Last updated
What is Nebius used for in a Bolna voice agent?
Nebius Token Factory, formerly Nebius AI Studio, serves open-weight models such as DeepSeek, Qwen and Llama as the LLM for a Bolna voice agent through an OpenAI-compatible API. What sets it apart is where the model runs: dedicated endpoints let you pick the deployment region, which matters when call transcripts must stay in a given jurisdiction. Bolna connects to it as a custom LLM using the Nebius preset.
How do I set up Nebius with Bolna?
- Sign up at tokenfactory.nebius.com.
- Open API keys, click Create API key, name it and copy it right away. It is shown only once.
- Browse the text models and copy a full model ID, for example
nvidia/Nemotron-3_5-Lightning. Prefer an instruct model over a reasoning model, and add-fastto the ID if the model has a fast flavour. - Optionally, create a dedicated endpoint in the region you need and use the model identifier Token Factory shows for it.
- In Bolna, open Settings → Integrations → LLM → Custom LLM → Add model, or choose + Add custom LLM from an agent's LLM Provider dropdown. Select Nebius.
- Enter the model ID, set Endpoint URL to
https://api.tokenfactory.nebius.com/v1, paste the key and set Family tonemotron,qwenordeepseek. - Choose custom on the agent's Intelligence tab, select the Nebius model, save and run a test call.
The Nebius setup guide in the Bolna docs covers tool calling and rate limits.
What is the difference between Nebius public and dedicated endpoints?
| Public serverless endpoints | Dedicated endpoints | |
|---|---|---|
| Capacity | Shared, multi-tenant | Reserved for your organisation |
| Rate limits | Dynamic limits apply | No standard limits; throughput depends on your capacity |
| Region | Global, can change with available capacity | Fixed, chosen by you: eu-north1, eu-west1 or us-central1 |
| Pricing | Per token | Per GPU-hour, billed per minute |
Bolna checks the key and model ID against the Token Factory model list when you save.
What can you build with Nebius and Bolna?
Agents with data residency requirements. Register a dedicated endpoint per region as its own custom model so agents for each market point at the right one. Azure OpenAI is another option when data residency matters.
High-volume campaigns. Public endpoints return 429 over the limit, which leaves the agent silent for that turn. Move high-volume agents to a dedicated endpoint, which has no standard rate limits.
Low-latency open-model agents. Use a fast flavour for the same outputs with lower latency. For callers in India, measure the first reply on a web call, since public endpoints have no fixed location.
Tool-calling agents. Check a model's supported_features in the model list before using it on an agent that calls tools. The custom LLM page compares Nebius with the other presets.