Why are the models rate limited so heavily in this age of agentic coding where coding loops are supposed to run unattended.
Deepseek V4 flash has better quality output than most of the models.
Is this the future of EU AI ? If it is It's a shame...
Seeking advanced system thinking professional, I need your advices on how if we start from scratch what is the optimal way for a looping coding agent to access things and take action ?
Is it software abstraction as a CLI in natural language ?
Is it nicely structured JSON exchange with a system ?
HTML/CSS (I highly doubt it ehe)
Something else entirely ?
How can we reinvent the digital world so agent interact with software and the web ?
Open to you wildest ideas
Hey, demo video above.
I've been using Claude Code for a lot of non-developer tasks, just a prompt and a few files (.md, .json, .py, etc). I found adopting developer logic for business usage quite efficient.
However I was unable to package those "filesystem agents" into something reusable, host it, and run it. So I started building this project.
Each terminal agent gets its own isolated filesystem, and you can call it over API/MCP or through the chatbot inside the app.
The app allows to wire agents into a DAG graph and let them loop across different workspaces.
Somewhere along the way I got a second idea: give the agents a shared drive on top of their "local" filesystem. Goal was to keep their own private workspace but agents could also pass files around to work together.
It's early and rough. Mostly putting it out there to see what people think, mainly curious whether the shared-drive/graph part actually lands or if I'm overcomplicating it.
Hello community,
Today I'm sharing a small app I built.
I feel like tools like Claude Code and Opencode CLI agents are primarily optimized for complex software and a developer workflow, working inside a local codebase, and deploying through Git-based setups.
What I needed was a way to spin up a hosted solution with multiple isolated “workspaces”, each with its own directory, agent, files, and terminal, So I could expose each one as a callable service.
Think of each workspace as a sandboxed environment owned by a persistent agent with its own appropriate toolset, grounded files, folders, dependencies and terminal and agent call being accessible through an API.
My goal was to turn CLI agent frameworks into reusable, service-oriented automation units.
I built PAODO Workspace as a personal power tool, and I'd love to open-source it and build on it with others.
https://github.com/alxcls/PAODO_WS
Any feedback and use cases would be much appreciated.