GitHub 2026-04-14 Weekly Trending Projects

Condenses Andrej Karpathy's observations about LLM coding pitfalls into four installable CLAUDE.md principles—explicit assumptions, simplicity-first, surgical changes, and goal-driven execution—helping teams limit unnecessary edits and make LLM-driven changes verifiable.

Hermes Agent is a server-deployable autonomous AI agent offering a terminal-first interface, multi-platform messaging gateway, persistent memory and skills hub, sandboxed terminals, and scheduling—suitable for long-running, auditable, and extensible automation in engineering and operations.

Multica delivers a self-hosted or cloud-managed platform to turn coding agents into assignable, monitored teammates with reusable skills—aimed at engineering teams seeking controlled automation.

DeepTutor centers on an agent-native architecture that combines TutorBots, persistent memory, and RAG knowledge bases to deliver an extensible personalized tutoring and collaborative workspace—well suited for autonomous learning, teaching automation, and knowledge-driven conversational education products.

Archon is a knowledge and task management platform for AI coding assistants, offering an MCP interface, document retrieval and task collaboration to unify agent context and improve model-assisted development outcomes.

MarkItDown converts varied documents into structured Markdown for LLM pipelines, enabling automated text workflows while requiring care with optional dependencies and plugin compatibility.

AI Edge Gallery is an on-device offline generative-AI sandbox for mobile, supporting local Gemma 4 inference, model management, and multimodal interaction—suitable for privacy-sensitive development and performance evaluation.

LiteRT-LM is Google's edge-focused large-model inference framework emphasizing multimodality and hardware acceleration, suited for deploying high-performance offline or near-edge inference on mobile, desktop, and embedded devices.

PersonaPlex, built on Moshi and the Helium backbone, delivers low‑latency full‑duplex speech dialogue with combined text‑role and audio voice control, suited for research, real‑time interactions, and customer‑service prototyping and evaluation.

SEO Machine is a Claude Code–based end-to-end SEO writing and optimization workflow that combines research, writing, review, and performance data to produce brand-aligned long-form SEO content; however, license and maintenance activity should be confirmed before production use.