Pulse Weekly #1

Created: 2026-04-07 | Updated: 2026-04-07 | 6 min read

I went through GitHub Trending across multiple programming languages (Go, Python, Rust, TypeScript, JavaScript, Java, C++, C, Shell, Kotlin, Swift) at three timescales: daily, weekly, and monthly. Here’s what stands out when you cross-reference all three.

Note #1 - Re: Data: GitHub Trending is a live, constantly changing page. The data below was captured at a specific point in time on April 7, 2026. If you check the same page later, results may differ. “n/a” in a table means the repo didn’t appear on that particular trending page at capture time. All numbers represent star growth for the stated period.

Note #2: This post is originally intended to be a one-time thing, as I was genuinely interested in “what’s trending” and found some interesting stuff on GitHub. I used to do it regularly from various sources and decided to do it here for people who may share similar interests to mine and also for me, obviously, for future reference. And if this works well, I may do it more often. Unfortunately, I don’t have a way to measure visits (I don’t use web analytics), so my only metric is how painful or time-consuming the process is to write this post and how enjoyable it is.

#The Big Picture

AI coding agents and their ecosystems account for about a third of everything trending. Beyond AI: open-source alternatives to paid tools, on-device ML infrastructure, and a Telegram client scene with 5+ clients trending across C++, Java, Swift, and Kotlin.

#1. The Claude Code Ecosystem

Six Claude Code ecosystem repos are trending at once, not just Claude Code itself, but companion tools, skills collections, and agent directories built around it.

Two ecosystem repos, everything-claude-code (143K) and superpowers (138K), have more stars than Claude Code itself (110K). Skills repos like superpowers and mattpocock/skills are collections of reusable configuration files that extend Claude Code for specific tasks. agency-agents (73K) is a directory of pre-built specialized expert agents. awesome-claude-code-subagents catalogs 100+ subagents.

#2. Three AI Agents, Three Philosophies

Three AI coding agents show up across all three time periods with non-trivial growth. They all do similar things but are built very differently:

All three are terminal-based AI agents that read your codebase, run commands, and edit files.

Claude Code: Claude-only. Ships as CLI + VS Code + JetBrains + desktop + web + mobile. Exposes an Agent SDK and automates GitHub PRs via Actions. 110K stars, 51 contributors (all Anthropic).

Goose: 15+ LLM providers, 70+ MCP extensions, supports custom distributions. Moved this week from Block to the AAIF under the Linux Foundation (platinum members: AWS, Anthropic, Google, Microsoft, OpenAI). 38K stars, 439 contributors.

Hermes: built by Nous Research, connects to 14+ messaging platforms, runs on 6 backends (local/Docker/SSH/Daytona/Singularity/Modal). Integrates with Atropos for RL training, their tool-calling specialist improved parallel tasks from 10% to 46%. 29K stars, 303 contributors, 7 releases in 6 weeks.

#3. Google’s On-Device AI Infrastructure

Google has four on-device AI repos trending simultaneously:

At the same time, Apple Silicon ML repos are trending too:

LiteRT-LM is the successor to TensorFlow Lite for running language models on-device. The gallery app recently added Gemma 4 support and is gaining +1,100 stars per day, 13 contributors, all Google. On the Apple side, mlx-vlm (+277/day) brings vision-language models to Apple Silicon.

#4. openscreen: Open-Source Screen Recording

Not everything trending is AI. siddharthvaddem/openscreen is an open-source alternative to Screen Studio ($29/month) and it’s the second highest daily gainer overall:

At 24K stars with +15,791 monthly growth. Built in TypeScript (97.8%), 42 contributors, 11 releases. The creator describes themselves as “new to open source”. Other open-source alternatives to paid tools: Rectangle (vs Magnet), OBS (vs commercial streaming tools), Bruno (vs Postman).

#5. Rust in AI Infrastructure

Six Rust repos related to AI infrastructure are trending across the three time periods. All are focused on agent tooling, not model training:

Non-AI Rust repos are also trending: tokio (+13/day), biome (+10/day), RustPython (+23/day).

#6. Memory for AI Agents

Six repos focused on persistent memory for AI agents are trending across the three time periods. LLMs don’t remember anything between sessions by default, these projects try to fix that in different ways:

Memvid encodes memory into video frames and retrieves via semantic search over frame metadata. Onyx is the largest at 25K stars, it indexes enterprise data (Slack, Confluence, Google Drive) for AI search. Honcho is also used by Hermes (section 2) for persistent user modeling across sessions.

#Quick Hits

Things worth noting but not deep-diving this week:

• Telegram client arms race, 5+ Telegram clients trending simultaneously across C++, Java, Swift, and Kotlin: tdesktop (+187/day), Telegram-iOS (+25/day), DrKLO/Telegram, Telegram-X, piko.

• Android root/mod tools, KernelSU (+39/day), Magisk (+32/day), APatch (+8/day), Shizuku (+37/day). Four root/mod tools trending at the same time.

• Game engines and frameworks, love2d (+211/day), JoltPhysics (+128/day), luanti (+68/day), raylib (+19/day). Four game dev repos trending at the same time.

• capcom6/android-sms-gateway, SMS Gateway API on Android, +697 stars per day.

• freeCodeCamp, Still gaining +700/day.

#My Take

AI dominates the trending pages, about a third of everything this week. No surprise there. But the two things I keep thinking about are the non-AI signals.

Open-source alternatives to paid tools keep showing up and going viral. openscreen hit 24K stars in weeks. Screen Studio costs $29/month, openscreen is free. Same thing happened with Bruno vs Postman, Rectangle vs Magnet. Every time a paid dev tool hits a certain price, someone builds a free version and it explodes.

Then there’s Google’s on-device AI stuff. The gallery app is doing +1,100 stars per day with 13 contributors, all Google employees (apparently). No tweets going viral, no community hype. They shipped Gemma 4 support and the numbers just went up. LiteRT-LM is doing +483/day, that’s the successor to TensorFlow Lite, but for language models on phones. Google is quietly building the runtime for LLMs that don’t need the cloud, and almost nobody is talking about it.

If this was useful, consider buying me a coffee :)