> For the complete documentation index, see [llms.txt](https://trust.memori.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://trust.memori.ai/en/plans-billing/supported-llm-providers.md).

# Supported LLM providers

External LLMs are only enabled when the customer explicitly chooses to: Memori's proprietary NLP engine (MemoriNLP) always tries first to answer with the knowledge provided by the Agent's creator, and queries an external generative model only when configured to do so.

**Natively integrated providers**

| Provider                           | Available models (indicative, continuously updated) |
| ---------------------------------- | --------------------------------------------------- |
| **OpenAI**                         | GPT family and o series                             |
| **Anthropic**                      | Claude family                                       |
| **Mistral AI**                     | Mistral Small, Large, Nemo                          |
| **Microsoft Azure OpenAI Service** | GPT models on Azure infrastructure                  |
| **Amazon Bedrock**                 | Claude models on AWS infrastructure                 |
| **Google Vertex AI**               | Claude models on Google Cloud infrastructure        |

**Custom providers**

AIsuru also supports the integration of **any model compatible with the OpenAI protocol**, whether cloud-hosted or running locally (for example via Ollama, LM Studio or equivalent solutions). This also allows open-source models and enables fully on-premise setups where the AI model never leaves the customer's perimeter.

**Data retention by providers**

Each provider applies its own data retention policy, generally used only to prevent service abuse:

| Provider                         | Declared retention              |
| -------------------------------- | ------------------------------- |
| **OpenAI**                       | 30 days (abuse prevention only) |
| **Anthropic**                    | 30 days (abuse prevention only) |
| **Mistral AI**                   | 30 days (abuse prevention only) |
| **Microsoft Azure**              | 30 days (abuse prevention only) |
| **Amazon Bedrock**               | No retention                    |
| **Google Vertex AI**             | 30 days (abuse prevention only) |
| **Local models / private cloud** | As defined with the customer    |

{% hint style="warning" %}
For full details on the contractual terms, provider terms of service and IP arrangements with each model provider, request the relevant Data Processing Agreement. See [DPA](/en/legal-documents/dpa.md).
{% endhint %}

**Why the customer chooses the provider**

Since the model configuration is tied to the individual Agent and is not imposed at platform level, a customer can:

* use their existing contract with a provider;
* have different Agents respond with different models, according to needs;
* opt for a model run entirely within their own perimeter, so that no data leaves it.

For the regulatory framework of the general-purpose AI models used by Memori, see GPAI: General-Purpose AI Models.