Back

AI Models

Loading the catalog…

All (0)

Local (0)

Cloud (0)

Any capability

All

LOCAL KEEP AI

Home (/) · Documentation (/docs) · Free Models (/free-models) · Pricing (/pricing) · Downloads (/downloads) · FAQ (/faq) · About (/about) · Support (/support) · Privacy (/privacy) · Terms (/terms)

Local Keep AI Model Catalog — Local GGUF, WebGPU & Frontier Cloud Models

Browse, benchmark, and launch hundreds of open-weight and commercial AI models across coding, reasoning, vision, mathematics, and multilingual generation.

On-Device GGUF & Browser WebGPU Models (Free Forever)

Download quantized GGUF weights on iOS, Android, macOS, Windows, and Linux or run WebGPU models directly inside Chromium/Safari browsers at zero token cost. Popular local models include Qwen 2.5 Coder (1.5B–32B), Llama 3.3 70B, DeepSeek R1 Distill, Gemma 4, Phi-4, and Mistral Small.

Frontier Cloud Models Billed Transparently by Output Tokens

Access Anthropic Claude Sonnet & Opus, OpenAI GPT-4o & o3-mini, Google Gemini Pro, DeepSeek V3/R1, and Qwen 3 Coder through one unified API and subscription without managing separate provider API keys.

Hardware Compatibility & VRAM Sizing Guide

Each catalog entry reports parameter count, quantization format (Q4_K_M, Q5_K_M, Q8_0, FP16), context window length (up to 128K–1M tokens), and recommended RAM/VRAM so you can pick the fastest model for your device.

Frequently Asked Questions

Do local models cost any output tokens?

No. Local GGUF models and in-browser WebGPU models run on your own CPU/GPU/NPU and are 100% free with unlimited usage.

Can I pull Hugging Face models directly from the CLI?

Yes. Run lk pull <model-id> in your terminal to download and quantize compatible GGUF and safetensors models.