Forkcast Weekly #2: openhuman, bun, RuView, UI-TARS-desktop, agentmemory, scientific-agent-skills

Episode Summary#

This episode covers six projects from personal AI assistants to scientific research tools. Features include OpenHuman’s local-first AI with 118+ integrations, the ultra-fast Bun JavaScript runtime, ByteDance’s desktop agent for GUI automation, and specialized agent skills for scientific research workflows.

Repositories Covered#

openhuman#

A local-first personal AI assistant with memory tree, Obsidian wiki integration, 118+ third-party integrations, and TokenJuice compression. Runs with Rust, React, and Tauri for a native desktop experience.

bun#

A fast all-in-one JavaScript runtime, bundler, test runner, and package manager written in Zig. Designed as a drop-in replacement for Node.js with significantly faster startup times, lower memory usage, and native TypeScript support.

RuView#

A Rust-based code review tool that provides structured, automated review of pull requests with custom rule sets. Helps teams maintain code quality standards consistently across their codebase.

UI-TARS-desktop#

ByteDance’s open-source multimodal AI agent that automates GUI interactions — it sees the screen, controls the mouse, and operates browsers and desktop applications. Features MCP tool integration and a vision-language model for understanding UI elements.

agentmemory#

A persistent memory system for AI coding agents that maintains context across sessions. Works with Claude Code, Codex, Cursor, and Copilot without client-specific dependencies.

scientific-agent-skills#

A collection of specialized AI agent skills for scientific research workflows. Covers literature review, citation graph analysis, peer review simulation, and methodology validation to accelerate academic and research processes.

superpowers#

An agentic skills framework and software development methodology that enforces TDD, spec-first workflows, git worktree isolation, and two-stage code review. With 200K+ stars, it defines how AI coding agents should work reliably across Claude Code, Codex, Cursor, and more.

CloakBrowser#

Every web scraper eventually hits Cloudflare, reCAPTCHA, or fingerprinting. Stock Playwright scores 0.1 on reCAPTCHA v3 — instantly flagged as a bot. 57 C++ source-level fingerprint patches (not JS injection). reCAPTCHA v3 score: 0.9 (human-level) vs Playwright’s 0.1. Cloudflare Turnstile: PASS. FingerprintJS: PASS. BrowserScan: NORMAL (4/4 checks). bot.incolumitas.com: 1 fail vs Playwright’s 13 fails. 30+ detection sites tested. navigator.webdriver: false (source-level patch). Humanize mode: Bezier mouse curves, per-character typing delays, realistic scroll acceleration. ~190MB RAM idle, ~280MB with 3 tabs. Drop-in Playwright replacement. Cross-platform: Linux, macOS, Windows. MIT wrapper + proprietary binary.

CLI-Anything#

Auto-generates command-line interfaces for any software, making it controllable by AI agents.

AiToEarn#

An open-source AI content marketing agent platform that automates the entire content pipeline — from creation to publishing across 14+ platforms (TikTok, YouTube, Xiaohongshu, Bilibili, and more) to engagement and monetization. Built with Next.js, Node.js, and Electron.

cal.diy#

Fully open-source, MIT-licensed scheduling infrastructure — the community fork of Cal.com without enterprise features.

supertonic#

Every TTS service is cloud-locked, expensive at scale, and fails on edge cases like financial expressions or phone numbers. You need TTS that runs offline and gets the details right. 31 languages in v3. ~99M parameters — much smaller than 0.7B-2B class open TTS systems. ONNX Runtime for cross-platform inference. Raspberry Pi demo at 0.3× RTF with zero network. Browser via onnxruntime-web (WebGPU/WASM). Handles financial expressions, phone numbers, technical units correctly where ElevenLabs, OpenAI, Gemini, and Microsoft TTS all fail. 11 inference runtimes: Python, Node.js, Browser, Java, C++, C#, Go, Swift, iOS, Rust, Flutter. Expressive tags: laugh, breath, sigh. MIT + OpenRAIL-M licensed.

hysteria#

Standard VPN protocols get throttled or blocked on restricted networks. TCP-based proxies choke on lossy connections with 5%+ packet loss, making them unusable in many regions. Customized QUIC protocol designed for unreliable and lossy networks. Traffic masquerades as standard HTTP/3 to evade censorship without collateral damage. Proxy modes: SOCKS5, HTTP Proxy, TCP/UDP Forwarding, Linux TProxy, TUN. Builds for every major platform and architecture. v2.9.1 released May 10, 2026. 81 releases, 1,310 commits. MIT licensed. Legacy v1 maintained separately. Custom authentication, traffic statistics, access control.

Open-Generative-AI#

Open-source alternative to AI video platforms with 200+ models and zero content filters.

supersplat#

Editing 3D Gaussian splats requires desktop software with GPU setup — there’s no fast browser-based tool for inspecting, cleaning, and optimizing splats before publishing Free browser-based editor, no download or install needed, WebGL + WebGPU rendering, 144 releases (v2.25.1 May 2026), 619 commits, multi-language localization support, MIT license

skills#

A collection of production-grade engineering skills for AI coding agents by Matt Pocock. Provides structured, reusable capability files that make agents more reliable at TypeScript, testing, and general software engineering tasks.

telegraf#

InfluxData’s plugin-driven server agent for collecting, processing, aggregating, and writing metrics and logs. Part of the TICK stack with 300+ plugins covering systems, cloud services, databases, and message queues.

easy-vibe#

An AI-powered programming education platform that makes learning to code more intuitive and accessible. Uses vibe-based interactions to lower the barrier for beginners entering software development.

react-doctor#

A code quality tool from Million.js that detects poor React patterns and anti-patterns in your codebase. Catches issues like unnecessary re-renders, missing memoization, and inefficient component structures before they impact production performance.

FadCam#

Open-source, ad-free Android multimedia recorder with background video recording, screen recording, and live streaming.

Kronos#

First open-source foundation model for financial candlesticks trained on 45+ global exchanges.

9router#

AI coding tools burn through paid API credits fast — you hit rate limits, get locked out, or run up a bill just from tool-call overhead 40+ API providers, 100+ models, RTK compression saves 20-40% tokens by compressing tool outputs, Caveman Mode saves up to 65% output tokens, 3-tier auto-fallback (subscription→cheap→free), OpenAI-compatible /v1 endpoint on port 20128, Node.js 20+ with Next.js 16 + React 19 + Tailwind CSS 4, 554 commits, 44 releases, MIT

Shadowbroker#

Aggregates open-source intelligence from 60+ live feeds into a single map interface with AI agent correlation.

supervision#

Reusable, model-agnostic computer vision tools so you don’t rewrite annotation, tracking, and dataset pipelines for every project.

skills#

A collection of production-grade engineering skills for AI coding agents by Matt Pocock. Provides structured, reusable capability files that make agents more reliable at TypeScript, testing, and general software engineering tasks.

agent-skills#

A production-grade engineering skills library for AI coding agents by Addy Osmani. Standardizes engineering practices and quality gates to make agents more professional and reliable across various development workflows.

LLMs-from-scratch#

LLMs feel like magic black boxes. Without understanding the internals of attention, pretraining, and finetuning, you’re just guessing at prompts. 7 chapters + 5 appendices. Implements GPT, Llama 3.2, Qwen3 (dense + MoE), Gemma 3, Gemma 4, Olmo 3, Tiny Aya from scratch. BPE tokenizer built from scratch. Covers KV Cache, GQA, MLA, Sliding Window Attention, Gated DeltaNet, MoE, FLOPs analysis, DPO alignment, LoRA (Appendix E), DDP (Appendix A). Companion 170-page PDF with 30 quiz questions per chapter. 17-hour 15-minute video course. Code runs on conventional laptops with auto GPU detection. 94.2k stars, 14.5k forks. Published 2024, ISBN 978-1633437166.

codegraph#

A pre-indexed code knowledge graph for Claude Code, Codex, Cursor, and OpenCode. Provides local semantic indexing and graph traversal that dramatically reduces file scanning and tool calls across large multi-language codebases.

n8n-mcp#

MCP server that lets Claude and other AI agents build n8n workflows covering 1,650 n8n nodes with detailed schemas.

Personal_AI_Infrastructure#

Your AI tools know nothing about your life priorities, your projects, or what matters to you. Every conversation starts from zero. Positioned as a Life Operating System. Pulse daemon on port 31337 with 22 API routes, voice, hooks, observability, cron, wiki API. Algorithm v6.3.0: 7-phase OBSERVE→THINK→PLAN→BUILD→EXECUTE→VERIFY→LEARN loop. 45 public skills, 171 workflows, 37 hooks. ISA (Ideal State Artifact) with 12 sections and 5 identities. Memory v7.6: WORK, KNOWLEDGE, LEARNING, RELATIONSHIP, OBSERVABILITY, STATE. No RAG by design — uses filesystem + ripgrep. 12 security gates on every release. MIT licensed.

agents-towards-production#

End-to-end, code-first tutorials for building production-grade GenAI agents from prototype to enterprise deployment.

video-search-and-summarization#

Reference architectures for GPU-accelerated vision agents and AI-powered video analytics with search, alerts, and summarization.

hello-agents#

A comprehensive Chinese-language tutorial series from DataWhale that teaches how to build AI agents from scratch. Covers agent architecture, tool use, memory systems, and multi-agent orchestration with hands-on examples. The most popular AI agent learning resource on GitHub.

Watch#

• Video: https://www.youtube.com/watch?v=ApSpMWdR4DY
• Cover: https://i.ytimg.com/vi/ApSpMWdR4DY/hqdefault.jpg

Notes#

Transcript and notes will be added from Forkcast output artifacts.