This video introduces "Cloud Code Local," a new system that allows users to run a 122 billion parameter AI model entirely on their MacBook's Apple Silicon chip for free. It leverages Apple's MLX framework and bypasses the need for cloud-based APIs, thus eliminating API fees and ensuring data privacy by keeping everything on the user's machine. The system is highlighted as being significantly faster than existing cloud solutions, including CloudCloud Opus, achieving 65 tokens per second locally compared to Opus's 40 tokens per second. The key innovation is the elimination of a proxy bottleneck; instead, Cloud Code Local communicates directly with a compact 200-line Python server running on the user's GPU. This direct communication results in a 7.5x speed improvement, reducing a task that previously took 133 seconds to just 17.6 seconds. The system also offers remote control capabilities, allowing users to interact with it from their iPhone via iMessage while the Mac performs the heavy computation.

Hasan Toor

Someone just made Cloud Code run 100% locally on a MacBook for $0 a month. [music] No API fees, no cloud or data leaving your machine.

It's called Cloud Code Local and it runs a 122 billion parameter AI model entirely on your [music] Apple Silicon chip using Apple's own MLX framework and it's actually faster than CloudCloud Opus. 65 tokens per second locally versus 40 tokens per second from Opus in the cloud at $0 a month.

The reason it's so fast is because everyone else building local cloud code setups uses a proxy to translate between Claude's API and local [music] models. And that proxy is the bottleneck. This project kills the proxy completely.

Cloud code talks directly to a 200line Python [music] server running on your GPU. And a task that takes 133 seconds with other setups takes just 17.6 seconds here. And it's 7.5 times faster.

You can even control it from your iPhone [music] over iMessage while your Mac does the work at home. Comment local and I'll DM you the link.

This Free Tool Runs Claude Code Faster Than Cloud Opus On Your MacBook #ai #artificialintelligence

Someone just made Cloud Code run 100%
locally on a MacBook for $0 a month.
[music] No API fees, no cloud or data
leaving your machine. It's called Cloud
Code Local and it runs a 122 billion
parameter AI model entirely on your
[music] Apple Silicon chip using Apple's
own MLX framework and it's actually
faster than CloudCloud Opus. 65 tokens
per second locally versus 40 tokens per
second from Opus in the cloud at $0 a
month. The reason it's so fast is
because everyone else building local
cloud code setups uses a proxy to
translate between Claude's API and local
[music] models. And that proxy is the
bottleneck. This project kills the proxy
completely. Cloud code talks directly to
a 200line Python [music] server running
on your GPU. And a task that takes 133
seconds with other setups takes just
17.6 seconds here. And it's 7.5 times
faster. You can even control it from
your iPhone [music] over iMessage while
your Mac does the work at home. Comment
local and I'll DM you the link.

Get Our Mobile App

This Free Tool Runs Claude Code Faster Than Cloud Opus On Your MacBook #ai #artificialintelligence

Transcription