ZCode leads at 3615 stars/day as agent tooling dominates
We don't tell you what's popular — popularity lags and is gameable. We tell you what's gaining speed right now, and whether it's worth your attention. 6 accelerating AI repos earned today's slot.
Top mover
ZCode is a coding agent harness from Z.ai (the GLM model makers), bundling the scaffolding around an LLM to read repos, edit files, and run commands. It's gaining traction because it ships as a self-hostable alternative to closed agent CLIs, letting teams point it at their own model endpoints. The caveat: it's tightly coupled to Z.ai's GLM models, so value depends on whether you're already in that ecosystem.
🤖 Agents & automation
A collection of prompt/skill definitions targeting Chinese-language job-search workflows (resume rewriting, application submission, interview prep) plus general dev productivity. It is gaining traction because it packages these as reusable 'Skills' rather than one-off prompts, and the job-market angle resonates with a large Chinese developer audience. The caveat: it is essentially curated prompt templates and scripts, not a novel runtime or model, so value depends heavily on the underlying LLM and the quality of each skill's instructions.
Cumora is a cross-platform team chat app where AI agents are first-class participants, not bots bolted onto a sidebar. It supports cloud-hosted brains or bring-your-own via Claude Code and Codex, letting teams share agent context inside normal conversation. The 3.8k-star traction reflects real demand for multi-agent collaboration surfaces, but it's early and the value depends heavily on how well BYO-agent orchestration holds up under concurrent team use.
A Chinese-language practical playbook for building agent workflows on WorkBuddy, covering Skills, MCP integrations, automation, and multi-agent patterns with real examples rather than API docs. It's gaining traction because teams adopting agent frameworks need concrete recipes for tool wiring and orchestration, and this fills that gap for the WorkBuddy ecosystem specifically. The caveat: it's a guide, not a library, so its value depends entirely on WorkBuddy's continued relevance and the recipes staying current as the platform's APIs shift.
NVIDIA NeMo's labs-OO-Agents lets you define agents as Python classes with typed methods and state, rather than chaining prompt strings or graph nodes. It's gaining traction because it maps agent design onto familiar OOP patterns (inheritance, composition, type hints), which reduces the boilerplate and debugging pain of graph-based frameworks. Caveat: it's a 'labs' project from NVIDIA, so APIs may churn and it's tightly coupled to NeMo/NIM ecosystem assumptions.
⚙️ Inference & serving
ninfer is a focused inference runtime that hand-optimizes kernels for specific model checkpoints on specific GPUs, rather than chasing broad hardware coverage like vLLM or llama.cpp. That narrow scope can yield real latency and memory wins on the exact model/GPU pairs it supports, which is why it's drawing attention from people running single-card deployments. The trade-off is coverage: if your checkpoint or GPU isn't on the supported list, you get nothing, and per-pair tuning means slower onboarding of new models.