Setting Up Your LLM

Last updated: August 23, 2026

Bring your own LLM is the advanced privacy option for Sisense Intelligence. Instead of using the model that Sisense manages for you, you connect your own LLM provider account, so the AI features run on a model you control, in your own cloud and region, under your provider's data terms.

The managed LLM is the simplest way to run these features: Sisense hosts the model for you, so there is no provider account to set up and nothing to maintain, which makes it the best default for most deployments. You reach for bring your own LLM when privacy, compliance, or data-residency requirements mean AI requests should stay inside an environment you own. You supply the provider, the model, and the credentials, and Sisense uses your model to generate queries, charts, and narrative insights. A custom LLM supplements the managed services rather than replacing them.

Sisense supports three providers for this: Azure OpenAI, OpenAI, and AWS Bedrock.

Advanced Privacy and Control

Bringing your own LLM gives you control that the managed option does not:

• Your data stays in your environment. AI requests go to your own provider account, in the region you choose, so prompts and results are processed under your cloud and your provider's data terms, not a model managed by Sisense.

• You meet privacy and compliance requirements. When your organization requires that AI processing happen inside infrastructure you own, this is how you satisfy that.

• You control cost and quota. Usage is billed through your own provider account, so you set the budgets and limits directly.

• You choose the model. Pick a provider and a model in the tier that fits: a strong model for the best quality, or a lightweight one for lower cost and latency.

The trade-off is setup and billing on your side. If you have no specific requirement to run AI in your own environment, the managed LLM is simpler.

How It Works

 Sisense Intelligence  --request-->  Your LLM provider    --response-->  Sisense
 (Assistant, narrative,              (Azure OpenAI, OpenAI,              (applies your
  semantic enrichment)                or AWS Bedrock, in                  governance, shows
                                      your account & region)              the result)

1. You configure your provider once, under Provider Configuration.

2. When an AI feature needs the model, Sisense sends the request to your provider, using the credentials you supplied.

3. Your provider processes the request in your own account and region, and returns the result.

4. Sisense applies your data permissions and governance, then shows the result in the feature.

Because the model runs in your own provider account, your prompts and results are handled under your cloud and your provider's data terms. That is the privacy benefit of bringing your own.

Supported Providers and Models

Sisense recommends models in two tiers. A strong model gives the best answer quality for complex assistant tasks. A lightweight model costs less and responds faster, which suits simpler computations.

The tier you need depends on the feature. The Assistant requires a strong model. Narrative and Semantic Enrichment can run on a lightweight model. If you use the Assistant, configure a strong model. If you only use narrative insights or semantic enrichment, a lightweight model is enough.

Recommended Models

Tier	Azure OpenAI / OpenAI	AWS Bedrock
Strong (best quality)	GPT-5.1	Claude Sonnet 4.6
Lightweight (lower cost and latency)	GPT-4.1 Mini	Claude Haiku 4.5

Sisense is optimized for the GPT family, hosted on Azure OpenAI or OpenAI. Claude models are available through AWS Bedrock.

All Supported Models

These are the models you can choose in Provider Configuration today. It is your responsibility to manage your model version.

Provider	Model	Version	Tier	Notes
Azure OpenAI / OpenAI	GPT-4.1	gpt-4.1-0414	Strong	Being deprecated by OpenAI in November 2026. Migrate to GPT-5.1.
Azure OpenAI / OpenAI	GPT-4o	gpt-4o-1120	Strong	Being deprecated by OpenAI in November 2026. Migrate to GPT-5.1.
Azure OpenAI / OpenAI	GPT-5.1	—	Strong	Recommended for the strong tier
Azure OpenAI / OpenAI	GPT-4.1 Mini	gpt-4.1-mini-0414	Lightweight	Being deprecated by OpenAI in November 2026.
Azure OpenAI / OpenAI	GPT-4o Mini	gpt-4o-mini-0718	Lightweight	Being deprecated by OpenAI in November 2026.
AWS Bedrock	Claude Sonnet 4.6	—	Strong	Recommended for the strong tier on Bedrock
AWS Bedrock	Claude Haiku 4.5	—	Lightweight	Recommended for the lightweight tier on Bedrock

Sisense supports Bedrock with these two models only; other Bedrock models are not available through Provider Configuration.

Some models offer extended reasoning and bill those reasoning steps as output tokens, which can raise cost significantly. Review your provider pricing before selecting one.

Before You Begin

You need all of the following before the setup steps will work:

Set Up a Provider

Under Provider Configuration, the Bring your own LLM section lists your custom providers. The flow is the same for every provider: add a provider, choose its type, and enter the connection details. The fields differ by provider type, so follow the section that matches yours.

1. In Sisense, in the Admin tab, select Sisense Intelligence > Provider Configuration.

2. If this is your first time here, read and accept the terms by clicking I agree to approve the terms and conditions in order to proceed with using the cloud-linked features.

   

3. Under Bring your own LLM, click + Add Provider.

   

4. In Provider Type, choose Azure OpenAI, OpenAI, or Amazon Bedrock, and enter the connection details for that provider (see below).

5. Click Save, then verify the connection.

Verify the Connection

After you save a provider, Sisense checks the connection. In the Bring your own LLM list, a green dot next to the provider means the model connected properly with the key you entered. If the dot does not appear, open the provider again and check the model, the base URL, and the API key.

You can also confirm it end to end:

1. Open an AI feature that uses the LLM, such as the assistant.

2. Ask a simple natural-language question against a model you have access to.

3. A successful query or chart response means the provider is working. If the request fails, check the credentials, the model name, and that the model is reachable from your account.

Cost and Quota

You are billed by your provider for every request Sisense sends to your model, under your own account and pricing. Sisense does not add a charge for bring-your-own usage. Set usage limits or budgets in your provider console if you want to cap spend, and be aware that reasoning-capable models can use far more output tokens than standard models.

Limitations

• You can add only one custom LLM at a time. Support for configuring multiple models together (a strong model and a lightweight one) is on the roadmap.

• Bring your own LLM supports Azure OpenAI, OpenAI, and AWS Bedrock only. Other providers are not available through this configuration.

• The assistant requires a strong model. Lightweight models suit narrative insights and semantic enrichment.

• On AWS Bedrock, Sisense supports only Claude Sonnet 4.6 (strong) and Claude Haiku 4.5 (lightweight). Other Bedrock models, including Claude Opus 4.6, are not available through Provider Configuration.

• OpenAI is deprecating GPT-4.1, GPT-4o, GPT-4.1 Mini, and GPT-4o Mini, and will stop supporting them in November 2026. Move to GPT-5.1 for the strong tier on Azure OpenAI or OpenAI before then.

• Availability of a given model depends on your provider account and region, not on Sisense.

Frequently Asked Questions