← Andrej Karpathy
Cole Medin
Finally, an Open Standard for the Karpathy LLM Wiki is HERE
⏱ 20 min video · 3 min read2 Jul 2026Worth watching
TL;DR
Google has released an Open Knowledge Format (OKF) — a minimal standard layered on top of Andrej Karpathy's LLM wiki concept — that lets anyone build, share, and consume personal knowledge bases in a consistent, agent-readable way. Cole Medin walks through what OKF standardizes (folder structure + YAML front matter), why it matters for personal agents, and demos a shareable AI coding knowledge bundle he built using the format.
Key points
1
Karpathy's LLM wiki gist (40,000 GitHub stars) has no standard, so every person's wiki is structured differently and cannot be shared or consumed reliably by other agents.
2
Google's Open Knowledge Format (OKF) adds a minimal standard on top: consistent folder/index structure and a small set of YAML front matter fields, with 'type' being the only required field.
3
OKF is analogous to what MCP did for agent-to-tool communication — it does the same for agent-to-knowledge-base communication, enabling shared and team-wide wikis.
4
To adopt OKF, copy the spec.md from the OKF repo into your coding agent; it can then build a new wiki or refactor an existing one to the standard in one shot.
5
Cole has published a shareable OKF bundle of his best AI coding YouTube videos that anyone can clone into their second brain (Obsidian, Notion, etc.) and immediately query.
Actionable insights
→
Copy the OKF spec.md into your coding agent and ask it to build or refactor your existing LLM wiki — the spec is self-contained and the agent can one-shot the task.
→
Use the 'type' YAML field (the only required OKF field) consistently across all documents so your agent can filter by category without reading every file.
→
Clone Cole's AI coding OKF bundle from GitHub using the provided prompt and spec, then immediately ask your agent questions about AI coding topics like the PIV loop or context engineering.
→
Add a top-level index document and per-folder indexes to your wiki so agents use progressive disclosure — reading broad indexes first, then drilling into specific entity pages only when needed.
→
Even if you never plan to share your wiki, adopting OKF now lets you benefit from community-shared ideas on entity page structure and organization under a common standard.
Notable quotes
“It's like what MCP did for agent-to-tool communication, this OKF is doing for agent-to-knowledge-base communication.”
“The fact that there's not much substance here might seem counterintuitive, but I think that is actually a good thing.”
“I know that Google is lagging in the AI race right now, Gemini is not as good as GPT and Claude, but they have been releasing some really good stuff on how to leverage LLMs effectively — and I think that's a totally different lane than building LLMs.”
Worth watching?
✅
Worth watching the full video?
Worth watching if you actively use or plan to build an LLM wiki — the live demo of the OKF bundle being queried by an agent is the clearest illustration of the standard in action, though the key concepts and setup steps are all captured here.
Saved you some time? The creator still deserves a like.
Want this for your own channels?
Add the channels you follow. Every new video summarised and in your inbox the moment it drops. From £4/month.
Try it free