Garry Tan Open-Sources GBrain Personal Knowledge System

Garry Tan Open-Sources GBrain Personal Knowledge System

Y Combinator CEO Garry Tan released GBrain on GitHub this week — a self-enriching entity graph-based personal knowledge system built in just 12 days by ingesting over 17,000 pages [3]. The system features 25 agent skills for compounding knowledge and integrates with tools like Obsidian and OpenClaw for persistent agent memory.

GBrain operates as what Tan describes as a "markdown-shaped Postgres database for AI agents," enabling long-term memory capabilities that go well beyond traditional RAG setups [4]. Early discussions in the OpenClaw community highlight its potential as a second-brain tool that moves beyond simple retrieval to active knowledge synthesis.

Andrej Karpathy Proposes LLM-Wiki Pattern for Proactive Knowledge Management

Former Tesla AI director Andrej Karpathy published a GitHub Gist yesterday outlining the "LLM-Wiki" pattern — a system where AI agents proactively build and update persistent Markdown wikis from raw documents [5]. As Karpathy noted on X: "Something I'm finding very useful recently: using LLMs to build personal knowledge bases" [6].

The approach enables compounding knowledge unlike passive RAG retrieval, with tools like Obsidian Web Clipper and LogR automating document ingestion [7]. Early implementations report 70x efficiency gains over traditional vector databases, sparking fresh debates about the future of personal knowledge management systems.

EU AI Act Bans Emotion Recognition in Workplace Settings

Article 5 of the EU AI Act, which became applicable February 2, 2025, prohibits emotion recognition in workplaces and educational settings, along with biometric categorization systems [8]. The regulations carry penalties up to 7% of global annual turnover, with exceptions only for safety or medical applications.

Enforcement remains fragmented across EU member states, and notably, in-house tools are not exempt from these restrictions [9]. This creates potential compliance challenges for Nordic and European enterprises using AI-powered meeting tools that might inadvertently capture emotional or biometric data during calls.

What This Means For Your Meetings

The meeting intelligence space is rapidly evolving toward more sophisticated, less intrusive capture methods. Fathom's bot-less approach signals that the industry recognizes meeting participants' growing fatigue with obvious recording tools, while the technical innovations from Tan and Karpathy point toward more intelligent knowledge synthesis rather than simple transcription storage.

The EU's emotion recognition ban adds a crucial compliance layer for European organizations. Many meeting tools now incorporate sentiment analysis or speaker emotion detection — features that could run afoul of Article 5 regulations. Nordic companies should audit their meeting intelligence stack to ensure compliance, particularly for any tools that analyze participant emotions or attempt biometric identification.

The shift from passive recording to active knowledge building represents the next frontier. Rather than just storing meeting transcripts, the most valuable systems will proactively connect insights across conversations, building institutional memory that compounds over time. Key takeaway: The future belongs to meeting tools that capture less obtrusively while building knowledge more intelligently — and do so within European regulatory boundaries.

Sources

1. https://techcrunch.com/2026/04/15/fathom-adds-a-bot-less-meeting-mode-in-a-bid-to-take-on-granola
2. https://www.businesswire.com/news/home/20251008689349/en/Fathom-Expands-with-New-Features-Designed-to-Extract-Maximum-Value-From-Every-Conversation
3. https://github.com/garrytan/gbrain
4. https://www.reddit.com/r/openclaw/comments/1sizujg/anyone_tried_gbrain_as_a_memory_solution
5. https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f
6. https://x.com/karpathy/status/2039805659525644595?lang=en
7. https://pub.towardsai.net/andrej-karpathy-killed-rag-or-did-he-the-llm-wiki-pattern-7824d876e790
8. https://artificialintelligenceact.eu/article/5
9. https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-5