How to set up an AI provider in rConfig V8

rConfig doesn’t ship its own AI model. You bring one: a cloud API such as Anthropic or OpenAI, or a model you run yourself with Ollama. This page shows how to connect each provider type, test it, and make it the active provider that every AI capability uses.

If this is your first time setting up the AI Center, start with Getting Started. It covers the steps before and after this one.

Prerequisites

• The aicenter.update permission. The AI Settings tab is shown only to users whose role has it, and saving a provider needs it. Admins have it. From V8.4.2 the tab follows the permission, not the role name, so a custom role with aicenter.update sees it too.
• AI_CENTER_ENABLED=true on the server (see below).
• Credentials for your provider: an API key for a cloud provider, or the URL of a model server the rConfig server can reach.

Enable the AI Center feature flag

AI_CENTER_ENABLED isn’t in a default .env file, and it defaults to false. Add it to /var/www/html/rconfig8/.env, then clear rConfig’s caches:

AI_CENTER_ENABLED=true

cd /var/www/html/rconfig8
php artisan rconfig:clear-all

This flag is the master switch

While AI_CENTER_ENABLED is false, the AI Center is missing from the sidebar, every /ai-center page returns a 403, and every AI button elsewhere in rConfig is hidden. There’s no switch for it in the UI; it lives in .env only.

Which provider should you choose?

If you need…	Choose	Why
The best answers with least setup	Anthropic or OpenAI	Paste an API key and go. Strong reasoning on long configs.
To stay inside your Google Cloud estate	Google Gemini	Key from Google AI Studio.
To stay inside an Azure tenancy with your own deployment	Azure OpenAI	Uses your Azure resource and data-processing agreement.
Another hosted model vendor	Mistral, xAI (Grok), DeepSeek, Groq, Cohere or OpenRouter	From V8.4.2. Paste the vendor’s API key and pick a model. OpenRouter reaches many vendors’ models with one key.
Configs that never leave your network	Ollama	Model runs on hardware you control. No API key, no internet needed for inference.
A self-hosted model server or gateway (vLLM, LM Studio, LiteLLM)	OpenAI-Compatible	Any server that speaks the OpenAI chat API.

Model size matters

Configuration analysis works best on capable models. Small local models (7B to 8B parameters) handle short configs and simple questions, but expect weaker answers on long, complex configs. If you run locally, use the largest model your hardware runs comfortably.

Step-by-step walkthrough

1. Open AI Settings

1. Click AI Center in the left sidebar.
2. Select the AI Settings tab.

If no provider exists yet, you’ll see No provider configured yet. with an Add Provider button.

2. Add the provider

Click Add Provider (or press Alt+N).

From V8.4.2, the form is built from a model catalogue that ships with each rConfig release. The catalogue decides which providers are listed, which fields each one needs, which models are offered, and which model is recommended. New providers and models arrive with rConfig upgrades, so the form on your server is always the reference for what your release supports.

Field	What to enter
Display name	A name you’ll recognise on the provider card, such as “Production Anthropic” or “On-prem Ollama”. Must be unique.
Provider	Every provider your release supports. Changing it clears the fields you’ve entered.
API Key	Required for hosted providers, optional for OpenAI-Compatible, not shown for Ollama. Stored encrypted and never shown again; the card only says Configured.
Endpoint	Azure OpenAI only: your resource’s endpoint.
Base URL	Ollama and OpenAI-Compatible only: the address of your model server, as seen from the rConfig server.
Model	For hosted providers, a list of current models with the recommended one marked; pick Custom model ID… to type any other. A text box for Ollama, OpenAI-Compatible and Azure (Deployment name). Leave it empty to use the provider’s default.

Each field’s help text in the form says where to find the value. If a saved provider uses a model its vendor has retired, rConfig flags it so you can pick a current one.

Then follow the tab for your provider type:

Hosted providers

Anthropic, OpenAI, Google Gemini and the other hosted vendors in the list (from V8.4.2 these include Mistral, xAI, DeepSeek, Groq, Cohere and OpenRouter) all work the same way:

1. Create an API key in the vendor’s console, for example the Anthropic Console, the OpenAI API keys page or Google AI Studio.
2. In rConfig, set Provider to the vendor and paste the key into API Key.
3. Pick a model. The recommended one is a good starting point for config work; a larger model gives better answers on long, complex configs at a higher cost.

OpenRouter reaches many vendors’ models with one key, and its model IDs take the form vendor/model. All hosted providers are cloud services: configs go to the vendor under its data terms.

Azure OpenAI

1. In the Azure portal, open your Azure OpenAI resource. On Keys and Endpoint, note the Endpoint and one of the Keys. In Azure AI Foundry or Azure OpenAI Studio, note the deployment name of the model you want to use.
2. In rConfig, set Provider to Azure OpenAI and paste the key into API Key.
3. Paste the endpoint into Endpoint: the resource root only, for example https://your-resource.openai.azure.com. rConfig adds /openai/v1 itself, and trims a longer URL pasted from the portal back to the root.
4. Enter your deployment name in Deployment name: the name you gave the deployment, not the model ID.

From V8.4.2 the endpoint is set in the form. Before V8.4.2 it came from AZURE_OPENAI_URL in .env; Azure providers saved without an endpoint keep reading it from there, so existing setups carry on working. Edit the provider and add the endpoint to move it into the form.

Ollama

1. On the model server, install Ollama and pull the model you want to use with ollama pull <model>.
2. By default Ollama listens on 127.0.0.1 only. If it runs on a different host from rConfig, start it with OLLAMA_HOST=0.0.0.0 so rConfig can reach it, and allow TCP port 11434 through the firewall.
3. In rConfig, set Provider to Ollama (local) and set Base URL to the server address, for example http://10.10.20.15:11434. Use http://localhost:11434 only if Ollama runs on the rConfig server itself.
4. Enter the model name exactly as ollama list shows it, including the tag. Before judging answers, check Ollama’s context window: see Sizing a local model.

OpenAI-Compatible

Use this for vLLM, LM Studio, LiteLLM, or any server that implements the OpenAI chat completions API.

1. Set Provider to OpenAI-Compatible.
2. Set Base URL to the server’s OpenAI-style base path, which usually ends in /v1, for example http://10.10.20.16:8000/v1.
3. Enter the model name exactly as the server lists it.
4. If your server requires a key, paste it into API Key (optional, from V8.4.2). Leave it empty for servers that don’t need one.

Add Provider form for a cloud provider

Add Provider form for Ollama

Click Add Provider to save. Saving doesn’t test or activate the provider.

3. Test the connection

Each saved provider appears as a card. Click Test on the card. rConfig sends the prompt “Reply with exactly one word: ok” using the saved credentials and model, and shows a toast:

• Connection successful. with the model’s reply and the round-trip time, for example Response: ok (840ms).
• Connection failed. with the reason, such as “Authentication with the AI provider failed. The configured API key appears to be invalid or missing.” See Troubleshooting for what each message means.

4. Activate it

After a successful test of an inactive provider, the card says Connection works — Activate this provider to start using it. Click Activate. The card shows an Active badge.

Only one provider is active at a time. Activating one deactivates the others. Every capability, chat, and narration uses the active provider; there’s no per-capability choice.

Two saved providers, one active

Saved is not active

If providers exist but none is active, AI Settings shows No AI provider is activated and every AI feature replies with the same message. This is the most common reason the AI Center “doesn’t work” after setup.

5. Tune the model controls (optional)

Below the provider cards, Model Controls applies to the active provider:

• Max Output Tokens: the longest a single answer can be, from 256 to 32768. Raise it if long answers get cut off. Leave it empty for the provider’s default.
• Temperature: from 0.0 to 1.0. Lower values give focused, repeatable answers, which suits config analysis. Leave it at the default unless you have a reason to change it.

Click Save Model Controls.

Max tokens is a length cap, not a spend budget

It limits the length of each response. It doesn’t cap monthly spend. Set spend limits in your provider’s own console.

Switching, editing, and removing providers

• Switch providers: click Activate on another card. The change applies to the next request.
• Rotate a key: edit the card and paste the new key. Leave API Key empty to keep the stored key.
• Delete: deleting the active provider leaves rConfig with no active provider. Activate another one straight away, or AI features stop working.

Common gotchas

Common provider setup mistakes

• URL works from your laptop but not from rConfig: rConfig’s server calls the model, not your browser. localhost means the rConfig server itself. Test with curl http://<ollama-host>:11434/api/tags from the rConfig server.
• Model not found: the model name must match the provider exactly. For Ollama, use the name from ollama list, including the tag. For Azure, use the deployment name, not the model family.
• Azure test fails with a connection error: check Endpoint is the resource root from Keys and Endpoint (for example https://your-resource.openai.azure.com). For an Azure provider saved before V8.4.2 with no endpoint, rConfig reads AZURE_OPENAI_URL from .env; edit the provider and add the endpoint instead.
• Azure model not found: Deployment name must be the name you gave the deployment, not the model ID such as gpt-4o.
• Rate limits and quota: providers enforce their own limits. rConfig shows the provider’s rate-limit or out-of-credit error rather than failing silently.

Which provider fits?

Choosing an AI provider and model compares cloud, Azure and self-hosted designs. Running Ollama? Read Sizing a local model first: its default context window is too small for long configs.

What’s next

• Turn AI capabilities on
• Ask your first question with Config Analysis & Explain
• Review what data leaves your network for each provider type