Here is today’s look at what developers are starring on GitHub, plus a note on the GitHub Changelog. On 22 November 2026 the daily language lists are full of open-source coding agents and agent frameworks, with a couple of established open-source CRMs and a browser-based previs tool for some variety.
Star counts below are as shown on the daily trending pages for Python, JavaScript, TypeScript and PHP on the evening of 22 November 2026 (IST). They change quickly, so treat them as a snapshot. Repos featured in our last few roundups are skipped, which is why some of the top names on the main GitHub Trending page aren’t repeated here.
Trending repositories
1. anomalyco/opencode
What it does: “The open source coding agent.” OpenCode is an AI coding agent you run from your terminal against your own codebase.
- Language: TypeScript
- Stars: 228,539 (187 today)
Why it matters: it has the highest total star count of any repo on today’s TypeScript list. If you’re comparing AI coding agents, an open-source one lets you read exactly what it does with your code and which model provider it calls, which is useful for teams with strict security reviews.
2. earendil-works/pi
What it does: “AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI.” The README describes Pi as “a minimal, extensible agent harness that you can make your own”, customised with extensions, skills, prompt templates and themes.
- Language: TypeScript
- Stars: 123,308 (163 today)
Why it matters: Pi deliberately skips features like sub-agents and plan mode and lets you add what you need. You can use it interactively, in print or JSON mode for scripts, over RPC, or through its TypeScript SDK. It requires Node.js 22.19 or newer.
3. twentyhq/twenty
What it does: “The open alternative to Salesforce, designed for AI.” Twenty is an open-source CRM that you can self-host or use in the cloud.
- Language: TypeScript
- Stars: 61,323 (44 today)
Why it matters: the README pitches it as “the CRM you build, ship, and version like the rest of your stack”. You scaffold an app with the twenty CLI and define objects, fields and views as TypeScript code, which suits teams that want CRM customisations in Git rather than in admin screens.
4. agentscope-ai/agentscope
What it does: “Build and run agents you can see, understand and trust.” AgentScope is a Python framework for building LLM-powered agents and multi-agent applications.
- Language: Python
- Stars: 36,473 (48 today)
Why it matters: its focus is on being able to see and understand what agents are doing, which is the part that usually hurts when agent prototypes move to production. It’s one of the more established Python options and has its own documentation site and roadmap.
5. mastra-ai/mastra
What it does: “The modern TypeScript framework for AI-powered applications and agents.” Mastra covers model routing across many providers, agents, graph-based workflows, human-in-the-loop suspend and resume, memory and RAG, MCP servers, evals and observability.
- Language: TypeScript
- Stars: 29,734 (18 today)
Why it matters: if your stack is React, Next.js or Node, Mastra keeps the whole agent pipeline in TypeScript. Workflows use .then(), .branch() and .parallel(), and you can start a project with npm create mastra@latest.
6. krayin/laravel-crm
What it does: “Krayin CRM is Free & Open Source CRM Built with Laravel for Customer, Lead, and Sales Management.” It’s built with Laravel and Vue.js.
- Language: PHP
- Stars: 24,193 (4 today)
Why it matters: for Laravel developers it’s a large, real-world codebase to learn from, and a practical starting point if a client needs a self-hosted CRM. The project has a modular structure, custom attributes and Docker installation docs.
7. awslabs/cli-agent-orchestrator
What it does: “Multi-agent orchestration for AI coding CLIs — Claude Code, Kiro, Codex, and more, coordinated in isolated tmux sessions.” CLI Agent Orchestrator (CAO) runs a local cao-server and gives a supervisor agent tools to delegate work to specialist agents.
- Language: Python
- Stars: 1,628 (11 today)
Why it matters: instead of wrapping each tool in a new API, CAO keeps the agents as full CLI processes with their native authentication and features. That’s an interesting design if your team already uses several coding CLIs and wants them to work on a task in parallel or in sequence.
8. HaD0Yun/CozyClay
What it does: “Open source previs software in the browser: block a scene, pose characters, author camera moves and cuts, then take the same shots to an AI video model.”
- Language: JavaScript
- Stars: 2,073 (12 today)
Why it matters: the README makes the point that AI video costs money per take and that a text prompt can’t hold a whole scene. CozyClay lets you decide positions, movement and camera cuts in 3D first, then hands the video model a first frame, a greybox reference clip and a prompt. It’s also a neat example of a substantial 3D app running in a browser tab.
Quick comparison
| Repo | Language | Stars (today) | Good for |
|---|---|---|---|
| anomalyco/opencode | TypeScript | 228,539 (187) | Open-source coding agent |
| earendil-works/pi | TypeScript | 123,308 (163) | Minimal, extensible agent harness |
| twentyhq/twenty | TypeScript | 61,323 (44) | CRM defined as code |
| agentscope-ai/agentscope | Python | 36,473 (48) | Python agent framework |
| mastra-ai/mastra | TypeScript | 29,734 (18) | TypeScript AI apps and workflows |
| krayin/laravel-crm | PHP | 24,193 (4) | Laravel-based CRM |
| awslabs/cli-agent-orchestrator | Python | 1,628 (11) | Coordinating coding CLIs |
| HaD0Yun/CozyClay | JavaScript | 2,073 (12) | Browser previs for AI video |
From the GitHub Changelog
There were no new entries on the GitHub Changelog on Saturday 21 or Sunday 22 November 2026, as of this evening (IST). The most recent posts are from Friday 20 November, which we covered in yesterday’s roundup. A quick recap in case you missed them:
- Security advisory permissions are now generally available: organisation owners can grant advisory work, including handling private vulnerability reports, through custom repository roles, without making people repository admins.
- GitHub Enterprise Server 3.23 release candidate: includes AI secret detection that runs on the appliance itself (public preview) and a default three-day Dependabot cooldown before version update pull requests.
- Pin your favorite repositories to your dashboard: a Favorite repositories section on the new dashboard, synced with pinned repositories in GitHub Mobile.
If you run GHES, the 3.23 RC is the one to read: a release candidate is the right time to test upgrades in staging.
Takeaway
Open-source agents dominate today’s lists: OpenCode and Pi for coding in the terminal, AgentScope and Mastra for building your own agents in Python or TypeScript, and CAO for coordinating several coding CLIs at once. If you’re picking a framework, try two of them on the same small task and compare how easy it is to see what the agent actually did. That visibility matters more in production than any feature list.
