> ## 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 Simplismart to Bolna voice agents
> Use a Simplismart shared endpoint or a dedicated, optimised deployment as the LLM behind a Bolna voice agent. Covers API keys, endpoints and model names.
[Simplismart](https://www.simplismart.ai) is an inference platform focused on optimising open models for speed. It gives you two ways to call a model, and the choice decides what you type into Bolna:
| | Shared endpoint | Dedicated endpoint |
| - | - | - |
| Infrastructure | Simplismart's shared pool | Isolated capacity reserved for you |
| Models | Pre-deployed models you can try in the Playground | Pre-optimised models from the Model Marketplace, or your own and fine-tuned models |
| Pricing | Per API call and usage | Deployment-based |
| Endpoint URL | `https://api.simplismart.live`, with no `/v1` | Shown on the deployment's **API** tab |
Start on the shared endpoint to try a model on a call, then move to a dedicated endpoint for stable latency under production load.
***
## Set up Simplismart
Sign up at [app.simplismart.ai](https://app.simplismart.ai) and open the **Playground**. Pick an LLM, for example Gemma 4 31B Instruct, and send a few prompts in the style of your agent to judge reply speed and quality.
Go to **Settings → API Keys → Generate Key** in the Simplismart console. Name the key, choose an expiry and copy it. The same key works for shared and dedicated endpoints in your account.
Simplismart keys stop working at midnight UTC on their expiry date, and every call after that fails. Pick a long expiry, or ask [Simplismart support](mailto:support@simplismart.ai) for a key that doesn't expire.
Use `https://api.simplismart.live` as the base URL, with no `/v1`, and the model's ID as the model name. In the **Playground**, pick a model and click **Get API details** to copy the ID exactly, for example `google/gemma-4-31B-it`.
In the [Model Marketplace](https://app.simplismart.ai/model-marketplace), deploy a pre-optimised model in one click, or add your own model. When the deployment is running, open its **API** tab and copy the endpoint URL and the model name from the code snippet. See [inference and monitoring](https://docs.simplismart.ai/model-suite/deployments/inference-and-monitoring).
***
## Register it in Bolna
Open **Settings → Integrations → LLM → Custom LLM → Add model**, or from an agent's **Intelligence** tab open the **LLM Provider** dropdown and choose **+ Add custom LLM**. Select **Simplismart**.
The endpoint field is left empty. The shared endpoint is only shown as a placeholder so a dedicated URL isn't overwritten.
* **Model name**: the Simplismart model ID, for example `google/gemma-4-31B-it`
* **Endpoint URL**: `https://api.simplismart.live`, or your dedicated endpoint
* **API key**: the key from **Settings → API Keys**
* **Family**: for example `gemma`
```bash theme={"system"}
curl --request POST \
--url https://api.bolna.ai/user/model/custom \
--header 'Authorization: Bearer ' \
--header 'Content-Type: application/json' \
--data '{
"custom_model_name": "google/gemma-4-31B-it",
"custom_model_url": "https://api.simplismart.live",
"custom_model_key": "",
"custom_model_family": "gemma"
}'
```
On the **Intelligence** tab choose **custom** and your Simplismart model, then save and test the agent.
***
## Things to know
* **Bolna can't check the key or the model ID when you save.** Simplismart doesn't serve a model list, so a wrong key or a misspelled model ID is saved anyway and first shows up as a silent agent on a call. Run the [streaming test](/docs/custom-llm-inference#test-your-endpoint-before-you-register-it) first, then place a web call.
* **Extra headers aren't supported.** Bolna sends only the `Authorization` header. If a dedicated deployment's API snippet also needs an `id` header, Bolna can't call it. Ask Simplismart for an endpoint that doesn't need one.
* **Tool calling.** Simplismart doesn't document tool support on shared endpoints, so test your agent's tools on a web call before going live. On a dedicated deployment, set the tool call parser when you compile the model. See the [optimization guide](https://docs.simplismart.ai/guides/optimization-guide).
* **Region.** For a dedicated deployment in a specific region, such as India, contact Simplismart. Running close to your callers cuts network time on every turn.
***
## When to move from shared to dedicated
* Calls show uneven time to first token during busy hours: shared capacity is used by other customers too.
* You need a fine-tuned or private model: only dedicated endpoints can run your own weights.
* You need a latency SLA for a high-volume campaign.
When you switch, register the dedicated endpoint as a **new** custom model, move your agents to it, then delete the old one. Agents resolve a custom model by its name, so a clear name for each model makes the switch easy to follow.
***
## Troubleshooting
The key has passed its expiry date. Generate a new key under **Settings → API Keys**, add the model again with the new key, move your agents to it and delete the old one.
Because Simplismart has no model list, Bolna can't catch a wrong key or model ID when you save. Check both with the [streaming test](/docs/custom-llm-inference#test-your-endpoint-before-you-register-it). Keys from a different Simplismart workspace won't work.
Copy the model ID exactly as Simplismart shows it, including case, for example `google/gemma-4-31B-it`. Only models Simplismart has pre-deployed are served on the shared endpoint.
Check the deployment's status in the Simplismart console. A stopped or scaled-down deployment can't answer calls, even though the model stays registered in Bolna.
***
## Related
* [Custom LLM inference](/docs/custom-llm-inference): fields, save-time checks and extra request params
* [Nebius](/docs/custom-llm-inference/nebius): another provider with both serverless and dedicated endpoints
* [Intelligence Tab](/docs/agent-setup/llm-tab)
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.