Shen Sean Chen's banner

Shen Sean Chen

@ShenSeanChen • 12,670 subscribers

founder @waku_agent • https://t.co/sPSpu8yPx6 • youtube(38k) https://t.co/rMP0jaxHK1 • ex-google/mit • pokémon/onepiece fan

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

I made a video breaking down AI agent observability and evals, and ran the whole thing on open source. If you build agents, sooner or later you watch one run and realise you have no idea what it just did, or what it cost you. That gap is what observability and evals are for. Least glamorous part of the stack, and the part you end up needing most. 🛠️ treg, OpenRouter for agent tools (4.2k): 🔑 Try treg: 🌟 Github Repo (2k): 💻 Portable memory: ☕ Join our community: Observability is a camera. It records what ran, how long, what it cost. Eval is a referee. It decides whether the run was good enough to ship. I ran it on Treg for the tool calls and Waku Agent for the memory. treg is an OpenRouter for agent tool calls, so your agent carries one API key instead of one per tool. Both open source, so you can read the code rather than take my word for it. Full video in the first reply 👇 You Can Build Anything. You Can Learn Anything. 💪

I made a video breaking down AI agent observability and evals, and ran the whole thing on open source. If you build agents, sooner or later you watch one run and realise you have no idea what it just did, or what it cost you. That gap is what observability and evals are for. Least glamorous part of the stack, and the part you end up needing most. 🛠️ treg, OpenRouter for agent tools (4.2k): 🔑 Try treg: 🌟 Github Repo (2k): 💻 Portable memory: ☕ Join our community: Observability is a camera. It records what ran, how long, what it cost. Eval is a referee. It decides whether the run was good enough to ship. I ran it on Treg for the tool calls and Waku Agent for the memory. treg is an OpenRouter for agent tool calls, so your agent carries one API key instead of one per tool. Both open source, so you can read the code rather than take my word for it. Full video in the first reply 👇 You Can Build Anything. You Can Learn Anything. 💪

33,734 次观看 • 21 天前

I dug into Pi because the greatest projects are usually built on top of one very simple building block. 𝗣𝗶 𝗵𝗮𝘀 𝟳𝟳,𝟬𝟬𝟬 𝗚𝗶𝘁𝗛𝘂𝗯 𝘀𝘁𝗮𝗿𝘀 𝗮𝗻𝗱 𝗲𝘅𝗮𝗰𝘁𝗹𝘆 𝗳𝗼𝘂𝗿 𝘁𝗼𝗼𝗹𝘀: 𝗿𝗲𝗮𝗱, 𝘄𝗿𝗶𝘁𝗲, 𝗲𝗱𝗶𝘁, 𝗯𝗮𝘀𝗵. 𝗦𝘂𝗽𝗲𝗿 𝘀𝗶𝗺𝗽𝗹𝗲. No MCP, no sub agents, no plan mode, no to-dos, no permission popups. Their site has a whole section called "what we did not build." So here's the whole logic by Mario Zechner: prompt in → agents.md + system prompt become one forkable JSON → loop starts → read / write / edit / bash → session saved as a tree → loop ends agent-loop.ts has only 792 lines. I think it's beautiful. Initially I was wondering if it's gonna be a little bit too basic. And yes out of the box it's weaker than Claude Code. But skipping MCP isn't purity, it's a context budget. One MCP can sit tens of thousands of tokens in your window every single turn, for a tool you use maybe 10% of the time. Pi moves that weight from always on to on demand. A skill is just markdown. An extension is one .ts file. A package ships both. I ran it in the terminal, then again as a coding sub agent inside my own harness Waku-Agent ( through a delegate_task tool. Same binary both times. You're paying setup time for a harness that stays yours. Full 22 min breakdown in the comments. Save the diagram 🔖 You Can Build Anything. You Can Learn Anything. 💪

I dug into Pi because the greatest projects are usually built on top of one very simple building block. 𝗣𝗶 𝗵𝗮𝘀 𝟳𝟳,𝟬𝟬𝟬 𝗚𝗶𝘁𝗛𝘂𝗯 𝘀𝘁𝗮𝗿𝘀 𝗮𝗻𝗱 𝗲𝘅𝗮𝗰𝘁𝗹𝘆 𝗳𝗼𝘂𝗿 𝘁𝗼𝗼𝗹𝘀: 𝗿𝗲𝗮𝗱, 𝘄𝗿𝗶𝘁𝗲, 𝗲𝗱𝗶𝘁, 𝗯𝗮𝘀𝗵. 𝗦𝘂𝗽𝗲𝗿 𝘀𝗶𝗺𝗽𝗹𝗲. No MCP, no sub agents, no plan mode, no to-dos, no permission popups. Their site has a whole section called "what we did not build." So here's the whole logic by Mario Zechner: prompt in → agents.md + system prompt become one forkable JSON → loop starts → read / write / edit / bash → session saved as a tree → loop ends agent-loop.ts has only 792 lines. I think it's beautiful. Initially I was wondering if it's gonna be a little bit too basic. And yes out of the box it's weaker than Claude Code. But skipping MCP isn't purity, it's a context budget. One MCP can sit tens of thousands of tokens in your window every single turn, for a tool you use maybe 10% of the time. Pi moves that weight from always on to on demand. A skill is just markdown. An extension is one .ts file. A package ships both. I ran it in the terminal, then again as a coding sub agent inside my own harness Waku-Agent ( through a delegate_task tool. Same binary both times. You're paying setup time for a harness that stays yours. Full 22 min breakdown in the comments. Save the diagram 🔖 You Can Build Anything. You Can Learn Anything. 💪

59,186 次观看 • 3 个月前

𝗜 𝘁𝗲𝘀𝘁𝗲𝗱 Grok Bot 𝗮𝗻𝗱 Muse 𝘀𝗶𝗱𝗲 𝗯𝘆 𝘀𝗶𝗱𝗲, 𝗮𝗻𝗱 𝗱𝗿𝗲𝘄 𝗯𝗼𝘁𝗵 𝗵𝗮𝗿𝗻𝗲𝘀𝘀𝗲𝘀 𝗯𝘆 𝗵𝗮𝗻𝗱. One from SpaceXAI, one from AI at Meta under Alexandr Wang. Neither is open source, so I tried both and drew the architecture myself from whatever docs exist. Tbh I struggled quite a bit to find real use cases that can surprise me beyond what we have on Claude Code and Codex, but maybe we are just getting started on personal agent harnesses. 🌟 Github Repo (1.8k): 💻 Portable memory: ☕ Join our community: Some quick differences I found: Grok Bot gives your whole account one cloud VM where every Bot shares the browser, the /workspace, the terminal and the credentials, so a Bot is isolating a personality rather than a context. Muse gives each person one secure VM with a single agent, and a Sentinel sitting outside the container holding your tokens. Strip the branding off either one though and it's the same chain: one VM → connectors → routines → approval gate → memory What surprised me is that Muse ships a memory.md you can read, edit and download, but Grok Bot doesn't. I didn't find where it keeps memory at all. So I gave both of them Waku Agent memory over MCP instead, saved a fact in one harness, and the other one recalled it cold in a fresh chat. I also killed the shopping test halfway. I felt Amazon was built for humans to read and click, and I wasn't handing an agent my login username and password. Maybe the fast checkout use case is a better fit for software / AI product purchases. Watch it, then save both harness diagrams below. 🔖 You Can Build Anything. You Can Learn Anything. 💪

𝗜 𝘁𝗲𝘀𝘁𝗲𝗱 Grok Bot 𝗮𝗻𝗱 Muse 𝘀𝗶𝗱𝗲 𝗯𝘆 𝘀𝗶𝗱𝗲, 𝗮𝗻𝗱 𝗱𝗿𝗲𝘄 𝗯𝗼𝘁𝗵 𝗵𝗮𝗿𝗻𝗲𝘀𝘀𝗲𝘀 𝗯𝘆 𝗵𝗮𝗻𝗱. One from SpaceXAI, one from AI at Meta under Alexandr Wang. Neither is open source, so I tried both and drew the architecture myself from whatever docs exist. Tbh I struggled quite a bit to find real use cases that can surprise me beyond what we have on Claude Code and Codex, but maybe we are just getting started on personal agent harnesses. 🌟 Github Repo (1.8k): 💻 Portable memory: ☕ Join our community: Some quick differences I found: Grok Bot gives your whole account one cloud VM where every Bot shares the browser, the /workspace, the terminal and the credentials, so a Bot is isolating a personality rather than a context. Muse gives each person one secure VM with a single agent, and a Sentinel sitting outside the container holding your tokens. Strip the branding off either one though and it's the same chain: one VM → connectors → routines → approval gate → memory What surprised me is that Muse ships a memory.md you can read, edit and download, but Grok Bot doesn't. I didn't find where it keeps memory at all. So I gave both of them Waku Agent memory over MCP instead, saved a fact in one harness, and the other one recalled it cold in a fresh chat. I also killed the shopping test halfway. I felt Amazon was built for humans to read and click, and I wasn't handing an agent my login username and password. Maybe the fast checkout use case is a better fit for software / AI product purchases. Watch it, then save both harness diagrams below. 🔖 You Can Build Anything. You Can Learn Anything. 💪

23,055 次观看 • 1 个月前

𝗜 𝗼𝗽𝗲𝗻 𝘀𝗼𝘂𝗿𝗰𝗲𝗱 𝗺𝘆 𝗹𝗼𝗰𝗮𝗹 𝗔𝗜 𝗽𝗲𝗿𝘀𝗼𝗻𝗮𝗹 𝗮𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁: 𝗔𝗴𝗲𝗻𝘁 𝗛𝗮𝗿𝗻𝗲𝘀𝘀 & 𝗟𝗼𝗼𝗽 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗶𝗻 𝗿𝗲𝗮𝗹 𝗰𝗼𝗱𝗲. Everything in real code you can read in an afternoon: Harness, Loop Engineering, Memory, Eval, Tracing. 100% local on your laptop. 💻 Github Repo: ☕️ Buy me a coffee: The whole system is one loop: message in → retrieval gate → agent run → tools fire → trace → eval → memory saved → skills grow - The loop is ~95 lines of plain Python - Your memory is one SQLite file. Open it. It's yours. - Evals built in: deterministic + LLM-as-judge, with a release gate - Telegram gateway, voice wake word, Apple Calendar tools In the video it books the World Cup QF on my calendar, remembers who my friends are, and answers me on Telegram. Clone it, star it, break it, PR it. Repo + 20-min walkthrough in the first reply 👇 Watch it, then save the architecture diagram below. 🔖 You Can Build Anything. You Can Learn Anything. 💪

𝗜 𝗼𝗽𝗲𝗻 𝘀𝗼𝘂𝗿𝗰𝗲𝗱 𝗺𝘆 𝗹𝗼𝗰𝗮𝗹 𝗔𝗜 𝗽𝗲𝗿𝘀𝗼𝗻𝗮𝗹 𝗮𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁: 𝗔𝗴𝗲𝗻𝘁 𝗛𝗮𝗿𝗻𝗲𝘀𝘀 & 𝗟𝗼𝗼𝗽 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗶𝗻 𝗿𝗲𝗮𝗹 𝗰𝗼𝗱𝗲. Everything in real code you can read in an afternoon: Harness, Loop Engineering, Memory, Eval, Tracing. 100% local on your laptop. 💻 Github Repo: ☕️ Buy me a coffee: The whole system is one loop: message in → retrieval gate → agent run → tools fire → trace → eval → memory saved → skills grow - The loop is ~95 lines of plain Python - Your memory is one SQLite file. Open it. It's yours. - Evals built in: deterministic + LLM-as-judge, with a release gate - Telegram gateway, voice wake word, Apple Calendar tools In the video it books the World Cup QF on my calendar, remembers who my friends are, and answers me on Telegram. Clone it, star it, break it, PR it. Repo + 20-min walkthrough in the first reply 👇 Watch it, then save the architecture diagram below. 🔖 You Can Build Anything. You Can Learn Anything. 💪

47,673 次观看 • 3 个月前

I'm open sourcing a template I built to launch a full-stack SaaS with Stripe Payment, NextJS, Supabase and Google OAuth, in ONE SHOT. It took me some time so I decided to share this to you all. The goal is to help everyone save time when launching your next SaaS, with EVERYTHING YOU NEED. You can build anything. You can learn anything. 💪 Github Repo: Try The App: Video Link: Template Features: 🔐 Authentication with Supabase 💳 Stripe payment integration 🌓 Dark mode support 📱 Responsive design 🎨 Tailwind CSS styling 🔄 Framer Motion animations 🛡️ TypeScript support 📊 Error boundary implementation 🔍 SEO optimized

I'm open sourcing a template I built to launch a full-stack SaaS with Stripe Payment, NextJS, Supabase and Google OAuth, in ONE SHOT. It took me some time so I decided to share this to you all. The goal is to help everyone save time when launching your next SaaS, with EVERYTHING YOU NEED. You can build anything. You can learn anything. 💪 Github Repo: Try The App: Video Link: Template Features: 🔐 Authentication with Supabase 💳 Stripe payment integration 🌓 Dark mode support 📱 Responsive design 🎨 Tailwind CSS styling 🔄 Framer Motion animations 🛡️ TypeScript support 📊 Error boundary implementation 🔍 SEO optimized

191,422 次观看 • 1 年前

I made probably the simplest MCP demo to show you what it actually does, which was open sourced by Anthropic in Nov last year (Yes I'm a little late to the game). I think this is a cool tech and will potentially save us a lot of time coz it allows tools and agents to communicate much more conveniently. Github Repo: YouTube: Buy Me A Coffee: You can build anything. You can learn anything. 💪

I made probably the simplest MCP demo to show you what it actually does, which was open sourced by Anthropic in Nov last year (Yes I'm a little late to the game). I think this is a cool tech and will potentially save us a lot of time coz it allows tools and agents to communicate much more conveniently. Github Repo: YouTube: Buy Me A Coffee: You can build anything. You can learn anything. 💪

26,066 次观看 • 1 年前

没有更多内容可加载