Short answer: Local Code is a private, open-source AI agent from the team behind the Draw Things image app, released to everyone on 28 October 2026. It runs open models such as DeepSeek 4.1 Flash and Qwen 3.8 directly on your Mac, with Python, office tools and a browser built in, so it can turn your files into charts, decks, documents and videos without sending them to the cloud. Local use is free.
By Agent Playbox Editorial · Published · Updated
Local Code AI agent: private, open source, on your Mac
Key facts
- Released to everyone and made fully open source on 28 Oct 2026, after a public beta that began on 12 Sep 2026 (releases.drawthings.ai)
- Combines an Apple Silicon inference engine, a coding harness and an agent runtime with Python, Node.js, a shell, LibreOffice, FFmpeg and a browser (github.com)
- Local models range from Qwen 3.8 27B (16 GB minimum) to DeepSeek 4.1 Flash 1017 (376 GB on disk, 96 GB recommended) (getlocalcode.app)
- Works offline; nothing leaves your Mac unless you pick a cloud model (getlocalcode.app)
- Runs inside Apple's App Sandbox: no full-disk access, no launching other apps, only folders you share (github.com)
- Free on your own hardware; optional cloud credits from USD 9.99 or your own API keys at no markup (getlocalcode.app)
- Open source under the GPL-3.0 licence (github.com)
What did the Draw Things team release on 28 October 2026?
The makers of Draw Things, a well-known Mac app for AI image generation, released Local Code to the public on 28 October 2026 and made the entire app open source (Draw Things). Local Code had been in public beta since 12 September (MacRumors). You can download it directly from getlocalcode.app today, with a Mac App Store release coming soon (Draw Things).
The pitch fits in one line from its homepage: “Serious AI work that never leaves your Mac” (Local Code). The team stresses that it all works without external installations, sign-ins, API keys or external MCP servers (Draw Things).
What can the Local Code AI agent do?
Real work on your own files
Local Code can generate images, videos and infographics with precise charts, create slides and explainer videos with motion graphics, write or review Python and JavaScript, and update spreadsheets, Word documents and PowerPoint files (Draw Things). Example tasks on its site include charting monthly revenue from a spreadsheet and flagging anything odd, turning next week’s calendar into a walkthrough video, and building a shareable web page for a bakery’s weekend menu (Local Code).
Your personal apps, with permission
You can connect it to Photos, Messages, Mail, Notes and Calendar and ask it to summarise, edit or reorganise content where supported (Draw Things). According to its documentation, Notes, Mail and Messages are read-only, while Calendar and Photos can be edited, and each works only after you grant access (GitHub). It can even drive a built-in browser to visit a website, log in and make an appointment (Draw Things).
Everything built in
Under the hood, one app bundles three things: an inference engine tuned for Apple Silicon, a coding harness tuned to each model and an agent runtime with Python 3.14, Node.js 24, a shell, LibreOffice, FFmpeg, a WebKit browser and the Draw Things command-line tool for images and video (GitHub). Our research analyst and content producer templates are good starting points for this kind of work.
Which models does Local Code run, and on which Macs?
Local Code sizes models to your Mac. Qwen 3.8 27B needs at least 16 GB of memory (32 GB recommended) and 10 to 21 GB of disk; DeepSeek 4 Flash needs at least 32 GB (64 GB recommended) and 89 GB of disk; and the flagship DeepSeek 4.1 Flash 1017 needs at least 32 GB (96 GB recommended) and 376 GB of disk (Local Code). An SSD streaming feature keeps part of a model on your SSD and loads it as needed, so big models can run on smaller Macs (Local Code).
The team says its quantised DeepSeek 4.1 Flash runs at usable speeds with as little as 36 GiB of resident memory, at around 350 tokens a second for prompt processing and about 18 for generation, and that its benchmark results closely match the officially reported ones (Draw Things). These are the developer’s own figures. If you prefer, you can also pick a cloud model from Anthropic, OpenAI, OpenRouter or any OpenAI-compatible server (GitHub).
How private and secure is Local Code?
With a local model, inference runs on your device and works offline, so prompts and files are not sent to a cloud AI provider (GitHub). If you choose a cloud model, your prompts, file contents and tool output do go to that provider, and the app says so plainly (GitHub).
The agent’s tools are compiled into the app and run inside Apple’s App Sandbox with Hardened Runtime on the Mac. The agent has no full-disk access, cannot launch other programs and works only in folders you share (Local Code). Because the full source is published under GPL-3.0, anyone can check these claims (GitHub).
How much does Local Code cost?
Running on your own hardware is free. If you want hosted models, you can bring your own API keys at no markup or buy prepaid cloud credit from USD 9.99, with no subscription. Teams can ask about setup, custom skills and on-premise installs (Local Code). Its site says it is ready “for Mac, iPad and iPhone”, with a direct Mac download available now (Local Code) and a Mac App Store release coming soon (Draw Things). Funding and user numbers are unknown.
How does Local Code compare with Underdog and Perplexity Computer?
On-device agents are having a moment. Underdog, which opened an invite-only beta on 6 October, is a 100% local personal AI focused on email, calendar, meeting notes and errands (TechCrunch). Since 20 October, Perplexity Computer can also run fully on Apple silicon Macs for Pro, Max and Enterprise subscribers (Perplexity).
Local Code stands out for being fully open source, free on your own hardware and packed with a complete toolbox for documents, data, media and code. See Local Code vs Underdog and Local Code vs Perplexity Computer for side-by-side facts, and read our story on Perplexity Computer going local.
What should you watch next?
The next milestone is the Mac App Store release, which the team says is coming soon (Draw Things). Because the code is open, community additions are also worth watching: the team invites users, or Local Code itself, to clone the repository and adapt it (Draw Things).
We will keep the Local Code profile updated as new versions ship. For more private, on-device options, see our Underdog launch story and the full agent directory.
Timeline: how did we get here?
- : K-AI: a personal AI agent that keeps its memory on your Mac. K-AI went live in public beta on 7 November 2026, free on Mac.
- : Aside Computer Use takes its AI browser agent into desktop apps. Aside announced Computer Use on 6 November 2026: its AI browser agent can now work in desktop apps on Mac and Windows, running in the background across multiple apps on macOS.
- : MyHeritage launches GAIA, an AI agent for family history research. MyHeritage launched GAIA, its Genealogy AI Agent, on 4 November 2026.
- : Draw Things releases Local Code, an open-source private AI agent for Mac. The Draw Things team released Local Code on 28 October 2026 and made it fully open source.
- : Perplexity Computer now runs fully local on your Mac. Perplexity said on 20 October 2026 that its Computer agent can now run entirely on Apple silicon Macs: the model, the agent and the task stay on the device, working across local files and apps like Notes, Messages and Mail.
Agent snapshot: what do we know so far?
Facts below come from the agent profiles in our directory, each linked to its source.
Local Code at a glance
- Vendor: Draw Things team (Liu Liu and team, makers of the Draw Things image app)
- Where you use it: Mac app (direct download; Mac App Store release coming soon); the site also lists iPad and iPhone
- Pricing and availability: Free on your own hardware (USD 0). Optional cloud models: bring your own API keys at no markup, or prepaid cloud credit top-ups from USD 9.99, no subscription. Team setups on request
- Funding: Unknown
Underdog at a glance
- Vendor: Conway Research (founded by Sigil Wen, per TechCrunch)
- Where you use it: Mac (invite-only Apple silicon beta); iPhone and iPad on the App Store (no invite needed per site FAQ); Windows BETA; Linux and Android on the way (underdog.ai). TechCrunch (6 Oct 2026) described Macs and Windows initially with Linux, iPhone and Android coming soon.
- Pricing and availability: 100% free because it runs on your computer (underdog.ai/pricing); monetization is a tiny take when the agent buys something with a one-time card (site FAQ). Never ad-supported (TechCrunch).
- Funding: First round raised April 2026 (underdog.ai); a16z led per a16z announcement (Chris Dixon) and TechCrunch; amount and valuation undisclosed. Investors named on site/TC: Andreessen Horowitz, Khosla Ventures, Hummingbird, Anthology fund (Menlo + Anthropic), SV Angel, Breakneck; angels include Patrick Collison, Naval Ravikant, Guillermo Rauch, Noam Brown (and Deedy Das per TechCrunch).
Perplexity Computer at a glance
- Vendor: Perplexity
- Where you use it: Web, iOS app, Perplexity desktop app for Mac and Windows (Personal Computer), Slack; fully local mode on supported Apple silicon Macs
- Pricing and availability: Included in Perplexity Pro (USD 17/month billed annually, with weekly Computer usage) and Max (USD 167/month billed annually, 10,000 credits a month), plus Enterprise Pro and Enterprise Max. Extra credits: 100 credits = USD 1. Fully local tasks on Mac do not use Computer credits
- Funding: n/a (Perplexity product)
Full Perplexity Computer profile
Related comparisons and guides
- Local Code
- Local Code vs Underdog
- Local Code vs Perplexity Computer
- Underdog
- Perplexity Computer
- Perplexity Computer runs fully local on Mac
- Research analyst template
Agents in this story
A private, open-source AI agent for Mac, iPad and iPhone that runs frontier-class open models on your own device, with Python, Node.js, a shell, office and media tools and a browser built in, so it can turn files into charts, decks, documents, videos and code without sending them to the cloud.
100% local personal AI for email, calendar, meeting notes, dictation and errands — runs on your device, not in the cloud.
Perplexity's agent for end-to-end projects: it plans the work, picks the right model for each step, connects to hundreds of apps and hands back finished research, reports, dashboards and apps. Since 20 Oct 2026 it can also run fully on your Mac.
Frequently asked questions
What is Local Code?
Local Code is a private, open-source AI agent for Mac, iPad and iPhone from the Draw Things team. It runs open models on your device with Python, office tools and a browser built in. It was released to everyone on 28 October 2026.
Is Local Code free?
Yes, running models on your own hardware is free. Optional cloud models use your own API keys at no markup or prepaid credits from USD 9.99.
How much memory does my Mac need?
At least 16 GB for Qwen 3.8 27B. DeepSeek 4 Flash and DeepSeek 4.1 Flash 1017 need at least 32 GB, with 64 GB and 96 GB recommended.
Does Local Code send my files to the cloud?
Not with a local model: it works offline and files stay on your Mac. If you choose a cloud model, prompts and files go to that provider.
Is Local Code open source?
Yes. The full source code is on GitHub under the GPL-3.0 licence.
Sources
Updated