Moving AI provider configuration to AI Core
Moving AI provider configuration to AI Core
This article explains how administrators move the AI provider connection from the earlier provider settings (OpenAI, Azure OpenAI or the legacy Matrix42 GenAI provider) to M42 AI Core. After the move, Neo, Actions and the writing assistance all use the language model that you have configured in AI Core, and you manage that model in one place.
What Changes in 26.2
- The AI provider settings are read-only. Administrators can view and copy the existing settings and switch between providers that are already configured. New provider connections are set up in AI Core.
- The provider that was called "M42 GenAI with Knowledge Discovery" is now called M42 AI Core. The provider configuration has a button for testing the AI Core connection and a link to the AI Core administration page.
- Existing direct connections to OpenAI and Azure OpenAI keep working until version 2027.1. Moving to AI Core before then keeps your Actions and Neo running without a break.
Before You Start
- AI Core is deployed for your environment and the platform can reach it. The AI Core components are a separate part of the M42 Professional platform, and the steps below cannot be done without them.
- The environment is upgraded to version 26.2.
- You need access to the AI Core administration page.
- Have the API key and the URL of your language model ready. If you no longer have the API key of the earlier provider, create a new one in your provider account.
- Note the model name that the current provider settings use. Open Administration → Neo → Provider configuration to view and copy the existing settings.
Move the Configuration to AI Core
Step 1. Add the Language Model as a Secret
AI Core stores the connection to your language model as a secret.
- Open the AI Core administration page. The provider configuration has a link to it.
- Go to Config → Secrets.
- Select Add secret, and enter a name, the API key and the URL of your language model.
- Save the secret.
Step 2. Allow Pass-through for Actions and Writing Assistance
Actions that work without an AI agent, as well as the writing assistance, send their requests directly to the language model. This is called pass-through mode. It keeps these existing features working after the move.
- In AI Core, go to Config → Passthrough Chat Settings.
- In Backend, select the backend that uses the secret that you added in Step 1.
- In Force model, enter the model name that you noted before you started.
- Select Save Configuration.
- To check the result, go to Chat & Test → Passthrough, send a message and make sure that the chat responds.
Step 3. Set the default backend for the default agent
Neo uses a default agent that is created in AI Core automatically. The agent is created the first time that Neo needs it, and it uses the default backend of AI Core. Set the default backend before that happens, so that the creation succeeds.
- In AI Core, go to Config → Agents → Defaults for new agents.
- In Default Backend, select the backend that you used in Step 2. If you use only one language model, this is the one that you added as a secret.
- Save the configuration.
If the default agent has not been created because no default backend was set, Neo shows an error. Set the default backend, wait for about a minute, and then send another message in Neo. The platform tries to create the agent again when it receives the message. The first message after the creation takes a little longer to answer.
Step 4. Switch the Provider and Test the Connection
- Go to Administration → Neo → Provider configuration.
- Select M42 AI Core as the provider.
- Select Test your M42 AI Core connection. The test succeeds when the platform can reach AI Core.
Step 5. Check Actions and Neo
- Run an Action on a data card. This checks the pass-through connection from Step 2.
- Send a message in Neo. This checks the default agent from Step 3.
Troubleshooting
The connection test shows the error "Invalid API key". In AI Core, go to Config → API & Authentication and select Reset to generate a new API key. Then go back to the provider configuration and test the connection again.
Neo shows an error when you send the first message. Check that Step 3 is done, and then send another message after about a minute.
Setting names and availability can change between versions.
Table of Contents