---
title: "Configure Generative AI (Default) Connection"
slug: "configure-generative-ai-default-connection-new"
updated: 2026-07-23T10:47:07Z
published: 2026-07-23T10:47:07Z
canonical: "knowledge.curiositysoftware.ie/configure-generative-ai-default-connection-new"
---

> ## Documentation Index
> Fetch the complete documentation index at: https://knowledge.curiositysoftware.ie/llms.txt
> Use this file to discover all available pages before exploring further.

# Configure Generative AI (Default) Connection

## Configure Generative AI (Default) Connection

This article focuses on setting up your Quality Modeller instance to utilize the Generative AI feature within the Curiosity Platform. We will delve into the specifics of Generative AI and the Language Learning Models (LLMs) that we collaborate with for our cloud and on-premise deployments.

### Default Model

#### Cloud Deployment

In the cloud version of Quality Modeller, Generative AI is pre-installed and automatically activated in your workspace. Should you wish to disable this feature, please contact your Curiosity account representative.

The cloud-based Quality Modeller utilizes OpenAI's `gpt-5.2` (subject to change). We advise you to review the OpenAI privacy policy prior to leveraging any of the Generative AI capabilities with Quality Modeller.

#### On-Premise Deployment

For on-premise deployments the Generative AI capability is disabled by default. To use this functionality it will need to be enabled in the docker-compose / environment variables for the installation.

Quality Modeller out-the-box supports Large Language Models from **OpenAI**, **Azure OpenAI**, and **Anthropic (Claude)**. Its flexible architecture means it can also integrate with homegrown LLMs or any provider-specific LLM your organization is using within its infrastructure, depending on the organisation's security policy.

### Requirements for Configuring Generative AI

- An account with one of: OpenAI, Azure OpenAI, or Anthropic.
- An API key for the relevant model.
- A chat-completions model (e.g. `gpt-5`, `claude-sonnet-4-6`).

To configure the service, the following environment variables need to be set. This can be done by creating the `docker-compose-custom.yml` file to have the custom configuration. If you already have a `docker-compose-custom.yml` file with some custom configuration elements, we will simply be editing that file instead.

```
cp docker-compose-basic.yml docker-compose-custom.yml
```

Inside the `docker-compose-custom.yml` file, the `api` service's `environment` section needs to contain the environment variables listed below. If the `api` service does not exist in the YAML file, it needs to be added first — the `docker-compose-ad.yml` file can be used as an example for that.

### Environment Variables — Core

The following variable names are the canonical, provider-agnostic ones (`GEN_AI_*`). The older provider-specific names (`OPEN_AI_*`, `ANTHROPIC_*`) are still honoured as fallbacks for backward compatibility, but new deployments should prefer `GEN_AI_*`.

| Variable | Default | Description |
| --- | --- | --- |
| `GEN_AI_ENABLED` | `false` | Generative AI enabled (`true` / `false`) for the instance. |
| `GEN_AI_SOURCE` | `OpenAI` | Provider to use. One of: `OpenAI`, `Azure`, `AzureAPIM`, `Anthropic`. |
| `GEN_AI_API_KEY` |  | API key for the chosen provider. |
| `GEN_AI_MODEL` | `gpt-5` | Chat model / deployment name. E.g. `gpt-5`, `gpt-4o`, `claude-sonnet-4-6`. |
| `GEN_AI_BASE_URL` | (provider default) | Override the provider endpoint URL. Leave blank to use the SDK default (`https://api.openai.com/v1` for OpenAI, `https://api.anthropic.com` for Anthropic). |
| `GEN_AI_PROXY_URL` |  | Forward-proxy URL if required to reach the provider endpoint. |
| `GEN_AI_ENABLED_EMAILS` |  | Comma-separated list of email addresses that should have access to the Generative AI features. If empty, all users have access. |

### Environment Variables — Anthropic-specific

Only relevant when `GEN_AI_SOURCE=Anthropic`. Defaults are tuned for typical Claude Sonnet workloads; raise the timeout for very long completions, raise the max-tokens ceiling when generating large structured outputs (flowcharts, test suites).

| Variable | Default | Description |
| --- | --- | --- |
| `GEN_AI_ANTHROPIC_MAX_TOKENS` | `16000` | Default `max_tokens` applied when a prompt doesn't specify one. Anthropic's API requires this field; setting it too low truncates large flowchart / test-suite responses. |
| `GEN_AI_ANTHROPIC_TIMEOUT_SECONDS` | `300` | HTTP timeout for the Anthropic SDK (applies to connect, read, write, and overall call). Raise for very long completions. |

### Example — OpenAI

```
api:
  environment:
    GEN_AI_ENABLED: "true"
    GEN_AI_SOURCE: "OpenAI"
    GEN_AI_API_KEY: "${OPENAI_KEY}"
    GEN_AI_MODEL: "gpt-5"
```

### Example — Anthropic

```
api:
  environment:
    GEN_AI_ENABLED: "true"
    GEN_AI_SOURCE: "Anthropic"
    GEN_AI_API_KEY: "${ANTHROPIC_KEY}"
    GEN_AI_MODEL: "claude-sonnet-4-6"
    # Optional -- raise for very large flowchart generation
    GEN_AI_ANTHROPIC_MAX_TOKENS: "32000"
    GEN_AI_ANTHROPIC_TIMEOUT_SECONDS: "600"
```

### Example — Azure OpenAI

```
api:
  environment:
    GEN_AI_ENABLED: "true"
    GEN_AI_SOURCE: "Azure"
    GEN_AI_API_KEY: "${AZURE_OPENAI_KEY}"
    GEN_AI_BASE_URL: "https://my-resource.openai.azure.com/openai/deployments/my-deployment"
    GEN_AI_MODEL: "my-deployment"
```

### Backward Compatibility

The previous variable names continue to work without change. The cascade is:

1. `GEN_AI_*` — canonical, provider-agnostic (preferred).
2. `ANTHROPIC_*` — convenience names when running Anthropic.
3. `OPEN_AI_*` — legacy names, kept working indefinitely.

For example, `GEN_AI_API_KEY` takes precedence; if unset, the system reads `ANTHROPIC_API_KEY`; if also unset, it falls back to `OPEN_AI_KEY`. Existing `docker-compose-custom.yml` files do not need to be updated.

Quality Modeller is Curiosity's flow-driven model-based tool which provides a range of accelerators and connectors for building flowcharts rapidly. Align all stakeholders to quality outcomes and create critical assets early, delivering superior software at speed.