GitHub 2026-11-03 Daily Trending Projects

A CLI for AI agents to operate desktop apps using accessibility-tree refs instead of pixel matching.

A memory plugin for Claude Code users that recalls session knowledge automatically with SQLite, Chroma, and lifecycle hooks.

Scrapling is an adaptive web-scraping framework for the modern web, combining stealthy fetchers, multi-session concurrent crawls, and adaptive element tracking to build resumable, real-time, high-performance crawling workflows.

Hyprland is a highly customizable dynamic tiling Wayland compositor focused on visuals and plugin extensibility; repository metadata and license are missing, so verify project health and compliance before adoption.

Extends a single Claude Code session into a studio-like template with 49 specialized agents and 72 skills covering design, implementation, and release—useful as a structured starting point but requires license and maintenance verification before adoption.

FreeToken lets developers run 290B+ MoE models locally on consumer GPUs by treating CPU, GPU, and memory as an elastic inference platform.

T3 Code is a minimal web GUI and cross-platform desktop for Codex/Claude coding agents, enabling quick validation of agent workflows; it is very early-stage with limited maintenance and contribution paths, suitable for exploratory trials and proof-of-concept.

Kronos is the first open-source foundation model for financial K-line sequences, combining hierarchical discretization and an autoregressive Transformer to provide reproducible baseline tools for quant research and probabilistic asset forecasting.

A differentiable geometry library for PyTorch vision teams, keeping batched augmentation, camera models, and feature matching on the GPU.

A pure-C inference engine for running 744B–2.8T MoE models locally, streaming experts from disk instead of fitting everything in VRAM.

A set of engineering skills for Claude Code, Codex, and other agents, smaller and more composable than full process frameworks.

Worktrunk manages Git worktrees for developers running Claude Code in parallel, replacing cumbersome path-based commands with wt.

Paperclip orchestrates AI agents with tasks, org charts, budgets and audits.

nanochat is a minimal, hackable end‑to‑end LLM training and chat framework that can rapidly reproduce GPT‑2‑grade models on a single machine or an 8×H100 node; it is well suited for research, teaching, and fast experimentation, but pay attention to license, contributor metadata, and hardware costs.

Heretic pairs directional ablation with TPE-based automatic tuning to offer a one‑click, low‑damage decensorship tool for GPU-equipped researchers and engineers—useful for probing model semantics and generative boundaries, but requiring careful legal and misuse risk assessment.