Loading video...

Video Failed to Load

Go Home

Used Karpathy's LLM Wiki to turn Claude Code into a self-evolving system. Fed it my frontend & CRM sources, and it built a fully updating dashboard with Shadcn packages and constantly improving and linking everything for smarter outputs! Full setup video:

133,755 views • 7 months ago •via X (Twitter)

22 Comments

Eugeniu Ghelbur's profile picture
Eugeniu Ghelbur7 months ago

I built an open-source Claude Code skill that implements @karpathy's LLM-Wiki concept. It's not a second brain, it's a living vault. Every source you feed it rewrites existing pages, resolves contradictions, and synthesizes patterns you didn't ask for. The vault doesn't grow. It evolves. 22 commands, scheduled agents, thinking tools that argue with you using your own notes.

Luke's profile picture
Luke7 months ago

The self-evolving part is the breakthrough. Most AI setups are stateless, you start over every conversation. Once the system compounds knowledge across sessions, everything changes.

MrDarkoo's profile picture
MrDarkoo7 months ago

Where's your GitHub link 👀 🤔

Everything AI's profile picture
Everything AI7 months ago

the risk with self-referential codebases is you optimize for what the LLM thinks is improving rather than what actually ships. karpathy's wiki concept is solid but feeding it your own generated code can compound errors fast

techarena.au's profile picture
techarena.au7 months ago

You turned Claude into a self-updating dev? Respect. Drop the vid and repo, imma feed it my dumpster CRM and watch it glow up lol

Latent Node's profile picture
Latent Node7 months ago

Did you try with OpenEvolve too?

Claudia's profile picture
Claudia7 months ago

the self-evolving part is what matters. I do something similar but with persistent memory files instead of a wiki. every session writes observations about what worked and what didnt. next session reads them on boot. after a few weeks the system knows things about the codebase that no single session could discover

Tal's profile picture
Tal1 month ago

100% on statelessness. Claude Code and Cursor don't lack intelligence — they lack persistent memory between sessions. I solved this for my daily NASDAQ bot by making the repository the source of truth (NOTES.md + auto-push WIP checkpoints). Session crashed? Just send "." and the agent inspects memory and resumes without asking. Demo in the video above!

Kyle's profile picture
Kyle7 months ago

Karpathy's wiki as the foundation is a smart move - that man just quietly ships the best learning resources. 👀

Yash Khivasara's profile picture
Yash Khivasara7 months ago

does it stay accurate after a few update cycles? every self-evolving setup i've tried eventually starts confidently rewriting correct stuff

Chat Data's profile picture
Chat Data7 months ago

The interesting shift is using the knowledge base as shared context instead of a one off prompt. Once the source docs stay fresh, the agent output gets much more consistent and reviewable.

Balvinder Kalon's profile picture
Balvinder Kalon7 months ago

the self-evolving part is what makes this interesting. most people use claude code as a one-shot tool. feeding it your own data sources and letting it iterate on its own context is where it gets weirdly powerful.

Vishal | Video editor's profile picture
Vishal | Video editor7 months ago

feeding ur own crm sources into claude code + karpathy's llm wiki is such a big brain move honestly

Klayb's profile picture
Klayb7 months ago

agents can generate anything now but most of them still reset every session, so the real unlock is systems that actually retain and structure knowledge over time. been trying this with llmwiki and it feels closer to that direction

tamhn's profile picture
tamhn7 months ago

this is the pattern i've been trying to articulate. the wiki isn't just storage, it's a reasoning scaffold. when the context is structured right the agent becomes capable of things it couldn't do from scratch each time.

Mohamed Anis's profile picture
Mohamed Anis7 months ago

This is what agentic AI was always meant to look like, not just answering questions, but actively building and improving your systems in real time.

Kelvin Lee's profile picture
Kelvin Lee7 months ago

I've added it to my #piagent running gemma4 locally 100% 🦾

Kn's profile picture
Kn7 months ago

We've actually made a similar format in terminal based of Karpathy's idea of LLM Knowledge Bases. Give it a spin and let me know what you think:

lifcc's profile picture
lifcc7 months ago

The manual cycle of 'Claude breaks rule → figure out why → add fix to CLAUDE.md' gets old after the 10th session. The real pain is it never gets shorter — just bloats until you forget what half the rules mean. Does your LLM wiki auto-consolidate, or do you still prune by hand?

Gregor's profile picture
Gregor7 months ago

Feeding it your own codebase is where it gets actually useful, not just toy demos.

toni's profile picture
toni7 months ago

31K views on a dev workflow tweet signals hungry builders more than casual interest. When technical content outperforms entertainment, it's market demand not algorithm luck. Infrastructure play.

Christian Wink's profile picture
Christian Wink7 months ago

What are Shadcn packages? @grok

Related Videos

I still don’t understand why people keep building a Claude second brain by hand when these 10 repos already built the whole system They turn your Claude sessions into a wiki, let Claude read and write your vault from your phone, and one of them comes straight from Obsidian’s own CEO 01 llm-wiki → your Claude Code and Codex sessions become a searchable wiki, with an MCP server to query it 02 llm-wiki-claude-skill → one skill file teaches Claude to ingest sources, answer from them, lint the wiki and compile it into clean pages 03 obsidian-skills → Obsidian’s own CEO teaches Claude how Obsidian really works 04 llm-wiki → a Claude Code plugin with init, ingest, compile, query and lint, where every change is committed to git 05 obsidian-dev-second-brain → your tickets, sprints and project docs live in the vault as notes, and Claude keeps them in sync with the code 06 obsidian-mcp-server → 14 tools to read, write, search and surgically edit notes, tags and frontmatter 07 mcpvault → safe vault access with a read-only mode, and Obsidian doesn’t even need to be running 08 obsidian-claude-code-mcp → connect Claude Code to your open vault with one /ide command 09 obsidian-web-mcp → reach your vault from your phone or any MCP client, with OAuth and a Cloudflare Tunnel 10 obsidian-mcp → a git-backed vault with 18 tools, running locally or on AWS Lambda Nobody is going to build your second brain for you, and every chat you close without one is knowledge thrown away. Bookmark this, pick one repo from 01 to 05 and ingest your first source tonight. A month from now you have a wiki, and everyone who skipped it is still opening a blank chat.

Ryven

10,722 views • 14 days ago