Build Your Own 100% Local AI Second Brain
Andrej Karpathy's concept of agentic engineering, which emphasizes production-grade agent development over 'vibe coding,' now has a dedicated tooling suite from Google. The Google Agents CLI provides a unified interface for the entire agent lifecycle, including scaffolding, evaluation, and deployment. By integrating the Agent Development Kit (ADK) and the A2A protocol, the tool simplifies the management of fragmented services like model provisioning, retrieval layers, and observability. This release aims to make the complex infrastructure required for enterprise-ready AI agents more accessible and practical for developers. Andrej Karpathy's 'LLM wiki pattern' has popularized the concept of an AI-managed second brain where agents maintain and cross-reference notes automatically. While manual integration of tools like Obsidian and Claude Code is possible, the open-source project Rowboat provides an out-of-the-box solution for this workflow. Rowboat indexes various data sources into a living knowledge graph and offers specialized work surfaces for email, meetings, and coding to facilitate AI collaboration. The tSNE algorithm typically suffers from quadratic runtime complexity, making standard implementations like those in Sklearn inefficient for datasets exceeding 20,000 points. openTSNE is presented as an optimized Python alternative that scales effectively to millions of data points. Benchmarks demonstrate that openTSNE is approximately 20 times faster than Sklearn, processing one million points in roughly 15 minutes compared to Sklearn's two-hour requirement for a much smaller dataset. The section also provides resources for further learning on GPU acceleration and manual t-SNE implementation.
閱讀原文 ↗目錄
Karpathy’s Agentic Engineering finally has proper tooling (built by Google)
Andrej Karpathy's concept of agentic engineering, which emphasizes production-grade agent development over 'vibe coding,' now has a dedicated tooling suite from Google. The Google Agents CLI provides a unified interface for the entire agent lifecycle, including scaffolding, evaluation, and deployment. By integrating the Agent Development Kit (ADK) and the A2A protocol, the tool simplifies the management of fragmented services like model provisioning, retrieval layers, and observability. This release aims to make the complex infrastructure required for enterprise-ready AI agents more accessible and practical for developers.
- Agentic engineering is defined by Andrej Karpathy as a discipline involving spec design, evaluation loops, and security oversight.
- Google's Agents CLI centralizes the agent development lifecycle, covering scaffolding, evaluation, and deployment in one interface.
- The Agent Development Kit (ADK) is model-agnostic, supporting Gemini, Gemma, and third-party models like Claude via Model Garden.
- The A2A protocol is used within the framework to handle coordination and communication between different agents.
- The evaluation layer in the CLI utilizes an LLM-as-judge scoring method to test agents before they are deployed.
- Deployment targets such as Agent Runtime and Cloud Run can be configured directly through the Agents CLI.
- The tool automates the setup of infrastructure-as-code (IaC) and CI/CD pipelines for staging and production environments.
Build your own 100% local AI second brain
Andrej Karpathy's 'LLM wiki pattern' has popularized the concept of an AI-managed second brain where agents maintain and cross-reference notes automatically. While manual integration of tools like Obsidian and Claude Code is possible, the open-source project Rowboat provides an out-of-the-box solution for this workflow. Rowboat indexes various data sources into a living knowledge graph and offers specialized work surfaces for email, meetings, and coding to facilitate AI collaboration.
- Andrej Karpathy proposed the 'LLM wiki pattern' where AI agents maintain a compounding knowledge base to avoid re-reading raw notes.
- Rowboat is an open-source AI co-worker that has achieved over 15,000 stars on GitHub.
- The system uses background agents to index emails, meetings, and notes into a self-updating knowledge graph.
- Rowboat supports integrations with existing tools including Obsidian, Slack, X, and Fireflies.
- The platform features specialized 'work surfaces' such as an email client, meeting note taker, and a dedicated code mode.
- Advanced features include scheduling agents via cron jobs and implementing guardrails for human approval of agent actions.
Accelerate tSNE with openTSNE
The tSNE algorithm typically suffers from quadratic runtime complexity, making standard implementations like those in Sklearn inefficient for datasets exceeding 20,000 points. openTSNE is presented as an optimized Python alternative that scales effectively to millions of data points. Benchmarks demonstrate that openTSNE is approximately 20 times faster than Sklearn, processing one million points in roughly 15 minutes compared to Sklearn's two-hour requirement for a much smaller dataset. The section also provides resources for further learning on GPU acceleration and manual t-SNE implementation.
- tSNE runtime complexity is quadratically related to the number of data points.
- Sklearn's tSNE implementation becomes difficult to use with datasets larger than 20,000 points.
- openTSNE is an optimized Python implementation designed to handle millions of data points.
- openTSNE is 20 times faster than the Sklearn implementation according to provided benchmarks.
- openTSNE can project 1 million data points in approximately 15 minutes.
- Sklearn takes two hours to process 250,000 data points.
- GPU acceleration can be used to speed up other machine learning algorithms beyond tSNE.