c*******

if you steal from the shape store they send you to prism

@TermiusHQ @Tailscale what's the easiest way to use a setup like this where multiple people have access the same codex?

Termius + Tailscale + tmux You don't need your laptop to build with Claude Code, Codex, or any other AI coding agent: → Run your agent inside tmux to keep the session alive → Use @Tailscale for secure access to your laptop → Connect with Termius over SSH from iPhone, iPad, or Android Start coding at your desk. Continue on the go.

@kazuki_sf_ is there a reason you don't use the native remote codex feature? or run a single instance of codex at home, and use native apple screensharing to access it, and then pick up remotely from your phone?

I open sourced a small tool I had been using only for myself. The problem: I would be deep in a Claude Code session on my laptop and have to leave. Closing the lid meant losing momentum, so I either stayed put or started over later. session-baton hands the live session to a machine I keep switched on at home. It carries three things: my code including uncommitted edits, the full conversation with the agent, and a baton that marks which machine is in charge. Only one machine holds it at a time, so two divergent copies never quietly merge. Back at my laptop, one command brings it home and I resume where I stopped. While I am out, I can drive the session from the Claude app on my phone. What is different from a cloud agent service: it runs over my own SSH between two machines I already own. No vendor cloud, no server to operate, no account, no broker. My real local environment comes along, the repo, the .env file, the local tools, because the machine on the other end is mine. Who it is for: one developer with a laptop and an idle machine at home, that old MacBook in a drawer. If you need a fleet of parallel agents, guaranteed uptime, or team controls, you want something bigger. None of the ideas here are new. Remote development over SSH is old, and I am sure others have built versions of this. I wrote it because I wanted it for my own workflow, then spent the extra time to make it safe enough to hand to someone else. Dependency free bash, MIT licensed, and the repo publishes its own hardening notes, including the gaps that are still open. Let's token maxxing with session-baton + @claudeai github.com/kazuki-sf/sess…

@mfateev @GeoffreyHuntley why wouldn't someone just run codex in the cloud on a vm and have all of these agents running 24/7?

@GeoffreyHuntley There are many cases when running an agentic loop inside the Temporal workflow instead of activity is a better option. This avoids loss of LLM and tool results on failure as well as offers better observability, long running tools, humans in the loop integration.

second hot take, as it seems the first hot take landed. n8n was a silly idea, which i’ve always suspected was developed by young folks in silicon valley who are all in on the LLM craze who have never worked in corporate in their life thus didn’t have the knowledge that service busses are a solved problem and there’s deep prior art in the spooky land of enterprise. look up anything in the @dotnet space such as mass transit, nservicebus, [1] wcf if you wanna steal ideas for prompting. if you want a modern plug and play choice, @temporalio exists. use it and have a job invoke an agent as a process. don’t make the entire service bus non-deterministic. [1] lots of bad memories here, not great tech impl but the theory / education is generally on point.

@martin_casado the only thing that matters in the harness race is whoever can implement true multiplayer first google docs level coworking and edit history that lives outside of github people want to be able to seamlessly collaborate on projects natively to the harness they're building in

On harnesses, I vacillate between three beliefs: - the less harness, the better. Models are the magic - post training a model and harness is dramatically better and the model providers win - harnesses have real independent value from the model I have no idea which is right.

@0xPaulius can multiple people be monitoring the same codex instance choose how to steer the agent?

not seen codex/cursor do this built this app to run my fleet of agents while on the train instead of doomscrolling let’s u test Web/iPhone apps on device mac stays at home

being able to run other people's agent on your own machine, using your own codex sub but in the chat setting, it would allow someone to steer the agent in a certain way the multiplayer aspect that is missing is that multiple people can't steer the ship, it's still one person prompting or invoking their agents, no coworking past that

@organicdefector You mean running on others machines or just the Agent but on their own machine?

Buzz Desktop v0.5.0 is out! 🐝 Invite people with use-limited links, filter search with from:/in:/before:, bring your own agent harness (BYOH), a redesigned agent runtime settings page, a smarter project Inbox — plus a pile of stability fixes for Linux, Windows & scrolling. github.com/block/buzz/rel…

@edgarpavlovsky another hurdle is figuring out who to credit for a certain outcome when multiple employees are steering an agent

multiplayer ai is going to bring some really interesting agent custody problems i can see standard access permissions: view, operate, own an agent then you go deeper: 1) the agent's identity and core memory will be durable. what does that mean for information sharing through their core memory? 2) the agent will have cross-conversation context. when does it use or share context from which conversations, to whom (does the agent internally using context vs. sharing it explicitly make a difference? does it leak through reasoning? can it differentiate between using information in reasoning that the external user doesn't see explicitly vs reasoning they do?) 3) company brain access, agent learnings, agent dreaming all have the same problem (they're all information resources) sharing agents cross-company will then take this to an entirely new level multiplayer ai is really a permissions system problem

@wesbillman will you add the ability to invoke other teammates agents?

Buzz Desktop v0.5.0 🐝 Early support for your own harness is here. Try Cursor, OpenCode, Hermes, OpenClaw, or a custom ACP runtime. Feedback welcome! Community AI models now team up for stronger answers. Plus Inbox and Linux improvements. github.com/block/buzz/rel…

@JustJake with "shared context chats" are you referring to multiplayer harnesses?

People are truly not ready for the speed gains of: - Shared context chats - Recursive efficiency improvements - All the inference capacity coming online We truly are approaching the age of 10,000x more software

@edgarpavlovsky could've buzz solve this if they allowed other people in your community to invoke your agents?

We haven't seen great permission architecture for multiplayer AI yet The reality of any shared resource in a company is that it needs to be filtered through permissions of the user - human or agent. - The company brain should look different to every user - The skills library should look different to every user You'll also have two layers to work through sometimes: the permissions of the agent and the permissions of the humans who can interact with that agent An entirely new control plane problem we've never seen before

@liamihorne can anyone in the chat invoke the agents and steer the prompting?

Buzz is interesting, but the friction of moving off Slack seems too high for most teams Centaur still seems dramatically underrated for how good it is - a framework for multiplayer agents in your team’s slack centaur.run

@MOXDigital @liamihorne multiple people using the same agent being able to steer the conversation would be huge buzz only allows the person that added the agent to talk to it though, other people can't invoke your agents

@liamihorne what's the actual multiplayer part look like - agents talking to each other, or just multiple people using the same agent

@wadeowenwatts35 it still doesn't allow your teammates to invoke your agents only the person that added the agent can call to it the first harness that allows anyone in the chat to call specific agents and lets anyone steer the prompting will be NUCLEAR

The multiplayer agent harness is the real idea. Models are getting cheaper and more interchangeable; coordination, shared context, approvals, and proof of work become the product. Slack organized people. This could organize people and agents.

@edgarpavlovsky this is the only thing that matters buzz fell short because it doesn't allow anyone in the conversation to steer the agent only the agent owner can invoke it, I want to build agents that anyone in my team can call to

multiplayer ai shared context shared conversations shared artifacts multiplayer ai

@daria7y @trymira does this allow anyone in the chat to steer the agent and use ChatGPT subscription pricing?

YC just put "Multiplayer AI" on their wishlist. It's the bet we already made. @trymira lives in Telegram group chats, agent and people in one thread. The single-player chatbox was always the temporary shape.

@CalvinGrunewald welcome, now build a multiplayer agent harness

1/ Wanted to formally say hello on X! I'm an ex-Big-tech VP returning to full-time builder. And yeah, I’m loving every second of it. After years mostly lurking here, I’m finally posting properly. I’m a founder at ArchAstro. Here’s the short version of how I got here.

@punit_arani sol 5.6 ultra codex work one shotted white collar america

Company brain this. Company brain that. Who’s building the company hands? Who’s building the infrastructure that actually does the work rather than just think?

@jaredLunde for a smaller team, would running codex in the cloud be a solution for this? we want persistent chats, working in parallel, being able to choose who steers the prompt next while being able to see what teammates did the last move codex on a Mac mini, that our team remotes into

All agents will be cloud agents