---
title: "Bringing Your Own LLM"
description: "LLM providers, gateways, recommended models, credentials, and observability for CORRA AI."
---

CORRA AI uses the AI configuration defined in **Platform Settings → AI**. The configuration determines which LLM provider and model CORRA AI uses, along with the credentials and runtime settings required to connect to it.

## LLM providers

The AI configuration contains the provider connection and the default model used by CORRA AI. Supported LLM providers, and the models recommended for each, are:

| Provider | Recommended models |
| --- | --- |
| OpenAI | GPT-5 series |
| Anthropic | Claude Sonnet 4.5+, Claude Opus 4.5+, Claude Haiku 4.5+ |
| Google | Gemini Flash 2.5+, Gemini Pro 2.5+ |

These models do not have to be reached directly. They can also be used through a hosting provider or an OpenAI-compatible gateway, as described in [LLM gateways and hosting providers](#llm-gateways-and-hosting-providers).

## LLM gateways and hosting providers

Rather than connecting to a provider directly, the same models can be reached through a cloud hosting service or an AI gateway.

The following are supported:

| Connection | Type |
| --- | --- |
| Azure OpenAI | Hosting provider |
| Azure Anthropic | Hosting provider |
| Amazon Bedrock | Hosting provider |
| Azure AI Foundry | Gateway |
| Google Cloud Vertex AI | Gateway |
| Databricks AI Gateway | Gateway |
| Portkey | Gateway |
| Your own internal gateway | Gateway |

Any OpenAI-compatible endpoint can be used, so a gateway that is not listed here works in the same way. Gateways are commonly used to centralise credentials, apply routing or rate limits, and keep model traffic inside an approved network path.

## Configuring a connection

Configuration varies by connection type. Direct integrations typically require an API key, while hosted services such as Azure OpenAI and Amazon Bedrock require additional connection or cloud credentials. A gateway is configured the same way as a direct provider, with the endpoint pointing at the gateway.

The model configuration includes the model or deployment name, temperature, and maximum output tokens. For Azure OpenAI, for example, the configuration includes the Azure endpoint, API version, API key, and deployment name.

Credentials are stored encrypted. Secrets such as API keys, tokens, passwords, and other credentials are also redacted from tool-call logs.

## Model settings

The model configuration controls how CORRA AI interacts with the selected model.

**Temperature** controls the amount of variation in model responses. **Max Tokens** limits the amount of output generated by the model.

Context-window and output-token limits can also be defined for a model. When these limits are not explicitly configured, the platform uses the limits available for the model and falls back to a safe default where necessary.