Episode 193: Codex & Claude Code On Your Phone Introducing the OpenAgents Tricoder mobile app, enabling developers to easily use OpenAI Codex and Claude Code from their phone, anywhere in the world. The app is live now in TestFlight, coming soon for Android, and 100% open-source under Apache 2. How it works: 1. Download the app 2. Run a single npx command from your desktop and get a pairing QR code 3. Scan the QR code from the app and start chatting "I'm excited for developers to experience what I experienced last week, where I had probably my single greatest daily increase in my own personal productivity just getting this working, walking around my house and talking to it." Pairing devices can be done in either of two ways: 1. Local network -- Direct pairing over wifi, no accounts needed. Devices must stay nearby on the same network 2. @Tailscale VPN -- Enables you to stay connected anywhere. Leave your desktop on at home, use the mobile app from the grocery store. Just needs a free install of Tailscale on both devices. (Support for other VPNs coming soon) The desktop bridge communicates with Claude Code and Codex via the amazing new Agent Client Protocol from @zeddotdev. We'll add support soon for @opencode and other agents as users request them. Please forgive its very basic UI and inevitable bugs. The codebase is just <1 week old but now that the hard parts are done, we'll be rapidly adding features and polish based on user feedback. For now this app can be a nice supplement to your desktop coding work. Over the coming months, we're excited to explore replacing old-fashioned IDEs altogether. Much more to come. Please send us feedback-- and stay tuned! - Website: openagents.com - Docs: docs.openagents.com - GitHub: github.com/OpenAgentsInc/… - Join TestFlight: testflight.apple.com/join/dvQdns5B https://x.com/openagentsinc/status/1983960929575338281

Yes!

OpenAgents Episode 197: Reverse Engineering Cursor The year is 1900 and the world's "top" investors massively funded the company building the fastest horse. We'll keep building the Model T, but first let's see what we can learn from @cursor_ai. PRs welcome: https://github.com/OpenAgentsInc/openagents/tree/main/docs/re/cursor https://x.com/openagentsinc/status/1989020655602618370

OpenAgents Episode 196: Ditch the TUI Upgrade #1 of 10: "Replace janky fake terminals with a proper desktop app featuring a sidebar, chat history, and widgets for managing long-running agents." Here's what that looks like. And surprise! We're back on Tauri so this will be cross-platform (Windows/Mac/Linux/Android/iOS) from the start. 👍 Big props to @assistantui for a beautiful UI out of the box! https://x.com/openagentsinc/status/1988825875127607363

OpenAgents Episode 195: Designing 10x Better How can our product become ‘10x better’ than the competition? We discuss ten upgrades: 1. Ditch the TUI (and your old IDE) "There's no reason to be in the terminal because you're not using actual terminal commands. There's nothing about the TUI you couldn't also do from a desktop app. Give me a sidebar. Get rid of this screen flicker. I can't copy paste properly. All this stuff's already been solved. Give me a ChatGPT-style desktop app with history on the side, some widgets if I want to manage long-running agents." 2. Go Mobile "I want to have a desktop app that syncs to my mobile phone. So give me the exact same flow, the same UI components, the same controls. If I'm at home doing my usual workflow, let me have the desktop app, but if I want to go to the store, let me be able to do all of the same stuff from my mobile phone." 3. Code Overnight "I want to be able to let stuff run overnight. And not in this hilarious horrible way that you have to do with Codex: continue, continue, continue, continue, continue." 4. CLI Agents as Subagents "You know how Claude will delegate to a sub-agent and it'll have its own context window. When it issues a task, that's what it's doing. Why can't I just have my main OpenAgents chat delegate to Codex, delegate to Claude Code in the context of a single conversation with all of that context available to agents who want it?" 5. History & Memory "Give me the sidebar with the chats so I can actually see what's already happened. And like let me easily reference past chats. There should be a local SQLite database... Right now, all of the Claude Code and Codex conversations are just in these giant JSON blobs, not optimized for discovery at all." 6. Hassle-Free Interop (MCP or...) "All the hoops you have to jump through to get MCP working. People say the tool registry definitions are too long so now you need to create an API from it... stop it. Stop making me think in those terms. If you're in an app, you should be able to easily pull in integrations" - or let the agent do that itself. 7. Embrace Open Source "The leader here on the app side for coding agents is opencode. They've built a massive community of contributors... We want to do that kind of thing, but for our more opinionated agentic flow with OpenAgents." 8. Local+Swarm+Cloud Inference "If I've got a bunch of compute on my computer, if I've got stuff that my Foundation Models API via the M2 chip is able to do for summarization, for updating title summaries-- let's use your local compute where it's needed... There should be a mix of inference possibilities. People should be able to mix and match based on their cost preferences or have that done automatically." 9. Compute Fracking "You -- YOU -- have compute available to you that you are not doing anything with. Why are you not doing anything with it? You've got an M2 chip that's just sitting there. Are you using that compute right now to run agents? No? Then you're wasting electricity. Well there just hasn't been any software for you to run on your computer that was just able to have agents run around the clock, or -- if you don't have any economically sensible use cases -- why can't you sell that to your neighbor who will pay you a tiny fraction of a dollar, tiny fraction of a bitcoin, to use your compute? The big labs are not going to innovate in this direction because they are now drinking the NVIDIA koolaid, but we have no such limitations. So just remember: compute fracking, you heard it here first." 10. Revenue Sharing "To anyone who's like, Chris why are you spilling all the all the juice here? The sauce, you're giving it all away. The alpha, you're just leaking it out here. No, no, no. We're here to build network effects. So we're going to be open source. We're going to be building in public. We're going to monetize this. And how we're going to build the biggest network is because we're going to get developers paid by building an actual marketplace. This is going to solve so many of the coordination problems around why MCPs suck in implementation, because no one has an actual financial incentive to maintain them or make them good. Hey developers, you wrote an agent plugin. Would you like to put that in a registry and get paid a tiny fraction of a cent every time someone uses it? How do you do that? Oh, a built-in bitcoin wallet. We built this already. That's the benefit of having been around for two years and on episode 190+. We've built all of this stuff in prototype form and now we're just mashing it into a single product now that we've got the agentic coding go-to-market piece figured out." --- "What would you add to this list? Please tell us. We are aiming to get at least basic versions of every one of these ten upgrades in place by the end of this year, six weeks from now." WHAT WOULD YOU ADD? 😇 https://x.com/openagentsinc/status/1988293182942228779

Om nom nom

Nice! Feeding this to my coding agents 😂

Cookin

OpenAgents Episode 194: The Trillion-Dollar Question What percentage of global AI workloads in 2030 will run on Apple silicon? ChatGPT Pro thinks the range could be anywhere from 7% to 35%, depending in part on how broadly its edge inference APIs are adopted by developers. We've got a few things to say about that. We demo agentic search through a codebase using nothing but Apple's on-device Foundation Models API. It works-- even while offline. It's local and private with a marginal cost of zero. How far can we take this? What percentage of coding agent workloads can shift from cloud to the edge? We are about to find out. https://x.com/openagentsinc/status/1986655639431500034

Episode 192: OpenAgents Upgrades Itself We ask the OpenAgents Tricoder app to edit its own code. It commands OpenAI Codex over websockets via Tailscale, updating the app over-the-air via React Native expo-updates. Works! No coding or IDE necessary. Shall we automate this? https://x.com/openagentsinc/status/1981533017999814688