Building a Second Brain with DenchClaw

I've tried most of the second brain systems. Zettelkasten with Obsidian. Notion dashboards that became graveyards. Roam Research backlinking spirals. Notion AI. Mem. Logseq. The problem with all of them is that they treat knowledge as text — as notes to be linked. They don't treat knowledge as structured data about the world you actually operate in.

Your second brain, if it's going to be useful, needs to know about people, not just ideas about people. It needs to know about projects with statuses and deadlines, not just notes about projects. It needs to know about relationships between things, not just proximity of text blocks.

DenchClaw is a local-first, open-source AI CRM backed by YC S24. I've been using it as my second brain for the past year, and it's the first system that hasn't collapsed under its own weight.

Here's why, and how.

The problem with pure note-taking as a second brain#

The appeal of systems like Obsidian or Roam is the idea that knowledge lives in the connections between things. Backlinks, graph views, emergent structure. You write atomic notes, you link them, and over time a web of knowledge builds itself.

In practice, this breaks down for most people because:

Unstructured notes don't age well. A note from 2022 says "interesting idea about pricing." Is that still relevant? In what context? Who did I discuss it with? You can't query unstructured text the way you can query structured data.

People and projects need fields, not just paragraphs. When I'm thinking about a person — a founder I know, an investor in my network, a collaborator — I need to know their email, their company, how I met them, when I last talked to them. That's not note content. That's structured data.

The gap between notes and action is too wide. I write a note, then I have to remember to go back to it. A CRM with structured fields and views can surface the right information at the right time automatically.

What I needed was a system where the structured data and the free-form notes live together, with the structured data providing the scaffolding that makes the notes findable and useful.

That's exactly what DenchClaw's architecture provides.

DenchClaw's architecture as a second brain#

DenchClaw has two distinct layers:

Layer 1: Structured data. Objects (tables), fields (typed properties), entries (rows). Stored in DuckDB. Queryable with SQL. Viewable as Table, Kanban, Calendar, Timeline, Gallery, or List. This is the database layer — fast, queryable, reliable.

Layer 2: Entry documents. Each entry has an associated markdown file — a free-form document that lives alongside the structured fields. This is the narrative layer — where context, history, and nuance live.

Most knowledge management tools have one layer. DenchClaw has both, deeply integrated.

What I track in my second brain#

Here's the object structure I use:

People#

Every meaningful person in my professional and personal life. Fields: name, relationship category (investor, founder, friend, advisor, family), last contact date, how we met, current company, a "warmth" rating.

The entry document for each person is a running log of our relationship: notes from calls, things they mentioned, things I want to remember about them, open threads, promises made.

Companies#

Every organization I care about — my portfolio, competitors, potential partners, employers, communities. Linked to the people who work there.

Projects#

Everything I'm working on or thinking about. Fields: name, status (active, paused, complete, archived), category, start date, linked people and companies. The entry document is the project log — decisions made, things tried, lessons learned.

Ideas#

This is my capture inbox. Every idea that comes to me — product features, business models, essay topics, things to read. Fields: title, category, status (raw, developed, shipped, abandoned), energy level (how excited am I still?). Entry document is the full development of the idea.

Books and Articles#

What I've read and am reading. Fields: title, author, status, rating, date read. Entry document is highlights, notes, and connections to other things.

Decisions#

Major decisions I've made or am considering. Fields: date, context, options considered, decision made, rationale. Entry document is the full reasoning. This is invaluable for looking back — I can see the reasoning behind past decisions in full fidelity.

The power of relation fields#

The thing that makes this more than a collection of notes is DenchClaw's relation fields. Every project is linked to the people involved. Every idea is linked to the projects it might inform. Every decision is linked to the context (project, person, company) that drove it.

This creates a genuine knowledge graph — but unlike Roam's unstructured graph, mine is typed. I know that this link is "person works at company" not just "these two things are related."

In DuckDB, I can ask:

-- What ideas do I have related to my active projects?
SELECT i.title as idea, i.energy_level, p.name as project
FROM v_ideas i
JOIN v_projects p ON p.id = i.project_id
WHERE p.status = 'active'
ORDER BY i.energy_level DESC;
 
-- Which people are connected to my top ideas?
SELECT pe.name, pe.relationship_category, i.title as idea
FROM v_people pe
JOIN v_projects proj ON proj.id = pe.primary_project_id
JOIN v_ideas i ON i.project_id = proj.id
WHERE i.energy_level > 4;

No note-taking app gives me this. This is the query layer that unstructured knowledge tools lack.

Entry documents as the narrative thread#

The entry document system is where DenchClaw becomes genuinely different from a CRM.

For a person I know well, their entry document might be 2,000 words. A timeline of our relationship. What they're working on. What they care about. What we've discussed. What I want to ask them next time.

For a project I've been working on for a year, the document might be a living engineering log — every major decision with its rationale, every dead end with what I learned from it, every breakthrough with the context that made it possible.

For an idea I've been developing, the document is the full long-form exploration — the research I've done, the objections I've considered, the version that's actually getting closer to a thesis.

These documents are DenchClaw entry documents — markdown files that live on your disk, alongside the structured DuckDB data, associated with a specific entry. You can edit them in any text editor. You can grep them. You can version-control them. You can read them on a plane with no internet.

That's local-first as a second brain principle: your knowledge lives on your machine, in formats that last, accessible without dependency on any vendor's continued existence.

AI on top of structure#

The AI layer in DenchClaw amplifies what the structure provides. Because your data is structured, AI queries become dramatically more useful.

Instead of asking "what did I think about pricing models?" to a pile of notes, I can ask: "What are all my active projects in the B2B space, and what ideas do I have tagged as relevant to those, and which people in my network have experience with the problem?" And get a structured answer drawn from my DuckDB, not a semantic search over text.

Action fields let me build AI-powered workflows. An action on a People entry that says "generate prep notes for a meeting with this person, drawing from our interaction history." An action on an Ideas entry that says "develop this into a blog post outline." An action on a Project entry that says "generate a weekly status summary from the entry document log."

The local-first principle#

I want to say something directly about why local-first matters for a second brain.

Your second brain is a map of your mind. It contains the most personal professional data you have: everyone you know, every idea you've had, every decision you've made, every project you've worked on. Putting that data in a vendor's cloud means trusting that vendor — their security, their business model, their continued existence, their pricing decisions, their data policies.

I've watched tools that people built their entire knowledge management system on disappear or pivot. Roam's pricing drama. Notion's various capability changes. WorkFlowy going quiet. The risk isn't theoretical.

DuckDB is a single file on your disk. It will be readable forever with open-source tooling. Markdown documents are text files — they predate the web and will outlast every SaaS company currently operating. This is the format decision that makes a second brain actually durable.

Getting started#

npx denchclaw

Start with one object. People is the highest-leverage starting point — most knowledge work is ultimately about people. Create a simple People object, import your contacts, and start adding entry documents as you have conversations.

Add objects as you feel the need. Don't build the whole taxonomy upfront — let it emerge from actual usage.

FAQ#

How is DenchClaw as a second brain different from Obsidian? Obsidian is a note-taking tool with optional structured data (via plugins). DenchClaw is a structured data tool with integrated free-form documents. If you think in notes that link, Obsidian is better. If you think in entities that have attributes and relationships, DenchClaw is better. Many people use both.

Can I import my existing notes into DenchClaw? Yes — markdown files can be imported as entry documents. Structured data from Notion, Airtable, or other tools can be imported as CSV.

Is DenchClaw useful without any coding? Fully. The AI and browser agent handle automation. The UI handles all basic operations. SQL and code unlock advanced use cases but aren't required.

How do I make sure I actually use this consistently? The systems that stick are the ones you use when they're useful, not the ones you maintain as a ritual. Build DenchClaw into workflows that already exist — before calls, after meetings, during weekly reviews. The barrier is low (npx denchclaw and you're in).

What's the most valuable thing to track first? People. Specifically: the people in your professional network who matter to you, and when you last talked to them. The "who do I need to reach out to" view alone makes the whole system worth building.

Ready to try DenchClaw? Install in one command: npx denchclaw. Full setup guide →