On device
$0
Per token. Bring your own API keys at no markup.
Serious AI work that never leaves your Mac.
Frontier models and real tools, running on your own device.
- Up to 2x faster on Apple Silicon
- Works offline
- Nothing leaves your Mac
- Open source
Local execution
Runs here. Stays here.
- You ask
- Model on your GPU
- Tools in the sandbox
- Your files
Cloud model
Off. Nothing sent.
Security
Locked inside Apple's sandbox.
Hardened Runtime
App Sandbox
Folders you share
The agent
The agentCan
- Use folders you share
- Run Python, Node.js and shell
Can't
- Read your whole disk
- Launch other apps
- Use the cloud unless you choose
Fully open source, so you can check all of this yourself. View the code on GitHub
Models
Sized to your Mac.
Mac memory
Qwen 3.8 27BMin 16 GB · Rec. 32 GB10–21 GB on disk
DeepSeek 4 Flash 0731Min 32 GB · Rec. 48 GB89 GB on disk
DeepSeek 4.1 Flash 1017Min 32 GB · Rec. 64 GB376 GB on disk
SSD streaming keeps part of a model on your SSD and loads it as needed, so big models run on smaller Macs.
Or bring the cloud
- Cloud Compute · 32 models
- ChatGPT
- Claude
- OpenRouter
- OpenAI-compatible
Built-in skills
Watch it do the work.
Saturday & Sunday
The weekend bake
Pricing
Free on your hardware.
Cloud credit
$9.99
Top-ups from $9.99 for hosted models. No subscription.
Teams
Let's talk
Setup, custom skills, on-prem.
Self-serve
Plan your setup.
Official guides
Start with a useful task.
Clear setup steps, from the Draw Things team.
Get started on Mac
Choose a model, grant a workspace and check your first result.
Offline use & privacy
Separate local inference from network tools and optional cloud providers.
Choose a model
Understand unified memory, SSD storage and checkpoint downloads.
Practical workflows
Work with documents, spreadsheets, code and research, with review steps.
About Local Code & official downloads
Who makes it, where to get it and which platforms are available.
Common questions
What is Local Code?
A private AI agent for Apple Silicon, made by the Draw Things team. Frontier models and real tools, running on your own device. Meet Local Code.
Does it work offline?
Yes. Once a model is downloaded, Local Code works with no connection at all. For air-gapped setups, talk to us. See the privacy and offline guide.
Which Macs can run it?
Apple Silicon with 16 GB or more. Qwen 3.8 27B runs from 16 GB and is recommended at 32 GB. DeepSeek models run from 32 GB by streaming from your SSD; we recommend DeepSeek 4 Flash 0731 at 48 GB and DeepSeek 4.1 Flash 1017 from 64 GB. Choose a model.
How does DeepSeek run on a Mac with less memory than the model?
With SSD streaming, which Machine Settings calls Weights Streaming. DeepSeek models are mixtures of experts, and each token uses only a few of them. Local Code keeps the most-used experts in memory and reads the rest from your SSD. Giving them more memory makes the model faster: with 40 GB, 94% of the experts DeepSeek 4 Flash 0731 needs are already in memory. Less leaves more room for your other apps. How Weights Streaming works.
What does it cost?
On-device models cost nothing per token, and your own API keys have no markup. Hosted models use credit top-ups ($9.99, $19.99 or $99.99), no subscription.
