Stop Counting Tokens
Start Coding Without Limits
A local AI coding assistant that runs entirely on your Mac.
The future of AI is local
AI was supposed to make developers unstoppable
Instead it gave us new problems
- Arbitrary usage limits that wreck productivity
- Price hikes and unpredictable bills that blow up budgets
- Questionable privacy practices that copy and expose code
Finally, a Capable Local Coding Assistant
Native, offline, private, powerful, and easy to install
Get more done, without counting tokens.
function Button({ children, onClick }) {
return (
<button
className="btn-primary hover:scale-105"
onClick={onClick}
>
{children}
</button>
)
}
The Cortex LLM is more powerful than anything you can run locally today
We built an innovative pruning, healing, and PostNAS pipeline that replaces large quadratic operations with linear ones, significantly shaving memory requirements so we can ship you a larger coding LLM that uses fewer resources.
Cortex writes better code with less context rot
Our proprietary TRACE context engine is real-time and git-aware providing the right code and documentation for the next step, increasing accuracy and significantly reducing deviations.
The Local Difference
Powerful and Uncompromising
Unlimited Usage
No daily limits, no token counts, no "upgrade to continue" interruptions.
Complete Privacy
Your code never leaves your machine and no shady terms of service.
Native Performance
Built entirely on Rust, Swift, and Apple's hardware apis, Cortex is blazing fast.
Powerful Tool Calling
Cortex has been purpose built for tool use like web search, documentation fetching, codebase search, and more.
Parallel Agents
Run multiple AI assistants simultaneously — refactor code while building APIs while writing tests.
Zero Environmental Impact
No server farms, no water usage — just your laptop's normal power draw.
Compare the Real Costs
Get more from your coding agents. For less.
| Feature | Cloud AI | CORTEX |
|---|---|---|
| Monthly Cost | $20-200 | $20 (Free in Early Access) |
| Query Limits | 50 - 500 /day | Unlimited |
| Response Time | 2-5s + network | <500ms local |
| Offline Support | ✗ | ✓ |
| Private | ✗ | ✓ |
| Compliance Friendly | ✗ | ✓ |
| Water Per Query 1 | ~25ML | 0ML |
| Carbon Footprint 2 | Massive | Laptop |
1 20-50 queries to cloud AI services can consume between 13mL and 500mL of water.
2 AI data centers are generally considered significant carbon emitters.
Be One of the First
We're working hard to get CORTEX ready for launch. Sign up to join our private testing group.
With Help From Our Friends
System Requirements
CORTEX is distributed as a native binary for macOS (Apple Silicon). One download, one command to install.
┌─────────────────────┬─────────┬─────────┬────────────┐ │ SPEC │ RAM │ STORAGE │ CPU │ ├─────────────────────┼─────────┼─────────┼────────────┤ │ Required │ 16GB │ ~30GB │ M1 or M2 │ │ Recommended │ 24GB │ ~30GB │ M3 or M4 │ └─────────────────────┴─────────┴─────────┴────────────┘
Frequently asked questions
How does it compare to Claude Code or Cursor?
CORTEX has 100% feature parity with Claude Code—web search, fetch, code generation, refactoring, everything. It's built in Swift instead of being another Electron app, so it's faster and uses less memory. Plus, no usage limits.
What model does it use?
CORTEX uses a modified version of Qwen3's open source model. It starts as a specialized 80B parameter model refined, distilled, and fine-tuned specifically for code generation and tool use. It outperforms much larger models on coding tasks while fitting comfortably in RAM.
Can I use my own models?
Currently CORTEX is optimized for our customized model. We're working on support for other local models but we're focused on a consistent user experience and we believe we should always ship with the best model for the job.
Does it support all languages Claude Code or competitors do?
Yes, and then some. Python, JavaScript, TypeScript, Go, Rust, Swift, Java, C++, and dozens more. If Claude Code can do it, CORTEX can do it.
What about privacy and security?
Everything runs locally - the model, the context, the prompts, the responses. Your code and prompts never leave your machine.
Do you track any usage data or telemetry?
Yes, we have very limited telemetry data turned on by default including:
- License validation - to ensure you're a valid licensed user
- Tool usage - logs anonymous successes and failures of tool calls to help with tool use reinforcement training
- Crash Reports - if CORTEX crashes, we'll collect a crash report to help us fix the issue
That's it. We're committed to keeping your data private and secure by keeping it yours.
While in early testing, we're requiring telemetry to be enabled so we can improve the product and prepare for launch. After launch, we'll make it optional and you can turn it off at any time.