Lightning Fast Vibe IDE
for macOS
Unfork yourself.
A native AI development environment built from scratch.
No Electron. No VSCode. No compromises.
Just 100MB of pure, optimized intention.
Compare that to the 1GB+ download and 3-5GB idle RAM of... well, you know.
We Didn't Fork Anything. We Built Everything.
Every other AI coding tool starts the same way: take an existing IDE, bolt on some AI features, ship it, and pray the duct tape holds.
We asked a different question: What if we designed an IDE from the bare metal specifically for AI-assisted development?
The answer is Vibe Mode—a native macOS application where every byte serves a purpose. The inference engine compiles directly into the daemon. The terminal emulator shares memory with the app. The context engine indexes your project in milliseconds, not minutes.
This isn't an AI feature grafted onto a code editor.
This is what happens when AI is the architecture.
“The best tool isn't the one with the most features. It's the one that disappears when you're in flow.”
— Some gremlin who's been optimizing for six months straight
An IDE That Knows More Than You Do
(And we mean that as a compliment)
Project Omniscience
Your AI assistant doesn't just see your current file—it understands your entire codebase. Dependencies, patterns, that weird utility function you wrote six months ago. All of it. Indexed. Searchable. Contextual.
See around corners. Know what breaks before you break it.
Instant Inference
Run models locally on Apple Silicon with our native inference engine. 3-20ms to first token. No Python runtime. No pip install. No "waiting for model to load" spinner of death.
Your M-series chip finally earns its keep.
Ghostty Inside
Yes, that Ghostty. The mass-hyped, mass-praised GPU-accelerated terminal emulator—built directly into Vibe Mode. Not spawned as a subprocess. Not wrapped in an iframe. Native integration.
And the whole app is still just 100MB. We checked.
Infinite Context
Our Samsara context engine maintains conversation continuity without the "context window exceeded" interruptions. Your AI remembers the architecture discussion from this morning while you're debugging tonight.
Context that flows. Conversations that don't reset.
Security by Default
API keys in Keychain where they belong. Local inference that never phones home. Project-level isolation. Your code stays your code—we built paranoia into the architecture.
Privacy isn't a feature. It's a constraint we designed around.
Every Provider
Anthropic, OpenAI, Google, xAI, Groq, DeepSeek, OpenRouter, Together, Cerebras, and 10+ more. Plus local models via VibeMLX, Ollama, or LM Studio. Use what you want. Switch when you want.
No vendor lock-in. No "works best with our cloud."
Let's Talk Numbers
We're not here to name names. But we will name sizes.
| Vibe Mode | The Other Guys™ | |
|---|---|---|
| Download Size | 100 MB | 1+ GB |
| Memory at Idle | < 100 MB | 3-5 GB |
| Cold Start Time | < 1 sec | 3-8 seconds |
| Framework | Native | Electron |
| Terminal | Ghostty | xterm.js |
| Local Inference | Built-in | Subprocess |
| Context Indexing | Instant | "Indexing..." |
We're not saying Electron is bad. We're saying there's another way.
A way that doesn't require 8GB of RAM to edit a config file.
Built Different. Built Native. Built to Last.
No forks. No wrappers. Just raw metal and vibes.
VibeMLX
Native Inference Runtime
A 17MB Zig binary that does what Python frameworks need 500MB to accomplish. Loads HuggingFace models directly. Runs on Metal. Compiles into the daemon—no subprocess overhead, no IPC latency.
3-20ms time to first token. 900+ tok/s on small models.
See BenchmarksSamsara
Infinite Context Engine
Named for the cycle of rebirth. Your conversations don't die—they transform. Samsara maintains semantic continuity across sessions, intelligently compressing and retrieving context so your AI always knows what you've discussed.
No more "as we discussed earlier" when it doesn't remember.
Learn Morevibe-jinja
Bare-Metal Templating
We needed Jinja2 for chat templates. Python Jinja2 meant Python dependencies. So we rewrote it in Zig. Complete compatibility, native speed, zero external dependencies.
Because "good enough" isn't in our vocabulary.
View on GitHubJoin the Gremlins
Vibe Mode is currently in private beta. We're letting people in based on waitlist position—which you can improve by following us on social and inviting friends who appreciate good engineering.
The vibes are invite-only. For now.
Join the WaitlistAlready have access? Sign In