Server to live AI in 5 clear steps.
Seed AI is self-hosted on infrastructure inside your own network, so your business stays in control throughout the deployment process.
We help you plan the hardware, install the platform, configure your models and users, and prepare your team for launch.
One platform
built around your business.
Each deployment is tailored to the models, number of users, and internal workflows your business plans to support.
Get your GPU server.
Seed AI runs on a GPU-enabled Windows Server on your own network, and your IT team provisions that server.
The amount of GPU power you need depends primarily on the models you plan to run. Larger models generally require more GPU memory, while smaller models can run on more modest hardware.
Tell us which models you want to run and we will provide server and GPU specifications so your team can purchase the right hardware the first time.
View the GPU sizing guide Model size, GGUF file size, and estimated VRAM
A simple rule of thumb is that a model needs roughly its GGUF file size in GPU memory, plus approximately 20% overhead. Additional memory may be needed for longer conversation context and other runtime settings.
Bigger models require more VRAM. Smaller or more heavily quantized models require less.
| Model Size | Approx. GGUF File | Estimated GPU Memory | Typical Setup |
|---|---|---|---|
| 7–8B | About 5 GB | About 7 GB | One 8–12 GB GPU |
| 13–14B | About 8 GB | About 10 GB | One 12–16 GB GPU |
| 30–34B | About 20 GB | About 24 GB | One 24 GB GPU |
| 70B | About 42 GB | About 50 GB | One 48 GB GPU or two 24 GB GPUs |
| 120B+ | 60 GB or more | 70 GB or more | Two 48 GB GPUs |
These figures are a rough planning guide for common Q4 quantizations. Exact memory requirements vary by model architecture, context size, runtime settings, GPU offloading, and concurrent usage.
We help you install Seed AI .
Once your server is ready, we provide installation support and work alongside your IT team to get SeedAi installed and running on your network.
The deployment uses Windows Server and IIS, along with the required GPU drivers and supporting application components.
Upload your models.
Use the Seed AI model manager to upload supported GGUF model files and run them on your own GPU hardware.
Add multiple models for different tasks, departments, or performance needs. Your administrators can make models available to the appropriate users and agents.
Upload supported GGUF files from the model sources your company approves.
Maintain multiple models and switch between them based on the work being performed.
Replace or add models as your needs change and better options become available.
Configure your platform.
Customize Seed AI for your company, then configure the users, roles, models, agents, prompt templates, and knowledge bases your team will use.
Train your team and go live.
We give your staff practical training so they understand how to use the chat workspace, select approved models and agents, work with private knowledge bases, and follow your company’s AI policies.
After launch, the platform is in your hands. Your team can use your locally hosted models without Seed AI charging per-prompt or per-token usage fees.
Start with the right server.
Finish with AI you control.
Talk with us about the models you want to run, your existing infrastructure, and how Seed AI can be deployed inside your business.