01
Open the model card
Select the virt, go to the AI section, and choose Model. Confirm you are in the intended virt and can see both vendor and model selectors before changing parameters.
Model setup
Picking the right model controls cost, latency, and tone. Work through these steps every time you launch a new virt or test a fresh provider.
Preconditions
- Workspace role with access to the AI tab
- Pricing guidance from finance or procurement
- API keys for the providers you plan to activate
Steps
02
Choose a vendor and model
Use OpenAI or GigaChat for most customer care flows, while DeepSeek or Anthropic help with long reasoning. After you pick the vendor, confirm the API key status to ensure quota is active.
03
Tune temperature and token limits
Move the Temperature slider toward 0.3 for strict FAQ answers or toward 0.9 for more exploratory chats. Set Max tokens to 700 for short updates or 1500 for detailed consults, then note the expected spend in the sidebar.
04
Review parameters and save
Re-check the Vendor + Model pair, then validate Temperature, Top P, and max-token limits. If needed, adjust values and click Save.
WARNING
High token limits often hide gaps in knowledge. Before doubling the context window, check whether the virt can answer by using tools or by refining the instruction.
Practical Example
Example
Practical Example
A delivery network uses GigaChat for Russian-language order chats. The model is set to temperature 0.4 and a 900-token limit so responses stay short and accurate. After rollout, the team saw dialog cost drop by 18 percent, and first-response time improved by 2 seconds.