Grok Shell - Added Version Control on Mini Apps You can tweak , improve or revert to old version. Mini Apps are quite different because it can have its on Database and Api Integration Link: github.com/hexuria/grok-shel…

lets ship potetobot

Grok Shell -It's Grok Bot UI/UX but Build with Rust and Native Shell Uses OpenCodex For Inference atm. Soon I'll port it to use opengrok-server Unique Features - Support for Better office - Mini App 100% Free and Open Source! github.com/hexuria/grok-shel…

Devin max at 90% Claude at 99% Opus 5.5 is just too Good Looks like we gonna end the week sooner I hope someone can sponsor me So I can keep building opensource software Thanks you Universe :D I know someone will read this #x #algorithm do you work!

I've re-build the UI of Open Grok 100% with Native Shell and Mini Apps Compiled on Demand @grok @bot @poteto you guys should re-write it in Rust + GPUI + Native Shell This new Mini App Feature is a Killer ;D

This isnt @grok @bot But on Native Shell I can already see so many re-writes from Electron to Native Shell LOL

Ultra instrint 194/384 Jev Ultra Fast 120/384 That is on accuracy Test Im hell bent on making this the fast, most cheapest solution for your browser use agent. Text Model: Haiku 5.5 Decision Model: Jev

I'm the First Contributor on Native Shell :D Thanks for Accepting my PR Hope to Contribute more to this project!

Generative UI on Chat will be phase out soon Mini apps would be a trend, mark my words! Prompt your Agent to build a mini app after few minutes, it has its own database or api integration And you can start using it without any dependencies How Cool is that?

after #75 its performs much better :) @satvikpathak007

Browser agents don't need to THINK MORE. They need to THINK LESS. Same task. 6 runs. All passed. Ultra-Instinct vs Obscura + pageshot: ⚡ 1.9× faster (4.3 vs 8.2 min) 🧠 25% fewer tokens 🔧 41% fewer tool calls Built in Rust. Open source. github.com/hexuria/ultra-ins…

typesafe-sdk crates is #1 on downloads recently got bumbed to 0.3.0 - support systemone , cloudflare, openai - support images complete examples usage parity with python add it on your next project: cargo add typesafe-sdk

I don't believe using Skills , can make your coding Agent better Specially if you're using Frontier models Most of the time they perform worst

Im building Buwiz - AI Native Tax Form Compliance App Built to replace the old Klunky Legacy eBIRForms here in the Philippines Its 100% Free and Open Source

Im building Buwiz Forms Modern , Fast and Free Tax Filling Software Help people File Tax Easily and On Time

Those that are deeply Invested with GPUI subsecond Hot Patching works with GPUI You don't need a full rebuild :)

I've used all ULTRA in 2 Weeks I don't know if this is worth keeping @bot

🚀 BUWIZ FORMS V1 IS READY FOR RELEASE! Here's what's new: ✅ 39 OUT OF 43 BIR FORMS SUPPORTED The remaining 4 forms are affected by BIR server-side submission issues. ✅ 100% FORM VALIDATION PARITY WITH BIR This is the most important part. 🤖 CREATE YOUR OWN BUWIZ AI AGENT

Anyone from @cognition @DevinAI Im confused on this two usage limit? 8% is on desktop , 18 % is on cloud which limit is the correct ONE lol

how much of 20$ plan in devin would last lol lets see

That time I got re-incarnated to build opengrok Today's Quest: Added new office suites tools

I built Instinct, a mathematical decision engine written entirely in Rust. It uses mathematical scoring, pattern matching, and symbolic reasoning to make decisions without relying on LLM inference every time. When uncertain, it abstains instead of guessing.

@poteto shipped 2.5k PRs lasts month @warpdotdev 30 Engineers Shipped 2K PRs How much are those 30 Engineers Salary? Im sure warp can match you a Salary @grok @bot will miss you after that LOL @thsottiaux Left the room LOL

OpenGrok needs a browser agent that doesn't burn tokens repeating the same tasks or blindly click when pages change. So I'm building ULTRA INSTINCT ⚡ in 100% Rust. Observe. Verify. ABSTAIN when unsure. The endgame? Mastered tasks become instinct. ZERO LLM inference.

Here is a Quick Test Demo:

Open Grok is almost ready for the First Release Im using right now the 250$ free credits by @AnthropicAI In no time i can simply use OpenGrok to build itself

OpenGrok Features - 100% Better Knock Off of Temu @poteto LOL (please dont hate me) - Fine Grain Control on Agents - Templates - Agents Team - Recipes - Ultra Instinct - Portable Agent Tool Chain - Secret Vaults - Credential Swap - Egress Re-Route Traffic 100% OpenSource

Jev Ultra Fast ? I thought Rust is the only one That should always use the word Blazingly Fast Since Browser Use , use it ... There is no going back Its time For ULTRA INSTINCT! LOL

Im out ... Waiting for a Miracle Reset by @lydiahallie Save me LOL

Collectively known as OpenGrok I've Been Building Multiple Repos: - NativeChat - OpenGrok Server - Box - Open AI Gateway You can Easily Switch From Subscription to API Key Whatever @Grok @bot does , it can. 100% Built with RUST

If you’re an angel investor, you probably shouldn’t miss this one. I think I finally found a real AI infrastructure problem worth betting on. Scenario: Your agent needs Rust, Node, gh, or another tool to do its job. Solution: Agent Toolchain

Im officially out of @cursor_ai ... While @claudeai is at 2% before i hit limit. and @OpenAI at 78% left Lord Give me Unlimited Token Usage for the Greater Good of Mankind...

New OpenGrok update: Use the same Claude, Codex or Grok subscription across multiple computers with one OpenGrok account. If your subscription expires or hits its limit, inference automatically falls back to your API Gateway. Messages and artifacts stay intact and Sync

If your not building 24/7 then your just another Viber Coder

I’ve been shipping a ridiculous amount of features across NativeChat, OpenGrok Server, Box and OpenAI Gateway. Collectively known as OpenGrok. Everything is built from scratch in Rust.

You can now use @Opencodex with Open Grok as Inference Source Instead of OpenGrok Server. Still the Messages Persist to the Server. Only Downside of this is you need your laptop opened But the good thing is your not Gonna be banned.

i have 3 pro gift … if anyone wants it let me know :)

Even chatgpt cant understand you lol …

New feature of claude code keep your agent coding even if your already maxed out your weekly limit @AnthropicAI I thought this was a myth. I tested it at 99% gave all agents on all session a tasks now its 100% it still keep on going I am loving it <3

Still no reset on codex lol

Your node_modules is a mini black hole? Our Target Folder is a Super Gigantic Black Hole This shit has got to stop! 1 TB is getting eaten up by just 3-5 projects

I thought we have a reset? does Codex Plus have a reset? @thsottiaux

Went to sleep with 60 open PRs. Woke up with 7 left. The unlock isn’t just better coding agents it’s better verification. TLA+ • Lean • Kani • Loom • Bend • Benchmarks • Fuzzing • Security checks Agents that don’t just write code. They prove they didn’t break shit.

I know a reset is coming so I maxxed it out just now LOL

I did just now maxxed it out to 100% Thank you ;D

Your AI agent can use your Mac. Until you go AFK. Then macOS asks for permission. The agent can’t click it. You can’t click it. Because you’re not there. So I built Allowly. Phone → @Tailscale → Mac → HID → actual click. Your agent says: “macOS needs you.” You tap Allow from your phone. No fake clicks. No TCC bypass. Just a real HID click from wherever you are. Powered by @TypesafeAI JEV for the agent’s decision-making and computer/browser use.

Added important security feature when re-routing traffic from agents computer to your computer It basically prevent agents without permission to tunnel connection to your computer if its not approved Agent can still visit those sites normally if they are not rerouting traffic to your computer. This is a useful feature, if the App/ Website you are trying to access , blocks your access due to the Sandbox computer IP address isnt on your current IP.

Skill issue with AI SLOP @finkd

I'm building a real voice-controlled remote for Mac. And I'm using JEV for almost everything. 🧠 Decision making — auto allow / deny actions 🖥️ Computer Use — interact with the Mac 🌐 Browser Use — navigate and control websites, much faster 🧩 Generative UI — dynamically generate forms and controls 🎙️ Voice Control — control the entire workflow remotely The problem is macOS TCC. When an agent triggers certain macOS permissions, you still need a real physical interaction. A Computer Use Agent can't simply click the approval button. That's intentional. Great security. Terrible for unattended Macs. So I added another layer: JEV → Tailscale → remote hardware → Mac The hardware can physically interact with the Mac when TCC requires it. So the agent doesn't just see my Mac. It can reason, decide, interact, browse, generate UI, and physically control it. All remotely. No sitting in front of the machine. That's the goal: A Mac that an AI agent can actually operate — unattended.

Rebuilt OpenGrok using NativeChat (GPUI + RUST) Added: - GPUI Agent (Allow Agents to Control Client Dekstop App built with GPUI on Dev mode) - Integrated Grok Box - Update Box , Reset Box - Ability to Retain Data after Reset or Update - Added Recipes - Teach a Tasks Generate Recipes (Grok Bot Generates Just a Markdown ) (Note: once I get access to Jev by @typesafeai ill immediately add it to enhance the execution engine to use this model) - AG-UI Compatible Response - 120 FPS - Local Execution Daemon - Box Execution Daemon - Ability to Share Computer (Org , Group, Other Agent) - Stream Response for Text - Buffered Response for AG-UI

I've added headless mode for Buwiz Forms perfect for Sandboxed environment where you dont need a full Desktop Environment

🚀 A NEW ERA OF AI-POWERED TAX FILING IS COMING TO BUWIZ FORMS We’ve been quiet for the past few months. But we haven’t been sitting still. We’ve been LASER FOCUSED on building the infrastructure needed to make AI truly useful for businesses. Today, we’re proud to announce the AI infrastructure that will power Buwiz Forms (eBIRForms) moving forward. We’ve built 4 LAYERS OF AI INFRASTRUCTURE: 🧠 OPEN GROK Our AI desktop client, with a mobile app coming soon. ☁️ OPEN GROK SERVER Run your AI Agents 24/7 in the cloud without needing to keep your laptop running. 🖥️ GROK BOX A dedicated computer for your AI Agent. 🔌 OPEN AI GATEWAY Bring Your Own Key. Bring Your Own Subscription. Use the AI models that work best for you. This means you are no longer locked into a single AI provider. You can use Fable 5, Grok 4.6, GPT Astra, Gemini 3.8, and other AI models as the ecosystem continues to evolve. BUT HERE’S WHERE IT GETS REALLY INTERESTING… 🤯 YOUR AI AGENT CAN ACTUALLY DO THE WORK. Imagine telling your AI Agent: “Create a Tax Profile for Client A.” “Prepare the upcoming tax forms.” “Fill out the required BIR forms.” “Submit the forms to BIR.” “Download the completed PDF.” And much more. This isn’t just AI answering questions. Your Agent can actually OPERATE THE COMPUTER and execute the workflow. Because our applications run on macOS, Windows, and Linux, your Agent can even have its own dedicated Linux computer through GROK BOX. That means your AI Agent can have: 🧠 Its own intelligence 💻 Its own computer 🔧 Its own tools 📋 Its own workflows ⏰ Its own schedule And it can keep working even when you’re not sitting in front of your laptop. ⚡ IMAGINE AUTOMATING YOUR MONTHLY TAX OPERATIONS You could set a routine for your Agent: “Every month, prepare the required tax filings for Client A.” Your Agent can execute the workflow according to the rules and permissions you define. No laptop required. No repetitive manual work. No constantly switching between applications. Just an AI Agent with the infrastructure it needs to get the job done. 🏢 ONE AGENT FOR EACH CLIENT For accounting firms and organizations managing many clients, this opens up an entirely new way of working. You can create dedicated AI Agents for different Tax Profiles and clients. Each Agent can have its own context, workflow, tools, and responsibilities. THIS IS WHAT WE MEAN BY UNLOCKING AI INTELLIGENCE FOR YOUR ORGANIZATION. We are not simply adding a chatbot to Buwiz Forms. WE ARE BUILDING AN AI WORKFORCE AROUND IT. 🚀 NOW WE ARE LOOKING FOR: 🤝 EARLY ADOPTER ENTERPRISES Organizations that want to experiment with AI-powered tax operations before this becomes mainstream. 🤝 STRATEGIC PARTNERS Companies that want to build with us and help bring this technology to more businesses. 💰 INVESTORS Partners who believe AI Agents will fundamentally change how businesses operate and want to help us bring this technology to the public at scale. We’ve spent the last few months building quietly. NOW WE’RE READY TO SHOW YOU WHAT WE’VE BEEN WORKING ON. The future of tax compliance isn’t just software. It’s SOFTWARE + AI AGENTS + AUTONOMOUS INFRASTRUCTURE. And we’re just getting started. 🚀 BUWIZ FORMS Powered by a new generation of AI infrastructure. #BuwizForms #eBIRForms #AIAgents #AI #TaxTechnology #Automation #Fintech #Philippines #Accounting #FutureOfWork #OpenGrok #GrokBox

Reach for the stars, and if you don't grab them, at least you'll fall on top of the world. piped.video/watch?v=EPo5wWmK…

Grok Box is a new Layer (sandboxed environment) I built on top of Open Grok to replace box by @asciidotdev Normally when Agents try to Use your Computer It would do 1 tool call for each action Now this is useful for tasks that are exploratory But is very slow and eats a lot of token I added a new Feature on Grok Box where you can create Recipe That can be process by the Box execution daemon Recipe is as simple as it is you add bunch of actions to produce results deterministically it good for worflow that you wanna do everytime e.g. creating an invoice There are 3 ways to create Recipe 1. You Instruct your Agent to do this 2. You manually do this using GUI 3. You use the Teach a Task , You record your screen and all the steps gets recorded, you can still edit this on GUI and remove unnecessary steps You can use this new feature i added to simply reduced the amount of tool call and token And it speed up as well the execution of Agent when using your computer. Once recipe is Saved , Agents can just use it during a toolcall so instead of calling in 5 steps that would take 5-6 seconds with Recipe you make 1 Tool call with 5 actions that takes only roughly 1 sec Meaning if your agent can create bunch of recipe and execute it , it would make tasks very fast. Meaning you can accomplish more tasks in a given time.

Open Grok using its own computer ... execution is also much faster compare to Grokbot ... The price also to do this is significantly much lower At scale as you have option e.g. - 1 computer shared across the organization - per user computer - per bot computer Depending on. your budget you can pick a solution... Also you only pay when you are actually using the computer... This is way cheaper than renting your own VPS :D

Open Grok Server + Open AI Gateway + Open Grok as drop in replacement for Cursor Server and GrokBot. This can be deployed on any cloud provider ... like AWS, GCP , Azure , Digital Ocean, Akamai et.al. Right now its deployed locally with Docker... If you think your grok bot subsscription is always running out easily. Then its time to switch with Open Grok I'll open source it ... After I completed polishing it... If your interested comment 'Open Grok' ill give you private invite to the repo. also please share it to your network here in X Together lets Liberate @Grok @bot

I've been heavily working on Open Grok Server as Drop in replacement for Cursor. All of the provider works now. - Cursor - Open Grok Server - Codex - Claude Code The main difference Cursor and Open Grok Server allows you to turn off your computer and your Agent is remote and its has its own remote Computer (Sandboxed) It would work 24/7 even if your Main Laptop / Computer is Turn Off.... On the Other hand Codex and Claude, if you turn off your computer they will not work even if you use a remote computer , as the Agent lives on your Computer not on the Remote Computer you give it access to. Open Grok Server supports hundreds of Provider With the baked in Intelligence, harness , tools, mcp, skills, connectors etc. Its perfect for Organization that needs control and full ownership of the whole AI Infrastructure even to the Inference. If this interest you , DM me You can ask me anything ;D

Anyone of @spacexai that created @bot can beat Open grok performance on chat feed on heavy media chat session? Im just a guy from a third world country Tell me how on earth I beat you guys? @elonmusk you should be hiring me LOL

I benchmark opengrok vs grokbot github.com/hexuria/opengrok/… opengrok solves a lot of issues grokbot has on chat feed , specially on heavy media chat session If you think you need a better performant grokbot? Try opengrok its free and opensource Use your Cursor Account , like on Grokbot. Feel The Difference

i still dont get 100 after connecting 🥲

Im using @bot to build opengrok github.com/hexuria/opengrok yeah badly needed this as im at 100% usage thanks

Open Grok (@bot ) most up to date 0.3 Much performant than original @bot Don't Trust me , Build it and Try it! github.com/hexuria/opengrok This would come with a pair Rebuilt Backend With @rustlang , tonic , rig by @joshmo_dev and Integration to Open AI Gateway github.com/hexuria/open-ai-g… Which would allow you to BYOK and BYOS with @AnthropicAI @OpenAI @cursor_ai or any model with @OpenRouter For your Organization usage Only For now the Open Grok Server is closed source but once its 100% ported i will open source it to replace @cursor_ai GPRC in the coming weeks OpenGrok features: - @trycua with Multiple Computer (windows,Mac Linux) - @CopilotKit AG-UI Compatible - Sandbox environment @asciidotdev box - Deployed at your own Infrastructure @awscloud @googlecloud @Azure - Share Selected Messages to Team and Bookmark it - Better Chat Feed Optimization than original @bot - All features of @bot included Soon: - Add your teammates , clients on a channel

I’ve been quietly building two projects: Open AI Gateway and Barok Works. Open AI Gateway is a gateway that lets you serve subscription-based and LLM provider API keys to any client. That means you can use those models in tools like Claude Code, or connect them to other clients — including Barok Works (no pun intended 😅). Barok Works, on the other hand, is kind of like Claude Cowork or ChatGPT Work, but think of it as Grok Bot with superpowers. It can use models from Open AI Gateway, and it can even connect to Cursor so you can leverage your Grok Bot subscription in your own agent workflows. But the bigger idea is giving your agent access to more computers and environments: * Windows 365 integration — give your agent its own Windows machine * Local VM — give your agent a Linux VM * box.ascii.dev — give your agent a sandboxed Linux machine * Lume — give your agent access to a sandboxed macOS environment And there’s more features added: * AG-UI — render interactive components directly inside the chat * Open Connector integrations — give your agent access to services like Gmail and other platforms * AI Skills — reusable capabilities that agents can discover and execute The connector idea is basically: One MCP to connect your agent to all your other MCPs. The end goal is an agent that isn’t limited to just generating text or code — it can actually use models, tools, services, and computers as part of one workflow. Still very much a work in progress, but I’ve been quietly putting the pieces together. 🚀

i never renew my subs does that count? :D

built an better version of digits.app for my client

Agentic Screen Recorder now has Computer Control Allows AI Agent to do things you are supposed to do when recording. Launch App, Interact with it, Doodle and Annotate as if your the one Doing the Demo

Prompt O Matic is an AI Native Screen Recorder capable of Scribing / Doodling on your screen , Take Screenshot or do Screen Record And let AI Generate all the necessary Screenshots , Prompts. Download it now at: github.com/codeitlikemiley/s…