GitHub Weekly: Top 10 Trending Repos (September 21 – September 27, 2026)

TL;DR: This week’s trending list skews heavily toward AI agent tooling — coding agents, agent memory, agent orchestration, and agent-native CLIs all show up more than once. There’s also a reminder that the old guard (Next.js, PyTorch) never really leaves the list. If you build with or around AI agents, there’s a lot here worth a look.

This week we’re sticking with the classic format: the overall Top 10 trending repos on GitHub, ranked by the platform’s own weekly trending algorithm, for September 21 through September 27, 2026. No angle filter this time — just whatever the broader developer community starred the most. As you’ll see below, that turned out to mean a lot of agent tooling.

#1. anthropics/financial-services

github.com/anthropics/financial-services — Python, 37,916 stars

This is a repo from Anthropic (the company behind Claude) aimed at the financial services industry. Repos like this typically package up pre-built “skills” or workflows — things like data analysis templates, report generation, or domain-specific prompting patterns — so an AI agent can handle finance-specific tasks instead of starting from a blank slate every time.

Good for: Fintech engineering teams and quant or data teams looking to give Claude (or another agent) a head start on finance-specific workflows without building all the scaffolding themselves.

#2. paperclipai/paperclip

github.com/paperclipai/paperclip — TypeScript, 90,738 stars

Paperclip bills itself as “the open-source app everyone uses to manage agents at work.” In plain terms: as more teams run AI agents alongside human employees for real tasks, someone needs a dashboard to track what those agents are doing, assign them work, and keep an eye on results — that’s what this project provides.

Good for: Teams already running multiple AI agents in production who need one place to coordinate and monitor them, instead of juggling separate terminals or ad hoc scripts.

#3. vectorize-io/hindsight

github.com/vectorize-io/hindsight — Python, 38,377 stars

Hindsight describes itself as “agent memory that learns.” Most AI agents forget everything the moment a session ends — Hindsight is built to give agents a persistent, evolving memory across sessions, so they can build on past interactions instead of starting cold every time.

Good for: Anyone building long-running AI agents or assistants — customer support bots, coding agents, research assistants — that need to remember user preferences or prior context across sessions.

#4. cloudflare/security-audit-skill

github.com/cloudflare/security-audit-skill — JavaScript, 22,439 stars

This is a “skill” (a packaged capability) for coding agents that walks through a multi-phase security audit and produces findings that are independently verified and machine-readable, rather than a single pass of “looks fine to me.” That verification step matters — a lot of AI-generated security reviews produce false positives or miss real issues, so building in a cross-check phase is a meaningful design choice.

Good for: Security-conscious teams already using AI coding agents who want a more rigorous, structured audit process instead of a one-shot “review this code” prompt.

#5. Tencent/WeKnora

github.com/Tencent/WeKnora — Go, 30,736 stars

WeKnora turns raw documents into a queryable knowledge base — what’s usually called RAG, or retrieval-augmented generation, meaning the system looks up relevant chunks of your documents before answering a question instead of relying purely on what the model already knows. On top of that it adds an autonomous reasoning agent and a self-maintaining wiki, so the knowledge base can update and organize itself over time rather than needing constant manual curation.

Good for: Teams building internal knowledge bases or support chatbots over large document sets who want the system to stay current without someone continuously re-indexing it by hand.

#6. davila7/claude-code-templates

github.com/davila7/claude-code-templates — Python, 32,010 stars

A CLI tool for configuring and monitoring Claude Code, Anthropic’s command-line coding agent. Instead of hand-writing configuration files and guessing at settings, it gives you ready-made templates and a monitoring view so you can see what your coding agent is actually doing across a project.

Good for: Developers already using Claude Code day to day who want faster setup and better visibility into agent activity, especially across multiple projects.

#7. stablyai/orca

github.com/stablyai/orca — TypeScript, 79,940 stars

Orca describes itself as an ADE (agent development environment) for running a whole fleet of coding agents in parallel — Claude Code, Codex, Cursor, and others — using your own existing subscriptions rather than a separate paid platform. It’s available on desktop, mobile, and remote runtimes, which suggests the goal is to let you kick off and check on agent work from wherever you are, not just from a single terminal.

Good for: Developers running several coding agents at once on different tasks or branches who want one interface to orchestrate and check in on all of them.

#8. vercel/next.js

github.com/vercel/next.js — JavaScript, 142,817 stars

Next.js needs little introduction at this point — it’s the widely used React framework for building server-rendered and statically generated web apps. Its continued presence on the weekly trending list, years after its initial release, is really a testament to how much of the ecosystem is still built directly on top of it, with regular releases keeping it in front of new and returning developers alike.

Good for: Teams building React-based web applications who want routing, server rendering, and static site generation wired together out of the box, instead of assembling those pieces themselves.

#9. HKUDS/CLI-Anything

github.com/HKUDS/CLI-Anything — Python, 50,794 stars

CLI-Anything’s stated goal is “making all software agent-native.” In practice, that means wrapping existing command-line tools in a layer that lets AI agents discover and call them more reliably, rather than each agent needing custom, hand-written integration code for every tool it might use.

Good for: Teams building AI agents that need to interact with a wide range of existing CLI tools and don’t want to write bespoke integration glue for each one.

#10. pytorch/pytorch

github.com/pytorch/pytorch — Python, 103,433 stars

PyTorch is the widely used deep learning framework — it gives you tensors (multi-dimensional arrays optimized for math on a GPU) and the tools to build and train neural networks on top of them. Like Next.js, its regular reappearance on trending lists reflects how much of the AI and machine learning world — including several of the agent projects earlier in this list — is quietly built on top of it.

Good for: Machine learning engineers and researchers training or fine-tuning models, and anyone building AI tooling that needs a solid underlying framework instead of starting from scratch.

Put together, this week’s list is a pretty clear signal of where developer attention is right now: agent memory, agent orchestration, agent-native tooling, and the frameworks (Next.js, PyTorch) that everything else still quietly runs on top of. If you’re building anything agent-related, Hindsight, Orca, and CLI-Anything in particular are worth a closer look. We’ll see what floats to the top next week.