let it become part of you. Most of what modern feeds do is the opposite of this. You scroll, you react, you move on. Nothing is digested. Nothing is kept. Even the "saved for later" lists accumulate into unread graveyards.

Pokomos assumes the ingest step is already done. You recorded the voice memo. You copied the quote. You sent the message to the group. You published the post. Pokomos joins in at the next step — organization, retrieval, synthesis — and lets the ingested things compound over time.

This is the same pattern Karpathy described in his LLM Wiki gist: raw sources become a compiled, living wiki; the LLM maintains the cross-references; the human curates the sources and asks the questions. Daniel Miessler gets at the same thing from another angle with his Personal AI Infrastructure — the idea that each of us deserves infrastructure that understands us specifically, not the average consumer. Pokomos is the observation layer on top of that.

Three layers: Observe, Operate, Orchestrate

To be concrete, I think of Pokomos as three layers, roughly in the order I am building them:

Observe. See what the vaults contain. Browse signals, knowledge-base nodes, drafts. Group by tag, by vault, by time. Notice which agents have been active, which have gone quiet. Notice which topics are recurring. This is the passive layer and the most important one. Most of what I need from Pokomos lives here.

Operate. Act on the vaults. Edit a draft inline. Move a signal. Re-classify. Merge duplicates. Run a pipeline on demand. The point is not to do this constantly — the point is to handle the maintenance that otherwise accumulates.

Orchestrate (optional, maybe never). A meta-agent with context across all the vaults, able to answer cross-vault questions. Which ideas have I returned to most often this month? What does my reading have in common with what I'm recording? Which drafts are stuck? This is the closest thing Pokomos has to a Q&A layer, and it is the part I am most cautious about — the easiest place to add complexity that nobody asked for.

I explicitly do not want Pokomos to be a control plane that starts and stops agents. The agents have their own triggers — webhooks, crons, the Telegram bot that listens when I speak. Pokomos is not their boss. Pokomos is a window into the mess they leave behind.

The "simple human" framing

I want to say this plainly because it is the part that gets lost whenever we talk about AI and productivity: I am not trying to become a 10,000× anything. I am trying to become a person who keeps better track of his own thoughts. That is it.

The goal is to be a simple human, better organized. Not multiplied. Not optimized. Not turned into a content factory. Just organized, the way people have always tried to be organized, with the best tools available to them at the time. Luhmann had index cards. Montaigne had a commonplace book. Dostoevsky had Anna and her shorthand. I have Telegram, Whisper, Sonnet, Haiku, a bunch of markdown files, and Pokomos.

The machines are doing the bookkeeping. That is what they are good at. The human — the one simple human, not ten thousand of him — is still the one who decides what matters, what to revisit, what to think about next, and what, ultimately, to say. Nothing about the system takes that away.

Against feed-shaped solutions

If Pokomos ever starts to look like a feed — scrolling, infinite, pushing content at me that I didn't ask for — I will have failed. The whole point is to build an inverse of that shape. You should not open Pokomos to see what's new. You should open it to see what you've been thinking about lately. Those are different questions. One is about the world outside demanding your attention. The other is about returning attention to your own interior.

The difference is the difference between fast food and home cooking. You can eat fast food every day and never once know what you actually like. You can cook at home and discover, over years, that you prefer one ingredient over another, one combination over another, that your tastes have changed. The act of preparation is the knowing. Pokomos is home cooking for thoughts.

How this grows

In the near term: I add vaults, I refine the observation layer, I build small artifact generators (a newsletter from the last ten days of Hipsters messages, a cross-vault query that answers "what did I think about X in 2021 vs 2024"). In the medium term: I add richer retrieval, better linting of the knowledge base, maybe a proper Q&A agent that compounds good answers back into the wiki. In the long term: I keep it simple.

The risk is always the same risk: adding complexity for its own sake. Every new feature in Pokomos has to pass the test of whether it serves the core idea — observing, not aggregating; helping me think about my own thoughts, not burying me in somebody else's. I hope I can hold that line.

References and influences