Last updated
What is Cerebras used for in a Bolna voice agent?
Cerebras is a very fast LLM option for a Bolna voice agent, running models on wafer-scale hardware with gpt-oss-120b listed at around 3,000 tokens per second. Fast generation means the full reply is ready almost as soon as the first token arrives, which helps longer answers and agents that call tools mid-turn. The trade-off is a short catalog on Shared Inference.
How do I set up Cerebras with Bolna?
- Sign in to the Cerebras Cloud Console, open API Keys and create a key.
- Optionally, paste your agent's system prompt into the Cloud Console playground to check tone and length. It works without a key.
- In Bolna, open Settings → Integrations → LLM → Custom LLM → Add model, or choose + Add custom LLM in an agent's LLM Provider dropdown. Select Cerebras.
- Set Model name to a Shared Inference model ID such as
gpt-oss-120b(or your endpoint ID for Dedicated Inference), Endpoint URL tohttps://api.cerebras.ai/v1, paste the key, and set Family togpt-ossorqwen. - Choose custom on the agent's Intelligence tab, pick the Cerebras model and save.
The Cerebras setup guide in the Bolna docs explains Shared and Dedicated Inference in more detail.
Which Cerebras models work with Bolna?
| Model ID | Parameters | Context (free / paid) | Speed |
|---|---|---|---|
gpt-oss-120b | 120B | 65k / 131k | ~3,000 tokens/s |
qwen-3.8-27b | 27B | 64k / 128k | ~1,850 tokens/s |
Values come from the Cerebras model catalog as quoted in the Bolna docs. More model families are available through Dedicated Inference.
| Shared Inference | Dedicated Inference | |
|---|---|---|
| Model name in Bolna | Catalog model ID | Your stable endpoint ID |
| Limits | Per tier (Free Trial or Pay as You Go) | Your own deployment |
What can you build with Cerebras and Bolna?
Agents with long answers. Support or explainer calls where the agent needs a few sentences per turn, since the whole reply is generated almost at once.
Tool-heavy agents. Agents that fetch data or book slots mid-turn, where generation speed keeps the pause short. See the Cal.com integration.
High-volume lines on dedicated capacity. Move to Dedicated Inference when Shared Inference rate limits get in the way of sustained call volume.
Latency benchmarks. Compare Cerebras with Groq on the same agent. The custom LLM page lists every provider preset.