---
title: "Configure Generative AI (Custom) Connection"
slug: "configure-generative-ai-custom-connection-new"
updated: 2026-07-23T10:47:22Z
published: 2026-07-23T10:47:22Z
canonical: "knowledge.curiositysoftware.ie/configure-generative-ai-custom-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 (Custom) Connection

## Configure Generative AI (Custom) Connection

This article explains how to configure a custom Generative AI connection for your {{glossary.Test Modeller}} workspace using the Connectors page. A custom connection is required when you want to point the platform at your own LLM endpoint instead of (or in addition to) the default model shipped with your deployment.

Use a custom connection when you want any of the following:

- Use a provider or model that differs from the instance default (e.g. switch a single workspace to Claude while the rest of the platform stays on GPT-5).
- Route Generative AI traffic through your organisation's Azure OpenAI deployment or APIM gateway.
- Maintain separate API keys, quotas, or audit trails per workspace or per project.
- Override the model on a per-prompt-template basis (e.g. heavy structured-output prompts on a larger model, classification prompts on a cheaper one).

If you only need to set one provider for the entire instance, prefer the environment-variable approach described in **Configure Generative AI (Default) Connection** — it doesn't require a Connector entry and applies platform-wide.

### Prerequisites

- A workspace where you have the Administer permission.
- An API key (and any provider-specific extras — APIM subscription key, etc.) for the LLM you intend to point at.
- The Generative AI feature must be enabled for your instance. On cloud, this is on by default. On on-prem, set `GEN_AI_ENABLED=true` per the Default Connection guide.

### Create a Connector

Navigate to **Workspace → Connectors → New Connector**, then choose **Generative AI** as the connector category. The supported connection types are:

| Connection Type | When to Use |
|---|---|
| **OpenAI** | Direct OpenAI account (api.openai.com). The simplest setup. |
| **Azure** | Azure OpenAI direct — request goes straight to your Azure OpenAI resource using an API key. |
| **Anthropic** | Claude models via Anthropic's direct API (api.anthropic.com) or an Anthropic-compatible endpoint. |

### Common Fields

These fields appear on every Generative AI connector regardless of type. Provider-specific extras are listed in the per-type sections below.

| Field | Description |
|---|---|
| **Name** | Friendly identifier shown in the Connector list and the per-prompt provider picker. |
| **Source** | Connection type — see the table above. |
| **Base URL** | Provider endpoint. Leave blank to use the SDK default for the chosen source (`https://api.openai.com/v1` for OpenAI, `https://api.anthropic.com` for Anthropic). Required for Azure / APIM. |
| **Model** | Default model (or deployment name, for Azure) used when a prompt template doesn't specify its own. E.g. `gpt-5`, `gpt-4o`, `claude-sonnet-4-6`. |
| **API Key** | Provider API key. Stored encrypted at rest. |
| **Proxy URL** | Optional forward-proxy URL, in the form `http://user:pass@proxy.host:port`. Needed only if outbound HTTPS to the provider must transit a corporate proxy. |

### OpenAI

The default and simplest setup. Point at `https://api.openai.com/v1` (or leave **Base URL** blank to use the SDK default).

### Azure (Direct)

For Azure OpenAI resources accessed directly without an APIM gateway. **Base URL** should include the deployment path; the platform extracts the deployment name automatically.

```
https://my-resource.openai.azure.com/openai/deployments/my-deployment
```

The `?api-version=...` query string is optional — omit it for the stable Azure surface, pass `?api-version=preview` (or a dated preview slug) to opt into features like Responses API and structured outputs. If omitted, {{glossary.Test Modeller}} defaults to `preview` to preserve historic behaviour.

### Anthropic

For Claude models. Leave **Base URL** blank to use Anthropic's default endpoint (`https://api.anthropic.com`) or set it to point at an Anthropic-compatible proxy.


### Assigning the Connector

Once a Generative AI connector is saved, it becomes selectable in two places:

1. **Per prompt template** — on each prompt template (Settings → Generative AI → Prompt Templates), the **Connector** dropdown lets you pin that template to a specific connector. Useful for routing heavy generation prompts (e.g. flowchart creation) to a different model than lightweight prompts (e.g. classification).
2. **Workspace default** — set one connector as the workspace default to make it the fallback for any prompt template that doesn't specify its own connector.

If no connector is set on either layer, the platform falls back to the instance-wide default configured via environment variables.


### Testing the Connection

After saving the connector, open any prompt template that uses it and run **Test Prompt** to confirm the credentials and endpoint resolve correctly. Errors appear inline with the SDK-provided detail (e.g. `401 Unauthorized`, `invalid_api_key`, `model not found`).