Run open-source LLMs on a real machine
One-click recipes for the top open-weight models on Ollama, plus a web UI to talk to them. GPU-ready Linux boxes, no quotas, no API keys.
FeaturedDeepSeek
DeepSeek (Ollama)
DeepSeek is a Chinese AI firm disrupting the industry with low-cost, open-source large language models. Strong on reasoning and code generation.
Open recipe
Best for codingQwen 2.5 Coder 7B
Alibaba Qwen (Ollama)
Tool-capable code model tuned for programming tasks. Best paired with OpenCode for agent-driven coding.
Open recipe
Qwen 2.5 7B
Alibaba Qwen (Ollama)
Tool-capable, JSON-tuned chat model. A solid default for agent UIs (OpenCode, Nanobot, SuperAGI) that require tool calling.
Open recipe
Qwen
Alibaba Qwen (Ollama)
Qwen Chat is an AI assistant for everyone, powered by the Qwen series of open models from Alibaba.
Open recipe
Llama 3.2
Meta (Ollama)
Llama 3.2 is an auto-regressive language model that uses an optimized transformer architecture.
Open recipe
Llama 3.2 3B
Meta (Ollama)
Smaller, fast, tool-capable Meta model. Good fit for boxes with limited RAM that still need tool calling.
Open recipe
Long Context Llama 3.1:8b
Meta (Ollama)
The 8-billion-parameter Llama 3.1 checkpoint, served locally through Ollama. About 16 GB of RAM, suited to longer-context reasoning workloads on GPU-capable hosts.
Open recipe
Granite 3.2 8B
IBM (Ollama)
IBM Granite 3.2 8B. Tool-capable instruction-tuned model with strong instruction following. Pair with any agent UI that requires tool calling.
Open recipe
Gemma
Google (Ollama)
Google Gemma open-weight models served via Ollama. Lightweight, permissively licensed, fast.
Open recipe
TinyLlama
Community (Ollama)
A compact, open-source language model designed to deliver efficient performance on local devices with a surprisingly small footprint.
Open recipe
Bring your own UIOpen WebUI
Chat UI for any LLM
A polished chat UI that talks to your Ollama box. Pair it with any of the recipes here to chat with your open models in the browser.
Open recipeWhy on EasyEnv
Hosted on a real machine, ready in seconds
No API keys
No accounts, no rate limits, no usage caps. Your model, your machine, your data.
GPU when you need it
Pick a GPU-backed box for the bigger models, a CPU box for the small ones.
Pair with an agent
Each recipe wires neatly into OpenCode, OpenClaw, Open WebUI, and other agent UIs.
Tear it down
Run an experiment, then destroy the machine. No long-running infrastructure to manage.
Bring your own model
Don't see what you need? Pull any Ollama-compatible model on a custom machine and serve it the same way.