PAW v4.0.5 🐾
🧠 9 AI providers
🌐 21+ channels
🎤 Voice + wake word
🖥️ Desktop Hub — Win / Mac / Linux
🔗 MCP · Crews · Deep Research
⛓️ Solana + Purp SCL
🛡️ 6-stage safety. 260 tests.
They ship features. We ship an OS.
https://t.co/tbcTJC9OFl
SOMEONE ACTUALLY MEASURED HOW MUCH DUMBER CLAUDE GOT. THE ANSWER IS 67%.
the data shows Opus 4.6 is thinking 67% less than it used to.
anthropic said nothing until the numbers went public. then suddenly Boris Cherny (creator of Claude Code) shows up on the GitHub issue.
users are calling it "AI shrinkflation" (same price, less intelligence)
we already know from the leaked source code that they have an internal switch that keeps the models working to their full extent for anthropic employees.
in the last week Claude went from WOW to being a more restricted and expensive version of ChatGPT.
people are saying Anthropic is deliberately downgrading Opus to save compute for training Mythos, their next model.
Hey everyone,
I've been working on PAW Agents — an open-source AI framework that tries to solve something that's been frustrating me: you shouldn't need 5 different tools and 12 API keys just to have an AI assistant that works across your daily platforms.
Here's where it's at right now:
WHAT ALREADY WORKS
• 9 AI providers supported — OpenAI, Anthropic, Google, Groq, Mistral, DeepSeek, xAI (Grok), Cohere, and Ollama (fully local, $0)
• 21+ channel integrations — Telegram, Discord, Slack, WhatsApp, LINE, Reddit, Matrix, Twitter/X, GitHub, Email, SMS, Notion, Calendar, MQTT, RSS, REST, WebSocket, CLI, Desktop, Voice, and Browser
• A Desktop Hub (Electron) with a dashboard where you can monitor agents, switch providers, and manage everything from one place — no terminal required
• Built-in MCP server + client support
• Multi-agent Crews so agents can collaborate on tasks
• 45+ tools including web scraping, file operations, browser automation, and Solana blockchain integration
• Voice control with "Hey PAW" wake word
• 3 safety modes: Supervised (confirms actions), Autonomous (runs freely), and Free (no guardrails)
• Research mode and extended Thinking mode for complex tasks
• Plugin system and visual Workflow engine
• Purp SCL — a domain-specific smart contract language for Solana that compiles and validates directly through PAW
• 260 passing tests, zero TypeScript errors
• MIT licensed — completely free, no premium tier, no catch
You can run the whole thing locally with Ollama for free. No cloud, no subscriptions. git clone, npm install, npm start.
WHAT WE'RE WORKING ON
• PAWL — a desktop companion (think Clippy but actually useful and cute) that reacts to your workflow, shows notifications, and makes the experience feel alive
• Improving voice mode reliability across platforms
• Better workflow templates so people can get started faster
• More thorough documentation and setup guides
• Performance optimizations for lower-end machines
WHAT THIS ISN'T (yet)
I'm not going to pretend this is better than everything else out there. It's not. Tools like OpenClaw and others have more polish and bigger teams behind them. But here's what PAW is trying to do differently:
• One hub for everything — instead of juggling multiple apps, PAW connects your AI to every platform you already use, from one place
• Runs locally from day one — Ollama support means you can have a fully functional AI agent setup without spending a dollar or sending data anywhere
• Built to be extended — the plugin system and MCP support mean you can plug in whatever tools or workflows make sense for your use case
• 24/7 virtual assistant capable — like OpenClaw, PAW can run as an always-on agent, but across 21+ channels simultaneously instead of just one interface
The goal is to build something that gives everyone access to powerful AI agents — not just people who can afford premium subscriptions or have the patience to wire together 6 different frameworks.
WHERE YOU COME IN
This is a solo project right now and there's a lot to do. If any of this sounds interesting to you:
• Fork it and build something cool on top of it
• Open a PR — I review everything and I'm happy to work through contributions together
• File issues — bug reports, feature requests, UX feedback, all of it helps
• Try it out and tell me what's broken or confusing — honest feedback is the most valuable thing right now
• If you just want to follow along, star the repo and follow @PAWagents for updates
I built PAW because I wanted it to exist. I'm sharing it because I think other people might want it too — and because the best version of this is one that more people contribute to.
Happy to answer questions, take feedback, or just chat about the architecture. Don't hold back.
---
Links:
🐾 GitHub: https://t.co/XEA4Kww4Mm
🌐 Website: https://t.co/VGC7I2ffzb
🐦 X/Twitter: @PAWagents
Built an open-source AI agent framework that beats OpenClaw in 22 out of 24 categories.
Mandatory safety pipeline. Multi-agent orchestration. Workflow engine. Browser automation. Semantic memory. 30+ tools. Real-time dashboard. 79 tests.
The LLM never touches execution. Full writeup 👇
Introducing PAW agents.
I fixed everything i didn’t like about openclaw.
It’s still a work in progress but tests have provided me with some impressive results so far.
I’d love to collab with more devs in here to make this the most complete agent framework there is.
https://t.co/hYceKGlv1z
Introducing PAW agents.
I fixed everything i didn’t like about openclaw.
It’s still a work in progress but tests have provided me with some impressive results so far.
I’d love to collab with more devs in here to make this the most complete agent framework there is.
https://t.co/hYceKGlv1z
Introducing PAW agents.
I fixed everything i didn’t like about openclaw.
It’s still a work in progress but tests have provided me with some impressive results so far.
I’d love to collab with more devs in here to make this the most complete agent framework there is.
working on something rather interesting.
After taking pictures this weekend while out with my family, I hopped on Lightroom to edit the photos.
My adobe subscription ended.
So I was like, let me check how much I've spent on Adobe software the past 12 years. Let me tell you, it wasn't pretty.
So I decided to make my own Lightroom program.
More updates soon.
Currently working on version 2.1 for my @openclaw tool “Clawtrace”
If you’re running an openclaw agent, i highly recommend trying the tool. Zero dependencies, works 100% offline. No security risks.
Read up on Clawtrace here 👇🏽
https://t.co/9Tef3NVJam
#openclaw
I want to keep improving Clawtrace.
So what will i do?
1: Make a new @openclaw agent
2: Let it collect its own data to put into clawtrace
3: let it give me real-time updates on how it thinks i could improve clawtrace.
Should be interesting
I think I might have found a way to revolutionize how you look at your openclaw agent(s) + how to optimize it in terms of token usage.
This could be massive in my opinion
🚀 Introducing CLAWTRACE: the ultimate zero-dependency, client-side trace visualizer for OpenClaw agents! 🦞
CLAWTRACE lets you instantly load, explore, and debug agent traces right in your browser—no installs, no servers, no risk. All your data stays 100% offline, guaranteeing absolute privacy and security 🔒.
Why is it a game-changer?
✨ Visualize complex agent workflows with ease
✨ Perfect for security, research, and development
✨ Open source, MIT licensed, and community-driven
✨ Works on macOS, Windows, Linux, and any browser
Ready to supercharge your OpenClaw experience? Clone the repo, open clawtrace-info.html, and start tracing!
It has an actual interface you can use to easily navigate through the mind of your agent.
If you need anymore info on how to get started or general info check the GitHub repo 👉https://t.co/8HjJJIwI7t
It's fully open source and MIT licensed. Feel free to use it, re-work it and give me feedback!
Empower your agents, protect your data, and unlock new possibilities—all with CLAWTRACE!