You can approximate this very closely in Replit:

- full VM so can house any sort of input or file w/ great coding agent - generated shareable artifacts with privacy built in - transparently versioned with git with builtin realtime and merge-based collaboration - builtin cloud services: storage, db, AI gateway etc

@patrickc It’s Paperclip

- Ability to manage a set of input files (Markdown or similar), plus other general-purpose context. - Some concept of compiled outputs/inference results (which ideally can be shared externally).

Both of those yes, but agent native: input and output artifacts oriented toward compounding engineering workflows. So output is just reusable context as standards.

https://plannotator.ai/workspaces/

But for better or worse, zero intent to do any orchestration; integration instead.

@patrickc Check out my fork of Multica - added memory, channels, GitHub ship integration for multiple repos, internal and external MCP servers, all in a multi agent, multi human environment that can use oauth for agents instead of api.

https://github.com/prellr/multica

@patrickc You might like something you build and run in codex like this:

@patrickc RIP your DMs

We're building this: https://github.com/gofixpoint/amika

It's very focused on coding use cases right now, but very inspired by the question: what would UNIX look like if it was redesigned after LLM agents launched?

Main idea: new "app" is a VM + agent + your context and files.

Each app has bundled inputs, and you can inject more input files.

Each app outputs "artifacts". You can review/edit/redo these, or run them through an "artifacts CI" of sorts.

You (and teammates) can chat with the agent on the computer, or even SSH into the computer if you want.

@patrickc I'm working on dynamic workflows + typed artifacts @overmindai. coincidently, I saw this excellent manuscript that formalizes some of the ideas I've been thinking about beautifully

@patrickc In my (anecdotal, but enterprise deployed!) experience, this ache is a product grammar / customer mental modeI problem. I landed on something between Slack, Confluence and Figma, built around "work lanes" (a mental parallel to a channel/workspace/folder).

@patrickc Stilla is multiplayer, real-time collaborative, individual level permissioned and version controlled from the ground up across everything (chats, skills, agent workflows, wiki etc)

What are you looking to do?

@patrickc I think you'll enjoy what @sgrove has been working on!

@patrickc i built my own system with similar properties.

just used it to one shot a pokemon ROM binary to WASM static recompilation.

a task that should have taken a team of 10x devs a year to complete, my agent system did it autonomously in ~4 hours

@patrickc Pretty close to what we’re building with @typedotcom, lets you run Claude code / codex on a shared computer w/ snapshots, multiplayer skills, markdown/html, files, and even full hosted apps + deep slack integration

this is part of @inline_chat's vision as the interface + realtime interactions + multiplayer + context/artifacts/storage + share externally. it works much better than slack/discord/etc because threads are the primitive, can be linked/referenced as context, individually shared with external parties, and infinitely scalable while multiplayer with no conflict/bloat as you only see what you need. the agents/workflows building on top of it won't be built-in though. they'll interact via our apis.

@patrickc I built a prototype of this a few months ago. Never got around to finishing it off.

Built it as basically a Markdown only Google Docs with Jupyter Notebook like REPL and other integrations.

There’s definitely something there but the interface needs to be a lot more fluid.

@patrickc It’s a combination of workspace with shared AI Drive/Connections/workflow/agent/multi-agent ststem that publishable to Web/Telegram/Discord. Control by Claude Code/Codex

Here is yt playlist of 1 prompt -> shareable result

https://youtube.com/playlist?list=PLKkXz0f573pk3KkQpVobsMxBcpsjtan_Y&si=E4vjr34Wlwta0QR9

@patrickc we're literally building this at http://AgentSwarms.fyi

visual sandbox where you can wire up agents, manage context/files, and iterate on workflows without setup. outputs are shareable and everything runs in browser

would love to show you what we've got

@aiventurelog @patrickc Okay I'll build one today

Not sure it’s an exact fit but been thinking along these lines with https://schematics.run . The idea is you could define schemas/tools for different files in a file system. Maybe you want *.pdf to generate screenshots then use those to generate markdown and then use the markdown to generate json in a certain shape, track dependencies and relations across all the files etc

@amplituhedron @patrickc @overmindai Glad to hear!