Real local neural LLMs
Genuine inference runs on your GPU or CPU through a bundled runtime.
Stop renting your AI. Own it.
Vibe runs real neural coding models on your own PC. Build, verify, iterate, and keep your projects private with no cloud dependency, subscription, or token meter.
No cloud
Inference runs on your hardware.
No subscription
Buy once. No per-token bill.
No upload
Your code stays on your machine.
Works offline
Open it with Wi-Fi off and build.
An AI coding studio that lives on your PC.
Vibe ships with the local models and runtime it needs. Describe a product, game, utility, or prototype, and Vibe plans the work, writes files, runs checks, repairs failures, and keeps a visible trail of what changed.
Project root
Z:\Games\cozy-platformer
Current mission
Build → verify → iterate
Generate the player controller, validate JavaScript, patch missing exports, then re-run checks.
- Architect plan complete
- Gameplay files written
- Validation running
- Local model repairing
Built for local control, not rented compute.
Ships with the models
Coding, reasoning, and embedding models install with the app.
100% local and private
Your prompts, source, and projects never leave your machine.
Build verification loop
Vibe writes files, runs real checks, and self-corrects when validation fails.
Specialist task graph
Architect, engineer, artist, tester, and reviewer roles coordinate through a resumable queue.
Extensible local tools
Use project/user subagents, skills, hooks, MCP tools, and optional local image generation.
Describe it. Watch it build. Keep control.
Prompt the product
Ask for a cozy game, a shop inventory tool, a utility app, or a prototype in plain language.
Vibe plans the work
The task graph splits the build into roles, status, evidence, next actions, and files changed.
Models write real files
Local coding models generate code in a normal project workspace on disk.
Checks keep it honest
Validation, screen OCR, loopback browser checks, and self-repair keep work grounded in what runs.
A local studio crew in the box.
Vibe includes curated coding, reasoning, embedding, and image models so first launch does not start with an API key hunt or a model download maze.
Model
Role
Delivery
Neural models
qwen2.5-coder:7bCodingBase game
deepseek-coder-v2:16bCodingBase game
gpt-oss:20bGeneral chat / reasoningBase game
starcoder2:7bCodingBase game
nomic-embed-textEmbeddings / retrievalBase game
Image models
SDXL (Stable Diffusion XL)Image generation - runs on any GPU and CPUBase game
FLUX.2-klein-4B (+ Qwen3-4B encoder)Higher-quality image generation + legible in-image text, auto-selected on ~12 GB+ VRAM GPUsOptional DLC (auto-installs)
System requirements
Made for real Windows PCs.
- OSWindows 10/11, 64-bit
- DiskAbout 33 GB for the app and bundled models
- Memory16 GB minimum, 32 GB recommended
- GPU8 GB+ VRAM recommended; CPU works but runs slower
- NetworkNot required for local operation
FAQ
Straight answers, no cloud fog.
Is there a subscription?
No. Vibe is designed as a buy-once app with no subscription and no per-token bill.
Does my code get uploaded?
No. Vibe runs inference locally and keeps your prompts, projects, and source on your PC.
Does it work offline?
Yes. The core coding workflow works fully offline because the models and runtime ship with the app.
Is it perfect every time?
No AI is perfect. Vibe shows its work, runs real checks, and iterates so you stay in control.
