Mycelis
Mycelis
Model hosting
With Mycelis you do not rent a GPU — you choose the model. We start it on a suitable GPU, size it to the number of people using it and bill only while it runs.
No base fee · No minimum term · Pause any time
Included with every model
What you get
No GPU catalog, no drivers, no containers. You describe what you need; Mycelis does the rest.
From the size of the model and the number of concurrent users, Mycelis works out the memory required and picks a suitable GPU.
Raise the number of concurrent users at any time. The deployment is set up again on a GPU that fits.
Inference runtime, networking and HTTPS are configured for you. No Docker, no drivers, no ops.
Set a daily time window and the model starts and stops on its own — for example for your working hours.
Optionally stop after a period without requests, and start again automatically with the first request.
A daily runtime limit per deployment and a monthly budget per workspace keep costs predictable.
How it works
01
Pick an open-source model from the catalog and state how many people will use it at the same time.
02
A suitable GPU is assigned, the runtime starts and networking and HTTPS are set up. Loading the weights takes as long as the model needs.
03
Call the OpenAI-compatible endpoint or open the chat interface. Stop and start the model whenever you like.
Business model
No seats, no base fee, no minimum term. A hosted model is billed like a meter: it runs while your team needs it.
By the minute, only while it runs
Billed at the hourly rate of the GPU that was chosen, to the minute. The rate depends on the GPU size.
Stopped means no runtime cost
While a deployment is stopped, no GPU runtime accrues. Start it again with one click, a schedule or the first request.
No base fee, no minimum term
Pure pay-as-you-go. You pay for the components you actually use.
Balance and budgets
A deployment starts only with sufficient balance. If you set a monthly budget for the workspace, running deployments are stopped when it is exceeded.
from openai import OpenAI
client = OpenAI(
base_url="https://api.mycelis.ai/proxy/v1",
api_key="pat_..."
)
response = client.chat.completions.create(
model="your-deployment-name", # the model you deployed
messages=[{"role": "user", "content": "Hello"}]
)
print(response.choices[0].message.content)Common questions
No. You choose the model and how many people use it at the same time. Mycelis picks a suitable GPU from that.
The deployment is stopped and no GPU runtime accrues while it is stopped. The deployment and its settings stay in place, so you can start it again later.
That depends on the model: the weights must be loaded onto the GPU first, and larger models take longer. You can follow the progress in the dashboard.
Open-source models from the Mycelis catalog, and your own fine-tuned adapters. Commercial models such as GPT or Claude need no hosting at all — they run through managed or your own keys.
Yes. Set a daily time window and Mycelis starts and stops the model for you. You can also stop it manually at any time, for example over the weekend.
Ready to get started?
Create a free account, pick a model and let Mycelis do the rest.