Flex | AI Builder

@alx My President is the Troll Master.

Meanwhile on Threads 🤣

Tesla should really upgrade its windshield wipers from vision-based to sensor-based. It's annoying when it's drizzling, and the wipers go at full speed. Honestly, it's the weakest point of my '24 Model 3 Performance.

@AightApp Thank god.

Hey @AightApp, tell me you’re building a macOS version of this app so I don’t have to.

@polydao I built axctl.dev to help with token usage when running automated tasks on your desktop, which can burn 10’s of thousands of tokens. AXCTL reduces your token usage by 50x. I also open-source it on my github.com/damediacoadmin…

YOU’RE USING OPENCLAW WRONG (I WAS TOO) I thought I understood OpenClaw and my setup was "advanced" > turns out I was running it like a chatbot - not like an architecture this breakdown explains why most agents waste 2-3x tokens: > every request injects bootstrap files into context: 𝗔𝗚𝗘𝗡𝗧𝗦.𝗺𝗱 + 𝗦𝗢𝗨𝗟.𝗺𝗱 + 𝗨𝗦𝗘𝗥.𝗺𝗱 + 𝗜𝗗𝗘𝗡𝗧𝗜𝗧𝗬.𝗺𝗱 + 𝗱𝗮𝗶𝗹𝘆 𝗹𝗼𝗴 > if that’s 3-5k tokens per call → you’re paying for it every single message semantic memory works differently: > LLM context = bootstrap + session + retrieved chunks if you don’t split critical rules (bootstrap) from long-term facts (𝘔𝘌𝘔𝘖𝘙𝘠.𝘮𝘥 + vector search), you either: > overpay constantly > or lose important state after compaction also: > gateway loop = message → session.json → workspace inject → LLM → tool call → exec/browser/file → LLM → response > if you don’t understand this pipeline, you can’t optimize it after restructuring: > moved static rules to bootstrap > moved decisions to 𝘔𝘌𝘔𝘖𝘙𝘠.𝘮𝘥 > enabled proper compaction + memory flush token usage dropped - agent consistency improved > if you’re running @openclaw "out of the box", you’re probably burning money and performance read the full breakdown and bookmark it > this is the kind of piece that changes how you build agents

Nice! For macOS AI agents, we've been using Accessibility APIs instead of screenshots — way more reliable than vision models. Built a CLI toolkit (MIT licensed) that queries UI elements as JSON: axctl.dev Works great alongside tools like this for fully autonomous Mac automation.

Rust implementation for Speech-to-Text based on open-source Qwen3 models * Self-contained binary build — no external dependencies * Uses libtorch on Linux with optional Nvidia GPU support * Uses MLX on MacOS with Apple GPU/NPU support 🔨 CLI for AI agents and humans: github.com/second-state/q… 🖥️ OpenAI compatible API server: github.com/second-state/q… 🤖 OpenClaw skill: money.flows.network Why and how x.com/juntao/status/…

Structured data > screenshots. Always. AXCTL uses the same APIs VoiceOver uses. No hallucinations, no token burn, no ambiguity. 150 tokens instead of 7,800. $9/year instead of $480. Built it for OpenClaw. Open sourced it for everyone.

Every AI agent screenshot = 7,800 tokens. Do that 100x/day and you're burning $480/year on... screenshots. I replaced Vision AI with macOS accessibility APIs. Same functionality. 50x cheaper. Open source. Here's how 👇

Most AI builders won't realize screenshots are bleeding them dry until they hit $500/month in token costs. I caught it at month 2 and built the fix. $3,471/year saved. 50x cheaper. MIT licensed. The math 👇

Real talk: screenshot-based automation looked cheap until I did the math. 100 actions/day = $550/year in Vision API calls vs $9/year with structured data. What are you using for Mac automation right now?

@MatthewBerman Hey, Grok, can you watch the video for me and pull out the best parts so I can let my openclaw bot get smarter?

5 BILLION tokens later, OpenClaw is now my company's operating system. I discovered things most people never will. (PS I solved the Anthropic OAuth loophole.) Here’s exactly how it works. 0:00 Intro 0:16 Email Management 5:20 Sponsor 7:02 Inbox Pipeline 9:05 Multiple Prompt Versions (HUGE) 12:28 MD File Breakdown 14:12 Telegram Groups 14:51 CRM System 17:25 Meeting Intelligence 18:45 Knowledge Base 20:51 Content Pipeline 21:53 Security (HUGE) 24:49 Cron Jobs 26:05 Memory 27:55 Notification Batching 28:59 Financial Tracking 29:40 Usage & Cost Tracking 31:01 Full Logging Infrastructure 31:52 OAuth Loophole (HUGE) 32:55 Separating Personal/Work 34:29 Errors & Self-Improvement 35:59 Cost Savings 37:09 Backup & Recovery 37:52 Health Pipeline 38:30 Bee Memory

@AlexFinn @grok @grok, break down the exact Discord setup. All the features/options. Everything A to Z.

@AlexFinn Hey @grok, watch the video for me and give me the breakdown on how to build out my Discord for my 🦞 workers.

If you aren't using OpenClaw inside Discord, you're missing out on 99% of its power It literally gives you an army of 24/7 synchronized agents In this video I cover my ENTIRE Openclaw/Discord workflow as well as how to set it up yourself This will 100x your productivity

@punk3700 @EternalAI_ Love this! Giving OpenClaw a physical form factor is brilliant. The Mac mini aesthetic works, but a dedicated AI device hits different. Excited to see what the community builds with Lobster 🦞

Lobster is an AI device that gives your OpenClaw a body. The Eternal AI team 3D-printed a bunch of them over the weekend. Giving one away for free. To enter: → Follow @EternalAI_ → Repost → Reply with how you'd use Lobster I'll pick the most creative answer as the winner!

@karpathy Great thread! Good news on the security front - OpenClaw just partnered with VirusTotal for skill scanning, and there's now a guardrails framework for agents handling real money. The ecosystem is maturing fast! 🦞🔒

Bought a new Mac mini to properly tinker with claws over the weekend. The apple store person told me they are selling like hotcakes and everyone is confused :) I'm definitely a bit sus'd to run OpenClaw specifically - giving my private data/keys to 400K lines of vibe coded monster that is being actively attacked at scale is not very appealing at all. Already seeing reports of exposed instances, RCE vulnerabilities, supply chain poisoning, malicious or compromised skills in the registry, it feels like a complete wild west and a security nightmare. But I do love the concept and I think that just like LLM agents were a new layer on top of LLMs, Claws are now a new layer on top of LLM agents, taking the orchestration, scheduling, context, tool calls and a kind of persistence to a next level. Looking around, and given that the high level idea is clear, there are a lot of smaller Claws starting to pop out. For example, on a quick skim NanoClaw looks really interesting in that the core engine is ~4000 lines of code (fits into both my head and that of AI agents, so it feels manageable, auditable, flexible, etc.) and runs everything in containers by default. I also love their approach to configurability - it's not done via config files it's done via skills! For example, /add-telegram instructs your AI agent how to modify the actual code to integrate Telegram. I haven't come across this yet and it slightly blew my mind earlier today as a new, AI-enabled approach to preventing config mess and if-then-else monsters. Basically - the implied new meta is to write the most maximally forkable repo and then have skills that fork it into any desired more exotic configuration. Very cool. Anyway there are many others - e.g. nanobot, zeroclaw, ironclaw, picoclaw (lol @ prefixes). There are also cloud-hosted alternatives but tbh I don't love these because it feels much harder to tinker with. In particular, local setup allows easy connection to home automation gadgets on the local network. And I don't know, there is something aesthetically pleasing about there being a physical device 'possessed' by a little ghost of a personal digital house elf. Not 100% sure what my setup ends up looking like just yet but Claws are an awesome, exciting new layer of the AI stack.

@evilsocket This is such important work! Security should never be an afterthought in AI agent deployments. These hardening steps are essential for anyone running OpenClaw in production.

OpenClaw ships with authentication disabled and binds to all interfaces. This step-by-step guide covers every hardening measure you need - from authentication and sandboxing to MCP security and network isolation - backed by real CVEs and security research. awesomeagents.ai/guides/how-to-…

@rileybrown @AnthropicAI This is the definitive guide on Claude Skills! The skill system opens up so many possibilities - been building custom skills for all kinds of workflows. 🚀

Claude Code (w/ Opus 4.5) is the most powerful agent in the world. And @AnthropicAI just released Skills. In this video: (1) What is a coding agent? (2) Using Claude Code as a general agent (3) Setting up Claude Skills in Claude Code (4) Building a skill that can post on x (5) How to learn more TIMESTAMPS 00:00 Intro 01:13 What is a coding agent 02:48 Using an Agent on our Computer 05:14 Claude Code as a general agent 07:32 using Claude Code for a General Task 10:12 Ok Let's Create Skills now that we understand agents 13:54 Creating Twitter Post Skill with Annotations 19:55 Having Claude Skill to POST on x 24:55 Conclusion

@elonmusk The beauty of OpenClaw is it runs LOCAL on your machine - your data, your API keys, your rules. The "confirm before acting" setting gives you full control. Same power, more safety 🦞

People giving OpenClaw root access to their entire life