GitHub Trending Weekly 2026-05-13: AI Agent Toolchain Goes Official, antirez's C Comeback, Universal Linux LPE Resurfaces
Data window: 2026-05-05 to 2026-05-13 (rolling 7 days) Sources: GitHub Trending weekly + monthly, GitHub Search API, HN Algolia
TL;DR: The week's biggest surprise was antirez (Redis creator) topping the new repo chart with a pure C DeepSeek 4 local inference engine (HN 497 points, 157 comments). Growth champion was a Rust-based DeepSeek TUI at +21K stars. Anthropic's official financial-services repo gained 12K+ stars. The other big HN story was V4bel/dirtyfrag exposing a universal Linux local privilege escalation, with the main thread reaching HN 816 points and 331 comments. Sustained monthly chart presence for agent-skills and TradingAgents signals ongoing momentum.
Fastest Growing: Weekly Star Gains Top 15
Source:
github.com/trending?since=weeklyMarker: 🔁 = also on the monthly chart (sustained momentum signal)
| # | Project | +Stars/wk | Total Stars | Language | Created |
|---|---|---|---|---|---|
| #1 | Hmbown/DeepSeek-TUI | +21,752 | 26,402 | Rust | 2026-01-19 |
| #2 🔁 | anthropics/financial-services | +12,088 | 21,452 | Python | 2026-02-23 |
| #3 🔁 | addyosmani/agent-skills | +11,725 | 40,363 | Shell | 2026-02-15 |
| #4 | ruvnet/ruflo | +8,660 | 49,713 | TypeScript | 2025-06-02 |
| #5 🔁 | TauricResearch/TradingAgents | +7,259 | 74,383 | Python | 2024-12-28 |
| #6 | CloakHQ/CloakBrowser | +5,449 | 7,742 | Python | 2026-02-22 |
| #7 | VectifyAI/PageIndex | +4,555 | 30,841 | Python | 2025-04-01 |
| #8 🔁 | AIDC-AI/Pixelle-Video | +4,480 | 15,596 | Python | 2025-11-07 |
| #9 | decolua/9router | +4,263 | 9,316 | JavaScript | 2026-01-05 |
| #10 | docusealco/docuseal | +3,537 | 16,451 | Ruby | 2023-07-03 |
| #11 | bytedance/UI-TARS-desktop | +3,211 | 33,509 | TypeScript | 2025-01-19 |
| #12 | LearningCircuit/local-deep-research | +2,449 | 7,362 | Python | 2025-02-09 |
| #13 | rohitg00/agentmemory | +2,291 | 5,768 | TypeScript | 2026-02-25 |
| #14 | playcanvas/supersplat | +2,164 | 7,682 | TypeScript | 2023-10-19 |
| #15 | HKUDS/AI-Trader | +2,132 | 16,557 | Python | 2025-10-23 |
Top New Repos: Born This Week, Top 15
Source: GitHub Search API (
created:2026-05-05..2026-05-13, sorted by total stars)
| # | Project | Total Stars | Language | Created |
|---|---|---|---|---|
| #1 | antirez/ds4 | 8,056 | C | 2026-05-06 |
| #2 | V4bel/dirtyfrag | 4,318 | C | 2026-05-07 |
| #3 | vercel-labs/zero-native | 2,909 | Zig | 2026-05-08 |
| #4 | strukto-ai/mirage | 2,056 | TypeScript | 2026-05-06 |
| #5 | yaojingang/yao-open-prompts | 1,824 | Python | 2026-05-06 |
| #6 | XBuilderLAB/cheat-on-content | 1,801 | Shell | 2026-05-05 |
| #7 | huangserva/3DCellForge | 1,700 | JavaScript | 2026-05-10 |
| #8 | BigPizzaV3/CodexPlusPlus | 1,497 | Python | 2026-05-06 |
| #9 | zarazhangrui/beautiful-html-templates | 1,018 | HTML | 2026-05-05 |
| #10 | lightseekorg/tokenspeed | 974 | Python | 2026-05-06 |
| #11 | FULU-Foundation/OrcaSlicer-bambulab | 831 | C++ | 2026-05-11 |
| #12 | pixel-point/media-downloader | 594 | Swift | 2026-05-06 |
| #13 | haydenbleasel/files-sdk | 560 | TypeScript | 2026-05-08 |
| #14 | kitft/natural_language_autoencoders | 537 | Python | 2026-05-05 |
| #15 | thakur-works/DarkGPT | 532 | — | 2026-05-10 |
This Week's Highlights: Fastest Growing Top 15
#1 Hmbown/DeepSeek-TUI: A Terminal Coding Agent for DeepSeek
Coding agent for DeepSeek models that runs in your terminal
+21,752 stars this week | 26,402 total | Rust | MIT
This week's growth champion is a Rust-based terminal coding agent for DeepSeek. A May 6 HN post mentioning it ("Terminal coding agent for DeepSeek V4") got only 3 points, yet the repo surged 21K stars. The pattern: growth was driven by DeepSeek V4's launch event, not community discussion. With 2,213 forks and 367 open issues, adoption is real but issues are accumulating fast.
The key question: does your workflow already depend heavily on terminal operations? If yes, worth a try. If you live in VS Code or Cursor, other options may fit better.
#2 🔁 anthropics/financial-services: Anthropic's Official Financial Services SDK
Claude for Financial Services
+12,088 stars this week | 21,452 total | Python | Apache-2.0
On both the weekly and monthly charts (🔁 sustained momentum). Created in late February, the repo only started surging in early May, likely tied to financial institutions evaluating Claude's enterprise offerings. The 2,887 forks indicate institutional-level interest. Apache-2.0 licensing is a deliberate choice enabling commercial use.
If you work in fintech, this repo is worth studying closely. Not just the code, but how Anthropic defines agent boundaries and compliance logic for financial scenarios.
#3 🔁 addyosmani/agent-skills: Engineering-Grade AI Agent Skills
Production-grade engineering skills for AI coding agents.
+11,725 stars this week | 40,363 total | Shell | MIT
Monthly #8, weekly #3. Addy Osmani (formerly Google Chrome engineering) curated this agent skills library covering Claude Code, Cursor, Antigravity IDE, and more. It's positioned closer to the ground than official docs: not teaching API usage, but providing drop-in engineering skill instructions. With 4,446 forks, this content is being heavily customized.
The core question it answers: "How do I make AI agents work to engineering standards?" rather than "What can AI agents do?" That perspective shift is one of this week's most important signals.
#4 ruvnet/ruflo: Claude Agent Orchestration Platform
The leading agent orchestration platform for Claude. Deploy intelligent multi-agent swarms, coordinate autonomous workflows...
+8,660 stars this week | 49,713 total | TypeScript | MIT
A nearly 50K-star agent orchestration platform supporting multi-agent swarms, RAG, Claude Code and Codex integration. The 540 open issues are a warning sign. On HN, someone asked "Is anyone using ruflo?" (1 point, 1 comment). The disconnect between star count and community discussion usually means many watchers but few committed users.
#5 🔁 TauricResearch/TradingAgents: Multi-Agent LLM Financial Trading
TradingAgents: Multi-Agents LLM Financial Trading Framework
+7,259 stars this week | 74,383 total | Python | Apache-2.0
The highest total star count on this week's chart at 74,383. On both weekly and monthly charts (🔁).
Risk disclosure: This is a research framework, not a live trading tool. The 14,508 forks indicate massive developer interest in multi-agent financial patterns, but using LLM agents for live trading carries significant risks: hallucinations, latency, and black swan event handling. No real-world deployment cases have surfaced on HN yet.
#6 CloakHQ/CloakBrowser: Stealth Browser Passing All Bot Detection
Stealth Chromium that passes every bot detection test. Drop-in Playwright replacement with source-level fingerprint patches. 30/30 tests passed.
+5,449 stars this week | 7,742 total | Python | MIT
A direct Playwright replacement with source-level patches for Cloudflare, reCAPTCHA, and other detection systems. "30/30 tests passed" is a meaningful differentiator. Use cases range from legitimate web scraping and competitive monitoring to gray-area bot operations. Evaluate legality before deploying.
#7 VectifyAI/PageIndex: Vectorless, Reasoning-Based RAG
PageIndex: Document Index for Vectorless, Reasoning-based RAG
+4,555 stars this week | 30,841 total | Python | MIT
The core claim: you don't need a vector database; reasoning can replace vector search for RAG. For developers used to Pinecone or Chroma, this is an alternative worth serious evaluation. Its sweet spot is document indexing and targeted knowledge base queries, not general-purpose RAG. If your RAG problems center on "the vector search returns wrong documents" rather than "it's too slow," this approach may help.
#8 🔁 AIDC-AI/Pixelle-Video: Fully Automated Short Video Engine
AI Fully Automated Short Video Engine
+4,480 stars this week | 15,596 total | Python | Apache-2.0
On both weekly and monthly charts. Integrates ComfyUI, TTS, and image generation into a fully automated short video pipeline. The 2,248 forks signal strong engagement, particularly from the Chinese developer community. The problem it solves: full automation from text prompt to finished short video.
#9 decolua/9router: A Free LLM Router That Surged This Week
Unlimited FREE AI coding. Connect Claude Code, Codex, Cursor, Cline, Copilot, Antigravity to FREE Claude/GPT/Gemini via 40+ providers. Auto-fallback, RTK -40% tokens, never hit limits.
+4,263 stars this week | 9,316 total | JavaScript | MIT
9router connects developer tools (Claude Code, Codex, Cursor, and others) through a proxy layer to 40+ free or low-cost LLM endpoints, with auto-fallback and an RTK compression scheme that claims 40% token savings. The +4,263 star surge reflects a real developer pain point: stacked subscription costs across multiple AI services plus the uncertainty of per-vendor rate limits.
Assessment: 9router does not currently have a corresponding standalone HN discussion this week, so social-proof signals come mainly from GitHub star velocity and the project's own README. For any tool that sits in the path of your API keys, it's worth checking: whether requests transit through a third-party server, the health of its npm dependency tree (this week's TanStack NPM supply-chain incident at HN 1,075 points and 453 comments illustrates that risk well), and whether each provider's ToS allows proxy access. "Free" at the tool layer means you don't pay the API vendor directly. It does not mean there is no trust or security cost.
#10 docusealco/docuseal: Open Source DocuSign Alternative
Open source DocuSign alternative. Create, fill, and sign digital documents
+3,537 stars this week | 16,451 total | Ruby | AGPL-3.0
An established repo from 2023, back on the chart this week. With DocuSign's expensive pricing, docuseal offers a self-hostable alternative. Ruby on Rails stack with AGPL-3.0 license, suitable for businesses or individual developers needing full control over document signing workflows.
#11 bytedance/UI-TARS-desktop: ByteDance's Open Source Multimodal Agent Desktop
The Open-Source Multimodal AI Agent Stack: Connecting Cutting-Edge AI Models and Agent Infra
+3,211 stars this week | 33,509 total | TypeScript | Apache-2.0
ByteDance's UI-TARS desktop edition integrating computer-use, browser automation, and MCP server capabilities. At 33,509 stars and 3,323 forks, it's a seriously evaluated option for enterprise-level agent infrastructure.
#12 LearningCircuit/local-deep-research: Local and Encrypted Deep Research
~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs. 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
+2,449 stars this week | 7,362 total | Python | MIT
A fully local, fully encrypted deep research tool supporting arXiv, PubMed, private documents, and local LLMs (llama.cpp, Ollama). Note the ~95% SimpleQA accuracy claim comes with a caveat: it's under specific hardware (Qwen3.6-27B on a 3090). Best suited for academic researchers or enterprise teams who need data to never leave the machine.
#13 rohitg00/agentmemory: Persistent Memory for AI Coding Agents
#1 Persistent memory for AI coding agents based on real-world benchmarks
+2,291 stars this week | 5,768 total | TypeScript | Apache-2.0
Provides cross-session persistent memory for AI coding agents, supporting Claude Code, Cursor, Copilot, and Codex. It addresses a real pain point: every new session means agents lose all context.
#14 playcanvas/supersplat: Browser-Based 3D Gaussian Splat Editor
3D Gaussian Splat Editor
+2,164 stars this week | 7,682 total | TypeScript | MIT
A WebGL/WebGPU browser-based editor for 3D Gaussian Splatting. This is the only repo in this week's top 15 completely unrelated to AI agents, representing independent momentum in the 3D visualization vertical.
#15 HKUDS/AI-Trader: 100% Fully-Automated Agent-Native Trading
AI-Trader: 100% Fully-Automated Agent-Native Trading
+2,132 stars this week | 16,557 total | Python
Risk disclosure: The name and description are aggressive ("100% Fully-Automated"), but the repo has no license information, which is a red flag for financial tools. At 16,557 stars and 2,609 forks, there's significant attention. Evaluate carefully whether actual capabilities match the claims, and whether any live testing data exists.
Highlights: Top New Repos (Top 10)
New #1 antirez/ds4: Redis Creator's DeepSeek 4 Local Inference Engine
DeepSeek 4 Flash local inference engine for Metal and CUDA
8,056 stars | C | MIT | Created: 2026-05-06
This week's standout new repo. antirez, the original creator of Redis, built a local inference engine for DeepSeek 4 Flash in pure C, supporting Apple Silicon Metal and CUDA.
HN 496 points with 157 comments makes this the highest single-post HN score among all repos this week. The discussion centers on a renowned veteran developer choosing C, an "unfashionable" language, to build a performance-focused inference engine from scratch. In an AI ecosystem dominated by Python frameworks, someone going back to the fundamentals is itself a signal.
630 forks accumulated in one week, showing developers are actively studying and porting this engine.
New #2 V4bel/dirtyfrag: Universal Linux Local Privilege Escalation PoC
Universal Linux LPE (Local Privilege Escalation)
4,318 stars | C | Created: 2026-05-07
The main HN thread, Dirty Frag: Universal Linux LPE, accumulated 816 points and 331 comments — the second-highest HN buzz among this week's new repos. The name suggests lineage with Dirty COW and Dirty Pipe — a proof-of-concept local privilege escalation vulnerability.
Security note: Valuable as a reference for security researchers, but requires thorough understanding of technical and legal boundaries before any production use. The 641 forks (unusually high relative to 4K stars) indicate active security community research into this vulnerability.
New #3 vercel-labs/zero-native: Desktop + Mobile Apps with Zig and Web UI
Build desktop + mobile apps with Zig and web UI
2,909 stars | Zig | Apache-2.0 | Created: 2026-05-08
A cross-platform framework from Vercel Labs using Zig for the native layer and web technologies for UI. The combination is counterintuitive: Zig is a systems language (lower-level than Rust), but Vercel chose it for minimal binary sizes and precise memory control. HN discussion is still low, but the Vercel brand drove rapid star accumulation. Worth watching.
New #4 strukto-ai/mirage: Unified Virtual Filesystem for AI Agents
A Unified Virtual Filesystem For AI Agents
2,056 stars | TypeScript | Apache-2.0 | Created: 2026-05-06
Solves agent sandboxing: gives AI agents a virtual filesystem for safe read/write operations without touching the real filesystem. Supports Claude Code, LangChain, and OpenAI Agents. Agent sandbox isolation is a core enterprise deployment requirement that still lacks a standard solution. This direction has potential.
New #5 yaojingang/yao-open-prompts: Chinese AI Prompt Library
A comprehensive Chinese AI prompt library covering work, learning, content, marketing, and daily life scenarios
1,824 stars | Python | Created: 2026-05-06
A prompt engineering resource library for Chinese-speaking users covering work, learning, content creation, and marketing. With 283 forks, people are customizing their own versions.
New #6 XBuilderLAB/cheat-on-content: Turn Every Post Into a Calibrated Experiment
A workflow that turns every post into a calibrated experiment: score, blind-predict, retro, evolve.
1,801 stars | Shell | MIT | Created: 2026-05-05
Not a content generator but a "post-as-experiment" workflow: predict performance before publishing, review against actuals afterward, iterate. The 345 forks suggest meaningful adoption.
New #7 huangserva/3DCellForge: AI-Powered 3D Cell Generation Studio
AI-powered interactive 3D cell generation and exploration studio.
1,700 stars | JavaScript | MIT | Created: 2026-05-10
Accumulated 1,700 stars in just 3 days. Focused on 3D cell visualization for biological sciences. The 292 forks indicate rapid academic adoption.
New #8 BigPizzaV3/CodexPlusPlus: Enhanced Tool for Codex App
An enhanced tool for CodexApp, striving to make Codex better to use and more comfortable
1,497 stars | Python | Created: 2026-05-06
A third-party enhancement tool for OpenAI's Codex App. The existence of such tools itself indicates that Codex App has noticeable UX pain points.
New #9 zarazhangrui/beautiful-html-templates: HTML Slide Templates for AI Agents
A library of HTML slide templates designed so any coding agent can pick the right one and produce a beautiful deck on the user's behalf, automatically.
1,018 stars | HTML | MIT | Created: 2026-05-05
Positioned as a template library "for AI agents" rather than humans. The design philosophy: let agents semantically select appropriate templates and automatically produce polished slide decks.
New #10 lightseekorg/tokenspeed: Speed-of-Light LLM Inference Engine
TokenSpeed is a speed-of-light LLM inference engine.
974 stars | Python | MIT | Created: 2026-05-06
Targets maximum-speed LLM inference on Blackwell GPUs, supporting DeepSeek, GPT series, Qwen, and Kimi models. Part of this week's broader "local/edge inference" trend.
Monthly Trend Cross-Reference
Several sustained momentum signals on this month's chart deserve attention:
forrestchang/andrej-karpathy-skills leads the monthly chart at 112,987 monthly star gains (not on the weekly chart). This is Andrej Karpathy's observations on LLM coding weaknesses compiled into a CLAUDE.md file. The sustained interest confirms "how to stop AI agents from making basic mistakes" remains the highest-attention topic.
NousResearch/hermes-agent (monthly +88,781 stars, monthly #5) and mattpocock/skills (monthly +57,314 stars, monthly #2) are both sustained performers in the agent skills category, corroborating this week's addyosmani/agent-skills momentum.
multica-ai/multica (monthly +20,178 stars, monthly #13) is in the "managed agents platform" category, organizing coding agents into trackable, task-assignable teams. This direction is appearing on the monthly chart for the first time.
This Week's Trend Insights
AI agent toolchains are going official. Three repos with sustained monthly chart presence (anthropics/financial-services, addyosmani/agent-skills, TradingAgents) reflect the same trend: AI agent use cases are moving from "personal experiments" to "enterprise evaluation." Anthropic's official financial services repo gaining 12K+ weekly stars shows institutions are seriously studying Claude's viability in regulated industries. The concept of agent skills is evolving from personal community projects to official engineering standards, and that process accelerated this month.
DeepSeek explodes on both fronts of the open-source stack. Hmbown/DeepSeek-TUI (Rust-based terminal agent) topped the weekly growth chart with +21K stars, while antirez's pure-C ds4 inference engine led the new repos chart with 8,056 stars and HN 497 points. The same underlying model spawning two independent breakout repos — one at the user-experience layer (TUI), one at the inference layer (ds4) — shows DeepSeek is no longer just a model release event. It has become one of the core drivers of the open-source tooling ecosystem.
Systems programming reasserts itself in the AI era. antirez's ds4 (pure C, HN 497 points) and V4bel/dirtyfrag (pure C, HN 816 points and 331 comments on the main thread) signal that the systems programming community still commands real attention on a Trending chart dominated by Python frameworks. dirtyfrag is the highest-HN-scoring new repo of the week — a reminder that operating-system level security is not going away under the AI wave.
