JustVugg

Solo founder.

[JustVugg @justvugg]

We're building the universal CLI layer for AI agents.

MCP servers. Custom scripts. One shell.

If you like this project support us → https://t.co/EDg425CZEH

Let's turn every tool into a command, together.

#MCP #OpenSource #CLI

[JustVugg @justvugg]

@antirez If I'm not mistaken they use swarm too. Anyway kimi 2.6 is really amazing in coding and more.

[JustVugg @justvugg]

@TheEcomNomad Thank you. The fail-closed approach is exactly what makes judicex different. In legal work, generating plausible but unverified answers isn’t innovation — it’s risk. We prioritize evidence grounding and explicit uncertainty over forced outputs.
Self-hosted. Works with any LLM.

[JustVugg @justvugg]

judicex is a production-ready open-source Legal AI workspace. It supports evidence-grounded drafting, matter analysis and verifiable answers with a fail-closed approach.
Self-hosted, local SQLite. Works with any LLM or none.
https://t.co/R2kFWC5ORk
pip install -e .

[JustVugg @justvugg]

@gregisenberg I’m taking advantage of your post, the MCP discussion and getting found by agents to show you my Gutenberg-CLI project. Take any OpenAPI API, website or other source and transform it into a CLI or MCP for your agents. There are already usable library!

https://t.co/gR4rzplVRO

[JustVugg @justvugg]

@TheEcomNomad Exactly, signal vs noise is the hard part.
agentmw tackles it by detecting failures in real-time, aggressively compressing stale context, and distilling just 1-3 high-value patterns into a reusable library — instead of dumping everything.

[JustVugg @justvugg]

agentmw is a production-ready open-source middleware for AI agents. It detects failures in real time, compresses stale context, and builds a persistent reasoning library that improves with every execution.
Works with any LLM and framework.

https://t.co/VsLHSYEjwM

[JustVugg @justvugg]

Since everyone is showing their own version of Ferrari's AI-generated light, I'll show mine too. Ferrari certainly achieved one thing. Everyone's talking about Ferrari.

[JustVugg @justvugg]

@antirez have you tried duck duck go?

[JustVugg @justvugg]

@rauchg https://t.co/R2kFWC5ORk

[JustVugg @justvugg]

@rauchg https://t.co/gR4rzplVRO

[JustVugg @justvugg]

Judicex — open-source Legal AI workspace for lawyers.
Strictly grounded answers with verifiable citations.
Evidence-based drafting and matter analysis.
Fully private and self-hosted.
Repo: https://t.co/R2kFWC5ORk
What’s your biggest pain point with legal AI tools right now?

[JustVugg @justvugg]

@AniketVarshne cli-use can be installed for now by downloading the project and you should tell agents to use it and discover it while for mcp we added discovery which via glama api allows you to find mcp you need to use with cli-use!

[JustVugg @justvugg]

Turn any MCP server into a real CLI in one command.

cli-use add fs /tmp
→ cli-use fs list_directory

Lower token usage, faster agents, zero extra dependencies.

https://t.co/EDg425CZEH

#AI #CLI #MCP

[JustVugg @justvugg]

@antirez I've used it for very small projects. However, from an agentic point of view, I was very impressed. As for projects, I've always tried small projects and in various debugging situations.

[JustVugg @justvugg]

@foxlab_ai Thanks for sharing my project!!

[JustVugg @justvugg]

@pdfnative Perfect!!

[JustVugg @justvugg]

Gutenberg CLI turns any public API into proper, verifiable tools.
Clean Go CLI, MCP server, Claude skills, SQLite cache + real proofs.
Dry-run by default. Finally tools you can actually trust.
https://t.co/gR4rzplVRO
#AI #MCP #DevTools #OpenSource

[JustVugg @justvugg]

@ScuffCrypto Absolutely! We're building the universal CLI layer for AI agents.

MCP servers. Custom scripts. One shell. Everything as simple as possible!

[JustVugg @justvugg]

@m13v_ 100%. the hard part probably isn’t the compiler itself — it’s ownership.

once agent workflows become production infrastructure, the eject step starts looking less like “prompt engineering” and more like release engineering / ops hygiene.

[JustVugg @justvugg]

@m13v_ yep — once workflows hit CI/repeated automation territory, the economics change completely.

agents are amazing for discovering the path.
terrible for being the path forever.

[JustVugg @justvugg]

@m13v_ exactly. once a workflow converges, keeping it inside an agent loop is mostly paying inference costs for orchestration you already learned.

at that point the optimal abstraction starts looking a lot more like unix than like autonomous agents.