# AI Settings

> Settings > Integrations > AI Settings: enable the AI engine, choose from ~20 model providers (including self-hosted), manage API keys, and test the connection—system-wide.

Source: https://totalshiftleft.ai/help-center/product-documentation/ai-settings

## Overview

**AI Settings** is the system-wide control panel for AI-assisted test generation: whether the AI engine is on, which provider and model are used, the model parameters, and the encrypted API keys. These settings apply to the whole system—to control *what* gets generated for a single project, see [Test Generation Settings](/help-center/product-documentation/test-generation-settings).

Where to find everything: [Administration Settings](/help-center/product-documentation/administration-settings). Explore the capability at [multi-provider AI](/features/ai-test-generation/multi-provider-ai).

## Before you begin

- You must be an **Administrator**. Non-admins see **Access Denied** (*"You need Administrator privileges to access AI settings."*).
- Have the provider API key ready for the model you plan to use. Keys are encrypted at rest and visible only to administrators.
- Nav path: **Settings → Integrations → AI Settings**. The page header reads **AI Settings** with **Reset to Defaults** and **Save Settings** at the top-right.

## Step 1 — Turn the AI engine on

In the **Use AI Engine** card, toggle the switch on. The help text reads *"When enabled, Selected AI Model will be used for test generation. When disabled, default test generation engine will be used."* With it off, generation falls back to the deterministic engine everywhere.

## Step 2 — Configure Test Generation Settings

The **Test Generation Settings** card holds system-wide generation defaults:

| Field | What it does |
|-------|--------------|
| **Test Types (comma-separated)** | Which test types AI may generate. Allowed: `status, format, data, performance, security`. Leave empty to use all. |
| **Enable Custom Instructions** | When ticked, your instruction text is applied to generations. |
| **Custom Instructions** | Free-text guidance (up to **2000 characters**; a live counter shows usage). Disabled until the toggle above is on. |
| **Enable automatic test generation** | Opt in to automated generation flows. |
| **Enable smart test suggestions** | Opt in to contextual test suggestions. |
| **Test design techniques** | Checkboxes: **Boundary values**, **Equivalence partitioning**, **Decision table**. |

## Step 3 — Pick a model provider

In **AI Model & API Configuration**, use the **Model Provider** dropdown. There are ~20 provider families: OpenAI, Anthropic, Google Gemini, Meta Llama, xAI Grok, Mistral AI, Perplexity, Azure OpenAI, API Free, Cohere, Together AI, Groq, DeepSeek, AI21 Labs, Replicate, Hugging Face, and **Self-hosted (Ollama / LM Studio / vLLM)**.

Choosing a provider selects its first model by default. A **Security Notice** on this card reminds you that keys are encrypted and admin-only.

### Self-hosted (Enterprise only)

**Self-hosted (Ollama / LM Studio / vLLM)** is an Enterprise capability. On non-Enterprise plans the option shows *"— Enterprise only"* and selecting it raises *"Self-hosted LLM is available on the Enterprise plan only."* with an **Upgrade** link. When available, selecting it reveals:

- **Base URL (OpenAI-compatible /v1)** — with **Quick fill** buttons for **Ollama** (`http://localhost:11434/v1`), **LM Studio** (`http://localhost:1234/v1`), and **vLLM** (`http://localhost:8000/v1`). From Docker, use `http://host.docker.internal:<port>/v1`.
- An **Advanced — embedding model** section (collapsed) with an optional **Embedding model** field (default `nomic-embed-text`) used only for offline RAG / semantic search when no cloud embedding key is set.

## Step 4 — Choose the specific model

Under **Specific Model**:

1. Pick a model from the dropdown. Groups are **Current**, **✓ Available for your key** (or **✓ Installed on your server** for self-hosted), **★ Custom (added)**, and **— Catalog / — Presets**.
2. Click **Refresh available models** to query the provider for the models your key/server can actually use (tooltip: *"Query this provider for the models its key can use"*). Tick **Show all** to include snapshots, previews, and legacy models.
3. To use a model not in the catalog, choose **➕ Add a custom model…**, type the model id, and press Enter. Custom models appear as removable chips (click **×** to remove).

## Step 5 — Set model parameters

In the three parameter fields:

| Field | Range | Notes |
|-------|-------|-------|
| **Max Tokens** | 100–8000 | Maximum tokens for the AI response. |
| **Temperature** | 0–2 (step 0.1) | Creativity level. |
| **Timeout (seconds)** | 10–300 | API request timeout. |

## Step 6 — Manage API keys and test the connection

Under **API Keys Management**:

1. Enter the key in the **{Provider} API Key (Currently Selected)** field—the label and placeholder match the selected model's provider (e.g. `sk-...`, `sk-ant-...`, `AIza...`).
2. Click **Test Connection** (label shows **Testing…**). This *"Sends a tiny ping to {provider} to verify the key and model. Nothing is saved."* Success shows **✓ Connection successful** (with the reply); failure shows **✗ Connection failed** plus a **Code**. A successful test offers **Save These Settings**.
3. Expand **Edit All API Keys** (**Hide** to collapse) to manage every provider's key in one place, including a **Self-hosted API key (optional)** and **Self-hosted Base URL**.
4. **Azure OpenAI** models additionally require **Azure OpenAI Endpoint** (e.g. `https://your-resource.openai.azure.com/`) and **Azure OpenAI Deployment** (the deployment name).

The **Selected Model Information** box summarizes the model, provider, required key, and **Key Status** (Verified, Key set — not verified yet, or Not configured). Changing the model, a key, the base URL, or Azure endpoint/deployment clears a prior verification.

## Step 7 — Save (or reset)

- Click **Save Settings** (top-right; shows **Saving…**). On success: *"AI settings saved successfully."*
- **Reset to Defaults** opens a confirmation (*"Are you sure you want to reset all AI settings to defaults?…"*); confirm with **Reset**. This cannot be undone.

## Troubleshooting

- **Access Denied** — you're not an Administrator; ask an admin to make the change.
- **✗ Connection failed** — check the key, model, and (for Azure/self-hosted) endpoint/base URL; the **Code** line names the failure. Re-run **Test Connection**.
- **Key Status: "Not configured" / "Key set — not verified yet"** — enter the key and run **Test Connection** before saving so generation can use it.
- **A key field is blank with a warning after an update** — the stored ciphertext couldn't be decrypted (typically after an encryption-key change); re-enter the flagged provider keys and save.

## Best practices

- Keep one **stable** model/configuration per environment so results are comparable.
- Never commit API keys to source control or share them in tickets; rotate on your standard schedule.
- After changes, run **Test Connection**, then spot-check [AI-generated tests](/help-center/product-documentation/ai-generated-tests) on a sample endpoint.

## Related articles

- [AI-generated tests](/help-center/product-documentation/ai-generated-tests)
- [Test Generation Settings](/help-center/product-documentation/test-generation-settings)
- [User Policies](/help-center/product-documentation/user-policies)