> ## Documentation Index
> Fetch the complete documentation index at: https://www.bolna.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Connect Cerebras inference to Bolna voice agents

> Use Cerebras Shared or Dedicated Inference for thousands of tokens per second in a Bolna voice agent. Model IDs, context limits by tier and setup steps.

[Cerebras](https://www.cerebras.ai) runs models on its wafer-scale hardware and is built for raw generation speed. Its model catalog lists `gpt-oss-120b` at around 3,000 tokens per second. On a voice call, 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. Shared Inference offers a small set of models. More model families are available through Dedicated Inference.

***

## Models on Shared Inference

| 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 are from the Cerebras [model catalog](https://inference-docs.cerebras.ai/models/overview) at the time of writing; check it for the current list.

> [!NOTE]
>
> The context window is smaller on the free trial than on paid tiers. A long system prompt plus a long call can hit the free-tier limit before it hits the paid one.

***

## Set up Cerebras

**1. Create an API key in the Cloud Console**

Sign in to the [Cerebras Cloud Console](https://cloud.cerebras.ai) and open **API Keys** in the left navigation. Create a key.

**2. Try the model in the playground**

The Cloud Console playground works without a key. Paste your agent's system prompt and a few caller lines to check tone and length before you wire it up.

**3. Choose Cerebras in Bolna**

Open **Settings → Integrations → LLM → Custom LLM → Add model**, or choose **+ Add custom LLM** in an agent's **LLM Provider** dropdown. Select **Cerebras**.

![Bolna Add custom model dialog with Cerebras selected and api.cerebras.ai shown as the endpoint placeholder](https://mintcdn.com/bolna-54a2d4fe/0EXypen329Eb5XE2/images/custom-llm/providers/cerebras-selected.png?fit=max\&auto=format\&n=0EXypen329Eb5XE2\&q=85\&s=395fde6ac532d3391f6adcc406512764)

Cerebras selected in the Add custom model dialog

**4. Register the model**

**Dashboard**

* **Model name**: an exact Shared Inference model ID such as `gpt-oss-120b`. For Dedicated Inference, use your **endpoint ID** instead.
* **Endpoint URL**: `https://api.cerebras.ai/v1`
* **API key**: the key from the Cloud Console
* **Family**: for example `gpt-oss` or `qwen`

![Add custom model dialog with Cerebras selected, model gpt-oss-120b, endpoint api.cerebras.ai/v1, masked key and family gpt-oss](https://mintcdn.com/bolna-54a2d4fe/aV72vktown5Rn3f1/images/custom-llm/providers/cerebras-filled.png?fit=max\&auto=format\&n=aV72vktown5Rn3f1\&q=85\&s=0e2266c7d4d0550d64409ab8ecff7d1b)

Cerebras with gpt-oss-120b

**API**

```bash theme={"system"}
curl --request POST \
  --url https://api.bolna.ai/user/model/custom \
  --header 'Authorization: Bearer <BOLNA_API_KEY>' \
  --header 'Content-Type: application/json' \
  --data '{
    "custom_model_name": "gpt-oss-120b",
    "custom_model_url": "https://api.cerebras.ai/v1",
    "custom_model_key": "<CEREBRAS_API_KEY>",
    "custom_model_family": "gpt-oss"
  }'
```

**5. Use it on an agent**

Choose **custom** on the agent's **Intelligence** tab, pick the Cerebras model, then save.

***

## Shared vs Dedicated Inference

With Shared Inference, the `model` field is a catalog model ID and limits depend on your tier (Free Trial or Pay as You Go). With [Dedicated Inference](https://inference-docs.cerebras.ai/dedicated/overview), the `model` field is your **stable endpoint ID**, which routes to whichever deployment is active. Because Bolna sends the registered model name as `model`, register a dedicated endpoint with its endpoint ID as the model name. You can then roll out a new deployment behind the same ID without touching your agents.

***

## Troubleshooting

> **429 rate limit errors**
>
> Shared Inference enforces per-tier request and token limits. Check your limits in the Cloud Console, move to Pay as You Go, or ask Cerebras about Dedicated Inference for sustained call volume.

> **Context length exceeded late in long calls**
>
> The free tier has the smaller context window. Shorten the system prompt, or upgrade for the full context length.

> **Model not found**
>
> Model IDs are short and exact, for example `gpt-oss-120b`, with no organisation prefix. Copy them from the catalog.

> **404 on chat completions**
>
> Use exactly `https://api.cerebras.ai/v1` as the endpoint URL. Bolna appends `/chat/completions` itself.

***

## Related

* [Custom LLM inference](https://www.bolna.ai/docs/custom-llm-inference)
* [Groq](https://www.bolna.ai/docs/custom-llm-inference/groq.md): another provider built for low-latency inference
* [Intelligence Tab](https://www.bolna.ai/docs/agent-setup/llm-tab.md)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.