Pinned Tweet
Dear follower, this is my personal account and it's not a promote campaign nor SEO or news feed. Sometime may have some nude photograph i'm apologize. But i'm kind of gentle and respect no sexual harassment. Thank you. #me
2
3
Gue kira desktop app bakal makin sepi, tapi AI malah bikin kebutuhannya naik lagi.
NanoUI lagi ramai dibahas sebagai alternatif Electron buat bikin desktop app yang lebih ringan, dengan Zig di baliknya. Tapi lo nggak harus ngerti Zig dulu buat mulai, AI bisa bantu bikin prototype.
Masuk akal sih, misal makin banyak orang jalanin local agent, makin kepake juga UI desktop yang enak dipakai buat ngontrolnya.
🚨 GOOGLE JUST FOUND SOMETHING REALLY CONCERNING.
Hackers are now using AI agents to find software vulnerabilities, write exploits and execute them without human intervention.
Google discovered a framework called “exploit_agent” that could:
→ Read newly released software patches
→ Identify potential security flaws
→ Write its own exploits
→ Execute them automatically
And here’s the scary part.
Google also reported 22 zero-days exploited in August alone.
The framework contained Russian-language code comments, although the operators haven’t been identified.
We’ve been talking about AI agents doing work on their own.
Now we’re seeing what happens when that same capability is used for cyberattacks.
This is getting serious.
It's a good day! d1.2 and d1.2-ultra are on openrouter:
d1.2-ultra: openrouter.ai/liquid/d1.2-ul…
d1.2: openrouter.ai/liquid/d1.2 – at San Francisco, CA
udomsak chundang retweeted
Running several coding agents shouldn’t mean babysitting several terminals.
CompozyOS is a local control plane for builders managing coding agents such as Claude Code, Codex, and Cursor.
It helps you keep parallel agent work organized by running your existing agent CLIs as persistent sessions, showing which ones need attention, and putting approvals and scheduled check-ins in one place.
Key features:
• Persistent sessions – agent work outlives the terminal, with options to fork, rewind, or continue with another agent CLI
• Session board – groups work into Needs you, Working, Waiting, In review, and Done so you can spot the next decision
• Separate worktrees – gives parallel agents their own checkout of the same repository
• Human approvals – holds a tool call until someone allows it, bringing waiting work into Needs you
• Scheduled check-ins – always-on agents review their projects and collect reports in a main session
It’s open-source (MIT license).
Link in the reply 👇
Adding vision to a language model is more than one extra modality. With large images or long video, the vision encoder can use more compute than the language model itself, so efficient models need a different design. Our VLM team, including @rshubert98, @TianshuYu6, @sina_semnani, and @iamsamstevens, explains how they teach LFMs to see, from mid-training to the custom vision encoder built for LFM3.
udomsak chundang retweeted
Running coding agents shouldn’t mean juggling a terminal for every repo.
CodeAF is a terminal-based coding harness built for open models, for developers managing work across multiple repositories.
It helps you spend less time coordinating agents by splitting requests into tasks, running them in separate repository copies, and checking results before merging work that passes onto your working branch.
Key features:
• Multi-project workspace – see projects, conversations, running tasks, and questions waiting on you in one window
• Isolated task branches – hand off separate changes without putting every task in the same working copy
• Built-in specialists – request coding changes, pull request reviews, or local security audits from the chat
• Model choice and budgets – connect providers including OpenRouter and Ollama, assign worker/planner/checker models, and set spending limits
• Headless execution – run tasks from CI, cron, or scripts with JSON output and an exit code
It’s an early preview, so expect rough edges.
It’s open-source (Apache 2.0 license).
Link in the reply 👇
udomsak chundang retweeted
我们把几个月的心血,直接开源了
ServerOS 现已登陆 GitHub,任何人都可以免费使用、折腾,甚至把它改得更好
快来给我们点爱,或者尽情吐槽代码 👇
github.com/serveroshq/panel
udomsak chundang retweeted
We open-sourced LoopForge:a self-hosted self-evolving cloud platform for AI software delivery.
One "fix this bug" run is luck. LoopForge turns your team's delivery method into SOPs that coding agents can loop-run in the cloud:
→ Roles, not vibes. Plan / code / review / test are separate agents. The writer isn’t the reviewer.
→ Script gates + human sign-off. Builds and tests decide by exit code, not an agent’s word. Nothing ships without approval.
→ Engineering-grade RSI. Each run is scored 0–100; the replay surfaces what broke and writes improvement prescriptions back into SOPs, roles and gates — you review, then apply.
→ Bring your own agents. 10 swappable CLIs (CodeBuddy, Codex, Claude Code, Cursor…). MIT, self-hosted, your keys and code stay on your infra.
Code was never the bottleneck. The handoff is — and LoopForge catches the baton.
🔗 github.com/Tencent/LoopForge
udomsak chundang retweeted
🚨 Amazon uncovers AI agents hunting API keys across LiteLLM, Dify, and other AI apps.
Captured logs reveal the playbook: exploit weak spots → steal API keys → use those keys to power more attacks.
The campaign's reported results remain unverified.
Full story: thehackernews.com/2026/11/am…
udomsak chundang retweeted
Test your app’s UI — without adding a testing SDK to it.
Speedrun is an end-to-end test runner for builders testing Android apps, iOS apps on simulators, and websites using plain YAML.
It helps you automate user journeys by tapping, typing, scrolling, and checking what’s on screen from outside the app. Steps wait for the app to settle instead of relying on fixed sleeps.
Key features:
• YAML flows – describe actions and assertions in short, readable files instead of writing a test harness from scratch
• Mobile + web – test Android devices or emulators, iOS simulators, and websites in Chrome or Firefox
• No in-app SDK – built-in mobile agents and direct browser control avoid separate WebDriver installations
• AI agent sessions – let a coding agent inspect screens, perform actions through CLI or MCP tools, and save successful steps as a flow
• CI reports + debugging – export JUnit or HTML reports and keep screenshots and logs when tests fail
It’s open-source (Apache 2.0 license).
Link in the reply 👇
udomsak chundang retweeted
🧠 Open-Source Spotlight: AWS Context Ontology Accelerator
Draft business ontologies with AI, review them with domain experts, and serve context to agents through MCP. Built on RDF, OWL, and SHACL.
Apache 2.0 licensed.
@awscloud
openknowledgegraphs.com/soft…
udomsak chundang retweeted
Run coding agents in parallel — without putting every task in the same working folder.
Beluga is a desktop app for developers managing projects and coding agents, with Claude Code as its default agent.
It helps you work on multiple tasks at once by giving each task its own Git worktree, branch, and agent, with changes and diffs beside the conversation.
Key features:
• Per-task worktrees – run tasks side by side within one repository or across projects, while the main folder stays read-only for the agent
• Live agent activity – see file reads, commands, and sub-agents as they happen, and steer the agent mid-run
• Scratch tab – inspect screenshots, test logs, and API responses as the agent creates them
• Agent-to-agent requests – ask a friend’s agent to work on their Mac, with approval controls on their side
• Sentinel automation – describe scheduled or event-triggered work once and get a report after each run; merges require passing checks and your approval
It’s open-source (MIT license).
Link in the reply 👇
udomsak chundang retweeted
Rakazo is live on ProductHunt!
It's the #1 open source Grok Bot alternative.
Any model. Any sandbox. Works with your ChatGPT/Claude sub.
Easy to self-host. Or $15/mo for the cloud version.
If you want to support the project, I'd love a comment/upvote!
producthunt.com/products/rak…
45.7 万 Star 的逆向神器 REA,这回是真的杀疯了!!
以前只是帮 AI 拆软件,到了 v6.9.0,直接开一个单独的桌面让 AI 录操作、跑验收,就连硬件里的固件和微信小程序都能硬拆!
读完官方的更新说明,这次升级真的很猛:
▸ 桌面任务做成流水线:一行命令准备好 Linux 构建环境,装好经过校验的程序、把起止节点定下来,录一次就能分成好几条桌面任务,最后用自动校验加屏幕裁判完成验收
▸ 什么都能硬拆:.NET 单文件、Bun / Deno、Unity / Godot、微信小程序、红白机卡带,一直到 ESP 固件、Electron 缓存,还有离线 NTFS 日志,拿到就能拆
▸ 解决 Token 爆掉的问题:以前把逆向工具全挂进 Claude Code,上下文一下就被撑爆;现在会按需选工具、把结构精简、分页返回结果,没扫完就标成未知,绝不乱编
▸ 常用工具开箱即用:Claude Code、Codex、Cursor、LM Studio 都能原生接入
搞逆向、做桌面 Agent、折腾 Claude Code 的朋友可以看看。
仓库链接:github.com/morluto/rea/relea…
udomsak chundang retweeted
🧪 Open-Source Spotlight: atomRDF
Connect atomic structures and simulation workflows using materials-science ontologies. Move data between Python, JSON/YAML, and RDF.
Openly licensed under MIT.
openknowledgegraphs.com/soft…
#KnowledgeGraphs #MaterialsScience
udomsak chundang retweeted
Try different AI models and MCP tools without building your own chat UI.
AgentOne is an open-source desktop AI agent and research assistant for builders who want models and tools in one app.
It helps you connect models to your workflows by combining an MCP extension browser, support for hosted and local models, and configurable chats.
Key features:
• MCP extension browser – discover servers through the official MCP registry and connect custom local or remote servers
• Hosted and local models – bring your own API keys or use local models through Ollama and LM Studio
• Per-chat controls – adjust temperature, top-p, and other model settings separately for each conversation
• Parallel chats – keep multiple conversations running and switch models mid-session
• Cross-platform desktop app – built with Tauri for Windows, macOS, and Linux
It’s open-source (AGPL-3.0 license).
Link in the reply 👇
پروژهای که باعث میشه هرچقدر بیشتر از Agentها استفاده کنی، باهوشتر و کمهزینهتر بشن!
پروژه AgentRun که بهتازگی متنباز شده، به Agentهایی مثل Claude Code و Codex کمک میکنه از تجربه کارهای قبلی خودشون یاد بگیرن و اونها رو برای استفادههای بعدی ذخیره کنن.
اما چطور کار میکنه؟
فرض کن یک Agent در جریان انجام یک پروژه با یک مشکل فنی پیچیده روبهرو میشه و بعد از چند مرحله آزمون و خطا راهحلش رو پیدا میکنه.
AgentRun این تجربهها رو ثبت و سازماندهی میکنه:
- مشکلات فنی حلشده رو به یادداشتهای قابل استفاده تبدیل میکنه.
- روشها و راهحلهای موفق رو در قالب Skills ذخیره میکنه.
- در پروژههای بعدی امکان استفاده مستقیم از تجربههای قبلی رو فراهم میکنه.
اما جذابترین بخش ماجرا اینجاست:
وقتی یک Skill سه بار استفاده بشه، سیستم تلاش میکنه اون رو به یک Workflow تبدیل کنه.
یعنی بهجای اینکه Agent هر بار از صفر شروع کنه و دوباره مسیر حل مسئله رو طی کنه، میتونه از یک فرآیند آماده و تجربهشده استفاده کنه.
نتیجه چیه؟
- کاهش آزمون و خطاهای تکراری
- جلوگیری از بررسی دوباره مسائل حلشده
- صرفهجویی در مصرف Token
- استفاده مجدد از تجربههای موفق
- تبدیل تجربههای پراکنده به Workflowهای قابل تکرار
در واقع، AgentRun تلاش میکنه تجربه هر پروژه رو به سرمایهای برای پروژههای بعدی تبدیل کنه.
🔗 لینک کامل پروژه در GitHub:
github.com/Parcha-ai/agentru…
کانال تلگرام : t.me/yarad_ai
udomsak chundang retweeted
LLM engineer's handbook
(30 minutes a day, 10 weeks, 50 lessons)
a roadmap for llm inference serving where everything points at one service instead of scattering across demos. you get the mental model first, then serve a model, instrument it, load test it past 1000 concurrent requests, and tune it. you finish with a stack you configured yourself and a benchmark worth publishing.
here is what it covers:
→ the roofline model, and why decode waits on memory while prefill waits on compute
→ vLLM internals, PagedAttention and the scheduler, read from the code
→ Prometheus and Grafana for TTFT, inter-token latency, and queue depth
→ SGLang and RadixAttention prefix reuse, benchmarked against vLLM
→ load testing past 1000 concurrent requests
→ quantization across FP16, FP8, and INT4, on quality as well as speed
→ speculative decoding and KV eviction, including where the gains disappear
→ disaggregated prefill and decode, deployed on Kubernetes
→ a cost, latency, and quality router with per-request token budgeting
→ publishing a reproducible benchmark
the roadmap on GitHub: github.com/patchy631/time-to…
(don't forget to star 🌟)
i am also writing an article for each major topic. the first one is out, on how a GPU actually works.
the article is quoted below.
udomsak chundang retweeted
Your coding agents are on three machines. Stop checking them one by one.
GOAT is an open-source app for running coding agents across your own machines from one place. It’s a fork of T3 Code for builders working across laptops, desktops, and remote servers.
It helps you coordinate agent work by connecting each machine’s GOAT server to a desktop, browser, or phone interface, so you can follow tasks without switching between machines.
Key features:
• Multiple providers – use agents such as Claude Code, Codex, and Cursor in one interface
• Cross-machine work – start threads on SSH hosts or move a thread with its uncommitted changes
• Fleet view – see connected machines and counts of threads running or waiting for you
• Shared memory – turn finished conversations into memories and share them across devices through a hub
• MCP orchestration – let connected agents launch threads, watch progress, and receive completion notifications
It’s open-source (MIT license).
Link in the reply 👇
udomsak chundang retweeted
call me crazy but its pretty clear now.
rea + mitmproxy is going to expose more apps in 2026 than every data leak combined.
mitmproxy records everything an app sends over the network. rea reads that recording and points at the exact code that sent it.
so you dont just see "something phoned home". you see what left, where it went, and which part of the app did it.
if this lands on your timeline dont scroll past it.
heres the 60-second setup, free:
1. install mitmproxy and point one app on your own machine at it
2. save the session as a .har file
3. run npx rea-agents setup
4. tell claude code: "read session.har with rea. list every request that carries my data and the code that sent it"
run it on the apps you already trust. thats where it gets uncomfortable.
thats it. thank me later.
most people will never check what their apps send.
the ones who do will stop trusting the rest.
simple as that.





























