Reading
A knowledge management agent built on the LLM Wiki mode proposed by Andrej Karpathy, using a medallion architecture for hierarchical storage, supporting hybrid deployment of local Ollama inference and cloud Gemini, and fully usable offline.
Section 01
A knowledge management agent built on the LLM Wiki mode proposed by Andrej Karpathy, using a medallion architecture for hierarchical storage, supporting hybrid deployment of local Ollama inference and cloud Gemini, and fully usable offline.
Section 02
LLM Wiki Agent directly originates from a design pattern shared by Andrej Karpathy. The core idea of this pattern is to organize the knowledge base into Wiki-style single-concept pages, establish connections between concepts via [[wiki links]], and use graph traversal as the navigation mechanism.
Unlike traditional document management systems, the Wiki mode emphasizes discretization and association of concepts. Each page carries only one core concept, forming a knowledge network with other concepts through links. This structure is naturally suitable for the understanding and reasoning of large language models.
Section 03
The project implements a medallion architecture, dividing knowledge into three levels:
Section 04
A read-only core knowledge source containing verified authoritative information. The agent can read but will never overwrite content in this layer. In search ranking, Gold Layer results have the highest priority.
Section 05
A writable knowledge storage layer for the agent, used to save knowledge organized, summarized, and generated by the agent. This is the main place where the agent performs knowledge work, with priority in search results lower than the Gold Layer.
Section 06
Raw data source, including imported documents, web-scraped content, and other unprocessed information. As raw material for knowledge processing, it has the lowest search priority.
This layered design ensures knowledge quality control and traceability, preventing raw noise from contaminating the core knowledge base.
Section 07
A key highlight of the project is the flexible choice of inference modes, allowing users to freely combine based on privacy and performance needs:
| Usage Scenario | Dialogue Inference | Embedding Vector | Network Requirement |
|---|---|---|---|
| Fully Local | Ollama (Local) | Ollama (Local) | None |
| Hybrid Mode | Ollama (Local/Cloud) | Gemini (Cloud) | Only for Embedding |
| Fully Cloud | Ollama Cloud | Gemini (Cloud) | Network Required |
This design breaks the binary opposition of 'local must be slow' and 'cloud must leak', allowing users to choose the most suitable inference mode based on specific tasks. Sensitive content is processed using local models, while complex reasoning tasks can call cloud capabilities.
Section 08
The project builds a rich knowledge graph, supporting six edge types to describe relationships between concepts:
Each edge carries source information (link_text, link_kind, evidence), ensuring the interpretability and traceability of the graph.
Keep going with more reads from the same topic.
Splinter is a minimalist, high-performance key-value (KV) and vector storage system enabling zero-latency inter-process communication via shared memory and atomic operations. With only 766 lines of core code, it supports millions of operations per second and 768-dimensional vector storage, offering a new architectural approach for local LLM inference and data-intensive applications.
Folkering OS is the world's first AI-native bare-metal operating system, entirely written in Rust no_std without relying on Linux, POSIX, or libc. It can generate commands from scratch, compile them into WASM, and run them in 10 seconds, achieving true self-evolution.
This project explores how to use Large Language Models (LLMs) to detect logical vulnerabilities in Ethereum smart contracts, especially business logic flaws that are difficult for traditional static analysis tools to capture, such as economic manipulation and execution order errors.
Cross-generational hardware testing based on NVIDIA A10G and H100 GPUs, comparing and analyzing the throughput, latency, and concurrent scalability of three mainstream LLM inference frameworks. SGLang achieves a 3.4x performance advantage over vLLM on H100, while Ollama faces architectural bottlenecks in high-concurrency scenarios.