Presented by
Good morning, AI enthusiasts. If you use a third-party Claude-powered tool like OpenClaw, check your account before noon today. Anthropic just ended subscription coverage for external tools, requiring extra usage bundles or an API key to keep your workflows running.
The policy arrived without much warning, and it signals something larger about how Anthropic is drawing the line between its consumer subscription and developer API.
In today's recap:
Anthropic ends Claude subscription coverage for third-party tools
Karpathy's LLM builds and maintains its own wiki
Build your own LLM-powered wiki with Obsidian
Anthropic's $400M biotech acquisition, explained
4 new AI tools, prompts, and more
ANTHROPIC
Anthropic cuts third-party access from Claude subscriptions
Recaply: Anthropic just announced that Claude subscriptions will no longer cover usage on third-party tools, requiring users to buy discounted extra usage bundles or switch to a Claude API key for tools like OpenClaw.
Key details:
Third-party tools built on Claude's API, like OpenClaw, were previously accessible through a standard Claude subscription but now require separate billing, with usage counted outside the subscription.
The change affects any external tool connected via the Claude login system, though exact counts of affected services and bundle pricing weren't disclosed. Extra usage bundles are available at a discount for the transition.
The announcement came from @bcherny, described in their profile as working at Anthropic, suggesting a deliberate monetization shift rather than a technical update.
The change took effect on April 6, 2026 at 12pm PT, giving third-party tool users no prior public notice to prepare.
Why it matters: Anthropic never explicitly said third-party tools were included in the subscription. Developers assumed they were and built products around that assumption. Now they're finding out they were wrong, and their users are too. Small tool makers have two choices: absorb the API costs themselves or raise prices for users. Either way, the era of free Claude access through a subscription is over for third-party apps, and it happened without much warning.
PRESENTED BY HUBSPOT
Turn AI into Your Income Engine
Ready to transform artificial intelligence from a buzzword into your personal revenue generator?
HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.
Inside you'll discover:
A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential
Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background
Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve
Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.
AI RESEARCH
Karpathy shows how LLMs can build your wiki
Recaply: Andrej Karpathy just shared an LLM-powered knowledge base system where an AI compiles raw research documents into a self-maintaining markdown wiki, skipping vector databases entirely for datasets under roughly 400K words.
Key details:
Raw documents in a
raw/folder are compiled by an LLM into a structured wiki of.mdfiles in Obsidian, with the AI writing summaries, concept articles, and backlinks between related ideas.Karpathy's own research wiki spans around 100 articles and 400K words, with the gist earning 4,951 stars and 998 forks within days of posting and the X post receiving 48.7K likes.
Karpathy described the whole setup as a "hacky collection of scripts" and said he sees room for "an incredible new product" to replace it, hinting at a possible future launch.
The gist was posted on April 4, 2026 and quickly went viral. The system is available now as a DIY setup, with no commercial product announced yet.
Why it matters: Most people's relationship with LLMs is transactional. You ask, it answers, and nothing persists. Karpathy's system flips that. The LLM doesn't just answer questions, it maintains a living knowledge base that grows with every query. The key insight is that structured Markdown files are indexable enough for mid-scale wikis without vector embeddings, making the whole thing transparent and human-readable. It's the second brain concept, but the LLM does the filing.
TUTORIAL
Build your own LLM-powered personal wiki with Obsidian
Recaply: In this tutorial, you will learn how to set up Karpathy's LLM knowledge base system, turning your raw research documents into a self-maintaining markdown wiki without building a RAG pipeline.
Step-by-step:
Create two folders in your Obsidian vault:
raw/for source materials (articles, papers, PDFs) andwiki/where the LLM will write all content. Install the Obsidian Web Clipper browser extension to clip web articles directly as.mdfiles intoraw/, saving images locally alongside each file.Open a new chat in Claude or ChatGPT and paste this prompt: "You are my research librarian. Read all files in
raw/and compile a structured wiki inwiki/. For each concept, write a short article, add backlinks to related topics, and maintain anindex.mdwith one-line summaries of every article."Upload your
raw/files to the LLM (drag and drop or paste contents). Let it run the compilation pass. It will create.mdfiles for each concept, generate summaries, and write cross-references. The first pass works best with 10 to 20 focused documents on a single topic.Open the
wiki/folder in Obsidian and enable Graph View (View → Graph View) to see your knowledge map. For Q&A, start a new session, share yourwiki/index.mdplus the relevant articles, and ask questions. The LLM navigates the index to answer without needing a vector database.Run a monthly linting pass by prompting the LLM: "Review my wiki for inconsistencies, stale data, missing connections, and article candidates. Output a prioritized list of suggested changes." File the outputs back into
wiki/so every query adds to the base over time.
Pro tip: Prompt the LLM to check images in raw/ and update any wiki articles that should reference them. The LLM can analyze images via vision and insert references back into the relevant .md files automatically.
TOOLS
Trending AI Tools
🧠 Gemma 4 - Google's most capable open model family
🤖 OpenClaw - Self-hosted AI assistant
📝 Obsidian - Markdown-based personal knowledge management app
🎬 Seedance 2.0 - ByteDance's multimodal video generation model
NEWS
What Matters in AI Right Now?
Adaptive introduced Triggered Agents, a feature that fires AI agents automatically when events happen in connected tools like Shopify, Stripe, or Calendly, with each trigger spawning an agent that can research, draft emails, or update spreadsheets using the user's instructions.
xAI introduced Quality Mode on Grok Imagine, powered by its most advanced image generation model, adding enhanced detail, stronger text rendering, and higher creative control, now available on web and mobile.
Perplexity launched Computer for Taxes, a module that helps prepare U.S. federal returns on official IRS forms, review professionally prepared returns, and build tax dashboards, with knowledge updated to cover recent legislation including the OBBBA budget law.
Nous Research released Hermes Agent v0.7.0 with a redesigned memory system that's now an extensible plugin, offering six third-party provider options ready out of the box with setup via
hermes memory setup.Anthropic acquired stealth biotech startup Coefficient Bio in a $400M stock deal, with the approximately 10-person team joining Anthropic's health and life sciences division, extending its push into computational drug discovery alongside Claude for Life Sciences.
Anthropic rolled out Microsoft 365 connectors for Claude across all plans, letting users connect Outlook, OneDrive, and SharePoint to bring email, documents, and files directly into Claude conversations.
Microsoft's terms of service describe Copilot as "for entertainment purposes only," with the company calling the language "legacy" and promising an update, while OpenAI and xAI carry similar disclaimers warning users not to treat AI outputs as factual.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
