Weekly · open-source AI · week 33, 2026
WAI + 10 more trending open-source AI repos · week 33, 2026
This week's trending open-source AI on GitHub — WAI, pi-book, KADATH, and more: the newest, Hacker News talk, a subfield spotlight, and the fastest-rising, drawn out of the noise.
Newest this week
WAI Play is an AI platform that plays your web games and evaluates their quality—surfacing the experience issues automated tests never catch. Python-based, 251 stars, and clearly striking a nerve with anyone shipping interactive web builds. Check it out: https://github.com/waiterve/wai-play
View on GitHub →
pi-book is a collection of source-backed architecture notes on building an agent, every decision traced to a citation you can actually check. Refreshingly rigorous in a space full of hand-waving. 205 stars and growing
View on GitHub →
KADATH treats agents like a gene pool. The runtime breeds, evaluates, and improves autonomous agents across reproducible epochs, applying evolutionary pressure until the population converges on your optimization goal. The reproducibility angle is the standout—it makes evolutionary runs auditable instead of black-box luck. Python, 169 stars. Repo: https://github.com/i3T4AN/KADATH
View on GitHub →Talk of Hacker News
eighttrigrams/us-vs-them tackles a problem that's about to get very real: as agents edit text alongside humans, who wrote which line? Instead of littering files with markup, it derives line-level provenance straight from version history. Clojure, 51 points on HN. Clean approach to a messy future
View on GitHub →Subfield spotlight
ECC is an agent harness that layers skills, instincts, memory, and security onto the coding agents you already run, Claude Code, Codex, Opencode, Cursor, and beyond. Its research-first approach makes agents investigate before they act. 239K stars, JavaScript, open source
View on GitHub →
NousResearch/hermes-agent is a Python framework for agents that actually remember and adapt—persistent context instead of stateless prompts. At 228K stars, it's become a go-to reference for anyone building agentic systems
View on GitHub →
firecrawl (164,385 ★) is the context API for searching, scraping, and interacting with the web at scale. Instead of raw HTML, you get clean, structured data ready to drop into RAG pipelines and agents. TypeScript under the hood, and clearly a favorite—the star count speaks for itself
View on GitHub →Fastest-rising
PrimeIntellect-ai/prime-agent is a self-improving RLM agent for coding workflows and long-running autonomous tasks. The hook: instead of restarting from scratch each run, it refines its own approach over time—written in TypeScript. Up +216 stars this week. https://github.com/PrimeIntellect-ai/prime-agent
View on GitHub →
OmniRoute is a MIT-licensed AI gateway that unifies 290+ providers (90+ free) and 500+ models behind a single endpoint. Highlights: quota-aware auto-fallback, RTK+Caveman compression that saves 15-95% of tokens, MCP/A2A support, and drop-in compatibility with Claude Code, Codex, Cursor, OpenCode, Cline and Copilot. Built by 500+ contributors and up 76 stars this week
View on GitHub →
reverse-skill is a PowerShell skill router pack for reverse engineering and authorized penetration testing. It uses AI to route tasks, bootstrap the right toolchain on demand, and maintain a self-evolving knowledge base—all wired into Claude Code, Kiro, Cursor, Cline, and other AI coding clients. Up +57 stars this week
View on GitHub →Shipping hard
An open-source HTML project on GitHub. 89 commits & 14 PRs this week
View on GitHub →