The idea behind the project is much influenced by @MulticaAI , but with a simpler yet still powerful abstraction. I just think that a central server is not always necessary, and we might just drive a team of agents with Github only.
Introduce Grovie, a local-first, GitHub-native multi-agent platform.
With a simple and free setup, you can fully utilize the power of local agents like Codex and Claude code across multiple machines, and keep all information inside your github repo.
https://t.co/m2gF3JbUUI
There is no central server needed, and it's absolutely open source and free. All you need to do is to start grovie daemon in your machine. You can assign multiple local agents with different instructions across multiple machines. At present, Codex/ClaudeCode/Pi are supported.
@karpathy Following the workflow, I had Moxt turn 200+ LLM papers into a wiki—and visualize the whole field, clearly mapping how concepts and papers connect.
Better yet, the entire team can now access and collaborate on this shared knowledge!
https://t.co/ORDJslDN5l (public, no sign-up)
@karpathy I made this workspace public so you can see the full pipeline—from raw papers to summaries, concepts, and the visualized connections between them.
You can also check the skill that drives the entire process:
https://t.co/O427IYSqHD
@karpathy Following the workflow, I had Moxt turn 200+ LLM papers into a wiki—and visualize the whole field, clearly mapping how concepts and papers connect.
Better yet, the entire team can now access and collaborate on this shared knowledge!
https://t.co/ORDJslDN5l (public, no sign-up)
What if every person on your team had an AI teammate — one that thinks, writes, builds, and remembers everything?
Then more show up. AI researchers. AI operators. AI strategists.
All working right alongside you.
Build your AI team here. This is Moxt.
→ https://t.co/AW4bSLwWon