Sabitlenmiş Tweet
Shreyas Kapale
296 posts
Shreyas Kapale
@Shreyaskapale
Founding Architect @ Lyzr | Creator of GitAgent | believer of git native AI agent infra | Building the Infra to support AI agents.
127.0.0.1 Katılım Eylül 2015
123 Takip Edilen161 Takipçiler
@garrytan I appreciate gstack agent, but it would be really awesome if you can build it on gitagent.sh it will be portable, versionable from day 0. Let me know if you need help with setting this up.
English
Look ma, GStack is blowing up on Chinese Language Twitter!
在悉尼和稀泥@JamesAI
从 @garrytan 的 GStack 学习到的其实不止 Skill 的写法，更多的是 Workflow 的构建。一个可控的，步进式的，引导性的工作流，不管是实际产出，还是用户心智，对比 skill 都有极大的改进。我目前学习到的 5 个核心模式： 1. 统一的 AskUserQuestion 格式 每次提问都有 4 个固定板块: Re-ground（重申上下文）、Simplify（简单解释）、Recommend（带理由的推荐）、Options（带工作量估算的选项） 极大的降低了用户的输入门槛，从某种程度上来说，规范了用户的输入。而 AI 的效果，很大程度上是因为用户的输入层次不齐，造成产出不一样。 2. 全流程的 Skill 链路引导 每个 skill 结束时主动推荐下一个 skill，形成链，例如：/office-hours → /plan-ceo-review → /plan-eng-review → /ship 3. 进度条 + 状态追踪面板 QA 报告有评分、分类统计表、before/after 对比、ship readiness 仪表盘。进度和状态的可视化对于工作流是必须的。 4. 完成报告 + 数据统计 /retro输出 commit 统计、热点分析、focus score；/qa 输出修复数量和 before/after 截图。交付即复盘，且给了用户确定性。 5. Telemetry + 复盘数据积累 每次 skill 调用记录到skill-usage.jsonl，/retro 可以分析历史趋势。说实话我在我的写作skill 里面也做了数据追踪和迭代，但是我只追踪了写作规则，也就是功能层面的数据，并没有追踪工作流层面的效率数据。 所以基于 Gstack 的设计，我把我的写作工作流升级了。另外我也会对我的其他 skills 进行整合。 说实话 Garry 带来的是一套系统化的思维，也就是乔布斯之前说的“把点连成线”。而大家做过事的就知道，对比单点功能，流程化的效率和质量提升是指数级的。 从这个角度来说，确实做到了 "Markdown is code"，或者说 "Markdown is software"。
English
@ryanelliottcrow @theAIsailor Yes, if you provide a github repo and a pat token it syncs to that the state.
English
@theAIsailor Can it do resume uploads ?
English
Shreyas Kapale retweetledi
Meet ClawLess. The world's first serverless browser runtime for AI agents.
No server. No backend. No access to your host machine. Just open a browser tab and run a full Claw agent inside an isolated WASM sandbox.
Try Now: play.clawless.io
Github: github.com/open-gitagent/…
Built on GitAgent.sh standard.
cc @Shreyaskapale @lyzr__ai
English
@spencer_i_am @theAIsailor Because its built in gitagent standard if you provide a repo of GitHub it will sync state and persist indefinitely. Performance is directly dependent on the host spec, and stability is good but as its opensource we will keep improving let me know if you run into issues.
English
@theAIsailor I assume there needs to be or is already a strong memory system since this can't possibly run 24/7. Is there a limitation from a performance/stablity standpoint?
English
Nice catch, dont have an answer to that yet but because clawless is also an sdk + its built ok gitagent standard, it means your agent can store persistent store into git repo or because its sdk you can build a system to do that , or maybe even a plugin memory api that fetches only relavent data as needed? Its a open-ended thought.
English
@theAIsailor sandboxing at the wasm layer is smart for isolation but curious how you handle long context state when you can't persist to host. js heap gets expensive fast for any agent doing multi-step reasoning
English
@garrytan Exactly - that is why for the portability problem there is gitagent.sh : ) let’s collaborate on gstack
English
For agentic systems founders and dev tools founders:
People do not want to pay for raw markdown and they shouldn't have to.
But they may pay for orchestration, hosting, updates, collaboration, portability, analytics, and managed execution.
These can be great businesses.
English
Shreyas Kapale retweetledi
@logic @theAIsailor Great catch we are working on adding a OpenAI compatible inference system, so it will enable any model to run on clawless
English
@theAIsailor How does one BYOLLM outside of three biggies?
English
Because the agent is built on gitagent.sh standard, it holds the entire state in github repo , also gives option to store in local storage, or export it as a zip.
But best part is if you build with clawless sdk you can handle state storage on your server only runtime is at clients end!
English
@theAIsailor Interesting approach. How will it handle long-running agents and local storage in WASM sandbox?
English
@garrytan 😭 you could have just used gitagent, it already translates agent into all of them
registry.gitagent.sh/agent/shreyas-…
Here i did it for you
English
@GithubProjects Sorry :(
Already pushed - clawless.io to prod.
English
|￣￣￣￣￣￣￣￣￣￣￣￣￣￣￣￣￣|
|Don't Push To Production On Friday|
|＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿|
\ (•◡•) /
\ /
——
| |
|_ |_
English
Shreyas Kapale retweetledi
Gitagent by @Shreyaskapale at @lyzr__ai Research Labs live on ProductHunt Today!!!!
producthunt.com/products/gitag…
English
@Niraj_Dilshan @GithubProjects yup, do star the repo and post about it, help project reach its audience <3 as it solves a real problem.
English
@Shreyaskapale @GithubProjects running completely in the browser sandbox is wild. the edge just keeps getting pushed further out.
English
NVIDIA NemoClaw is an open source stack that simplifies running OpenClaw always-on assistants safely.
It installs the NVIDIA OpenShell runtime, part of NVIDIA Agent Toolkit, a secure environment for running autonomous agents, with inference routed through NVIDIA cloud.
English
@cuysheffield You should make it available on npm ! So it can run using clawless.io
English
Excited to share Visa CLI, the first experimental product from Visa Crypto Labs. Check it out and request access here
visacli.sh
English
@Niraj_Dilshan @GithubProjects Checkout clawless.io too dont even need a server to run the agent runs purely on browser sandbox with virtual filesystem.
English
@GithubProjects finally a proper secure runtime out of the box. making it easier to host local agents safely
English
English
@GithubProjects safety only starts getting real once the runtime owns permissions, not just prompts
English
@bitcoinlanparty @GithubProjects Yeah that is exactly why go serverless - run it on clawless.io no server required.
English
@GithubProjects Running your AI agents through NVIDIA's cloud is convenient right up until NVIDIA decides to change the rules
English
Shreyas Kapale retweetledi
@theAIsailor browser isolation is the only way to stop these agents from rm -rf'ing my host when they hallucinate an edge case. WASM saving laptops one tab at a time.
English
@battista212 You can do all that on clawless.io too
Plus you dont need a server at all, file system runs on your browser - isolated using webcontainer
English
NVIDIA NemoClaw launches: open-source stack for securing always-on AI assistants. OpenShell runtime sandboxes network requests, file access, and inference calls with declarative security policies. Runway releases GWM-1, a "General World Model" to simulate reality in real time.
English
General Intuition raised $133.7M at seed to build world models from gaming data. Bet: actions from Medal's 1B+ annual gaming clips transfer to real-world robotics. YLeCun's AMI Labs raised $1.03B, World Labs $1B, Decart $100M. World models are now a multibillion-dollar race. #AI
English
@NVIDIAAIDev I would rather use - clawless.io
No server required! - serverless, runs on browser directly!
Sandboxing purely using web container!
English
🦞 Make claw agents safer with our new NVIDIA OpenShell – an open source runtime to build with autonomous evolving agents.
🐚 OpenShell sits between your agent and your infrastructure to govern how the agent executes, what the agent can see and do, and where inference goes.
🔐 Gives you fine-grained control over your privacy and security while letting you benefit from the agents’ productivity.
Run one command—and make zero code changes. Then any claw or coding agent like OpenClaw, Anthropic’s Claude Code, or OpenAI’s Codex can run unmodified inside OpenShell.
Every SaaS company just became an agent company. The missing piece was never the agents — it was the infrastructure that makes them safe enough to deploy. That's OpenShell.
Technical blog to learn more ➡️ nvda.ws/4brnAPW
English