Virtual Chat agent

This docker-compose repository should set up a collection of working Moodle and solr containers that will run the Virtual Chat Agent demonstrator.

Installation

1. Clone this repository.
2. Run docker-compose build. This should build moodle, mysql and solr images.
3. Run docker-compose up to bring up 3 containers.

Solr installation.

1. Once the solr server is runnng, access the command line via docker exec -it server-solr-1 bash

2. On the command line run solr create -c moodle to create the Moodle collection.

3. From the /var/solr/orig copy the configuration files over the Moodle collection configuration.

   This will add in the configuration setings for the DenseVectorField types and the configuration for the Tika extraction library.

   cp -r /var/solr/orig/ /var/solr/data/moodle/conf/

4. Restart solr. This is easiest done by doing a docker-compose down and docker-compose up.

Moodle Installation

1. Access http://localhost:8080 and run the Moodle installation, or acccess the docker container and run php /var/www/html/admin/cli/install_database.php

AI Provider configuration

An AI Provider will need to be configured.

1. Access the local_ai AI Providers settings at http://localhost:8080/admin/category.php?category=local_ai_settings, or via the Local Plugins section.
2. Select the Providers option.
3. Click on the OpenAI API Based Provider
4. Provide a Name. This will be displayed to users selecting an AI provider.
5. Set the Base URL. For OpenAI this will be https://api.openai.com
6. Set your API Key. This is configured on the Open AI platform .
   1. Access OpenAI platform
   2. Select a Project from the dropdown
   3. Select the Dashboard option from the top right menu
   4. Select API Keys from the left hand menu.
   5. Use th "Create new secret key` button in the top right to create new key.
7. Expand Features
8. Check the Allow Chat option
9. Set the Completions path option. This is the relative URL to the Base URL to text completion / chat API. For OpenAI this is /v1/chat/completions.
10. Set the Completion Model to an appropriate value for the model you wish to use. gpt-4o is the current latest Open AI model.
11. Enable the Allow Embeddings
12. Set the Embeddings path option. This is the relative URL to the Base URL to generate an embedding from text. For OpenAI this is /v1/embeddings.
13. Set the Embedding model to appropriate text embedding model. For OpenAI this is text-embedding-3-large or text-embedding-3-small

The Content Constraints section allows you to select a Moodle Category or Course that this provider will be available in.

If you do not select anything this provider instance will be accessible throughout your site.

Global Search Configuration

Once basic Moodle is installed, you need to configure the Global Search.

1. Access the global search configuration page http://localhost:8080/admin/settings.php?section=manageglobalsearch

2. Select the Solr for Rag search engine. This should have been pulled from https://github.com/mhughes2k/moodle-search_solrrag as part of the docker build of the moodle image.

3. Click on the Save Changes button.

4. The 3. Set up Search engine option should display "no solr configuration found".

5. Click on the 3. Setup search engine option.

6. Under AI Settings, set the Choose Provider option to an AI Provider that has been defined. A global provider is probably best.

7. For Choose File Content extractor choose Solr with internal tika.

   A standalone tika server is not fully implemented.

8. Under Connection Settings, set the Host name to solr.

9. Set the Index name to moodle. Unless you changed this in Step 2 of the solr configuration.

10. Click Save changes.

11. Return to the Manage global search configuration page.

12. You should have all "green" statuses for Steps 1 - 3.

13. Access Step 5 Enable global search.

14. Set Enable global search to "Yes".

15. If your cron is running on a regular basis (as it should), the global search indexer should now start indexing content in moodle sites.

    You can run php search/cli/indexer.php within the moodle container to trigger the indexing process as well.

Next Steps

1. Restore or create a course

2. Create an instance of the "AI Chat (Reference)" activity.

3. Give it a name.

4. In AI Provider select an AI provider. This doesn't have to be the same AI Provider that the search uses. This provider is used to synthesise the response to the user.

5. You should get a prompt and be able to interrogate your course.