We getting close to releasing another update and have done a number of tests with Agents app that will make it functional. With any luck the update will be published this week.
In a nutshell, Agents is an app that allows you to create automations and RAG-bots that can be based on different AI models (open-source, or hosted APIs), combined together and given "tasks" that would either run on cron or by triggers (webhooks or UNA events).
Agents use UNA API as a reference resource to create code that would do something in UNA. It can also use external APIs.
The idea is to create a tool that would allow you to add functionality/automations without needing to make new modules or modify existing apps.
For example:
Create a a Helper based on, say, OpenAI API and prompt/tune it to focus on identifying spam in posts. Call it "SpamGuru".
Then, create an Automator (you can use OpenAI/AssistantsAPI or another model - like LLama, add the SpamGuru helper to it and prompt it to "look at every new Post or Comment and identify if it is spam, if it is likely spam, hide the post, flag it for review. The Automator will generate PHP code that will run according to your prompt, it will use UNA events and actions. You assign it to a pre-create profile (an admin or a "bot" profile) to attribute this action to an account.
Then, your can create Assistants, which are more advanced and can "learn" but saving some data to vector to maintain long-term, large context for inference. For example, we'll be looking for more complex automations, like "save all posts that are flagged as spam, check if more than 3 posts by one user are likely spam, then suspend the user. If the user has more posts that are likely not spam than posts that are likely spam, then report the user.
We are also trying a few "built in" interfaces to access Agents. For example, a Studio Assistant which will help to configure UNA, or site-search augmentation which allows people to ask about site content using natural language.
To effectively integrate with other APIs, there should be a function to configure the base API URL, including the API key. Additionally, an endpoint for `/v1/openai/models` should be available to allow selection of any model directly from the API endpopint within `/studio/agents.php`.
Supporting a variety of additional endpoints would be beneficial for handling different types of data, such as vectors, embedings, files, videos, images, audio, and more. This approach would enable users to view the available models for each endpoint. Supporting multiple base URLs would allow integration with different providers, enabling the use of various OpenAI-compatible models from diverse sources for any task.
For example, in Coozila! AGI, several endpoints are available—some compatible with OpenAI and others from different providers. To work with UNA, the base URL needs to be hardcoded directly into the agents code:
UNA agents are promising but could benefit from enhancements to expand their capabilities. A useful feature would be allowing the agent to post content or asist the user, such as video, audio, text, job listings, or ads, and to customize post creation, such as articles or other content types.
Integrating messenger functionality would also be valuable. For example, registered users wishing to speak directly with an agent or visitors seeking site information could use messenger to communicate. This would enable users to receive support related to products or services, or interact with AI for specific tasks like managing their accounts or responding to messages. This feature could be especially beneficial for companies, as it allows an AI to handle personalized welcome messages and other interactions when human agents are offline.
A well-trained agent can provide quick, relevant responses and offer useful links to visitors, potentially increasing the conversion rate from visitors to customers. Implementing features like personalized AI interactions and dynamic content posting could further enhance engagement and functionality.
Hello. Models in UNA Agents can have their own API Keys and Endpoints. Model selector is also available but each model requires its own specific integration and for now we have 2 models integrated. Currently AI Assistant can be used in Studio Help, Site Search and "Ask AI" block. Thank you for the ideas.
To enable the use of open-source models from various providers, it would be beneficial to allow direct modifications to the OpenAI API base configuration in UNA sdudio. This involves:
Enabling Direct Modifications:
Allow configuration of the OpenAI API base so it can be updated to include support for open-source models from different providers.
Implementing an Endpoint for Model Selection:
Add an endpoint for selecting from available models, including those from various open-source providers.
Testing and Validation:
Ensure that the new configuration allows seamless integration and usage of both OpenAI and open-source models.
Comment: I have directly modified the script to change the OpenAI API address and used local models or models from other free APIs. However, it would be more flexible if the OpenAI address were a variable that can be changed directly from the studio. This would simplify configuration and allow users to set up various local or online providers according to their preferences.
It seems this feature is locked to OpenAI, there is no place to enter in your own models or own API end points.
It also seems linked to specific user profiles/organizations. It would be good if it were linked to member levels so you can provide an AI integration for members of specific paid levels.
Providers is linked to Shopify - no other option? Can't add custom providers?
Helpers seem to be forced to use OpenAI models with prompts, no option for custom helpers. If you do create on and try "tune" it (whatever that means) it errors :
Fatal error: Uncaught Error: Call to undefined method BxDolAI::getHelper() in /home/XXXXXX/public_html/studio/template/scripts/BxBaseStudioAgents.php:116 Stack trace: #0 /home/XXXXX/public_html/studio/template/scripts/BxBaseStudioWidget.php(121): BxBaseStudioAgents->getHelpers() #1 /home/XXXXX/public_html/studio/agents.php(26): BxBaseStudioWidget->getPageCode() #2 {main} thrown in /home/XXXXX/public_html/studio/template/scripts/BxBaseStudioAgents.php on line 116
Automators : Seem to link Providers to Helpers. So at this stage is limited to Shopify and a gpt with a prompt ?
Note : Even with OpenAI there is no place to insert your API Key?
G
I do not have that app
It looks like something AI related.
C
I have it but not used it yet as I can't find enough information regarding it
Hello! Here @Andrey Yasko answered about it. It seems to be a very interesting function: https://unacms.com/i/hello-are-there-instructions-and
We getting close to releasing another update and have done a number of tests with Agents app that will make it functional. With any luck the update will be published this week.
In a nutshell, Agents is an app that allows you to create automations and RAG-bots that can be based on different AI models (open-source, or hosted APIs), combined together and given "tasks" that would either run on cron or by triggers (webhooks or UNA events).
Agents use UNA API as a reference resource to create code that would do something in UNA. It can also use external APIs.
The idea is to create a tool that would allow you to add functionality/automations without needing to make new modules or modify existing apps.
For example:
We are also trying a few "built in" interfaces to access Agents. For example, a Studio Assistant which will help to configure UNA, or site-search augmentation which allows people to ask about site content using natural language.
C
That is one awesome tool, you guys are amazing this will be fantastic as it is developed.
Sounds like it better fits to the zeitgeist. Good on you Andrey.
To effectively integrate with other APIs, there should be a function to configure the base API URL, including the API key. Additionally, an endpoint for `/v1/openai/models` should be available to allow selection of any model directly from the API endpopint within `/studio/agents.php`.
Supporting a variety of additional endpoints would be beneficial for handling different types of data, such as vectors, embedings, files, videos, images, audio, and more. This approach would enable users to view the available models for each endpoint. Supporting multiple base URLs would allow integration with different providers, enabling the use of various OpenAI-compatible models from diverse sources for any task.
For example, in Coozila! AGI, several endpoints are available—some compatible with OpenAI and others from different providers. To work with UNA, the base URL needs to be hardcoded directly into the agents code:
[https://agi.coozila.com/api/docs/](https://agi.coozila.com/api/docs/)
UNA agents are promising but could benefit from enhancements to expand their capabilities. A useful feature would be allowing the agent to post content or asist the user, such as video, audio, text, job listings, or ads, and to customize post creation, such as articles or other content types.
Integrating messenger functionality would also be valuable. For example, registered users wishing to speak directly with an agent or visitors seeking site information could use messenger to communicate. This would enable users to receive support related to products or services, or interact with AI for specific tasks like managing their accounts or responding to messages. This feature could be especially beneficial for companies, as it allows an AI to handle personalized welcome messages and other interactions when human agents are offline.
A well-trained agent can provide quick, relevant responses and offer useful links to visitors, potentially increasing the conversion rate from visitors to customers. Implementing features like personalized AI interactions and dynamic content posting could further enhance engagement and functionality.
Hello. Models in UNA Agents can have their own API Keys and Endpoints. Model selector is also available but each model requires its own specific integration and for now we have 2 models integrated. Currently AI Assistant can be used in Studio Help, Site Search and "Ask AI" block. Thank you for the ideas.
To enable the use of open-source models from various providers, it would be beneficial to allow direct modifications to the OpenAI API base configuration in UNA sdudio. This involves:
Comment: I have directly modified the script to change the OpenAI API address and used local models or models from other free APIs. However, it would be more flexible if the OpenAI address were a variable that can be changed directly from the studio. This would simplify configuration and allow users to set up various local or online providers according to their preferences.
It seems this feature is locked to OpenAI, there is no place to enter in your own models or own API end points.
It also seems linked to specific user profiles/organizations. It would be good if it were linked to member levels so you can provide an AI integration for members of specific paid levels.
Providers is linked to Shopify - no other option? Can't add custom providers?
Helpers seem to be forced to use OpenAI models with prompts, no option for custom helpers. If you do create on and try "tune" it (whatever that means) it errors :
Fatal error: Uncaught Error: Call to undefined method BxDolAI::getHelper() in /home/XXXXXX/public_html/studio/template/scripts/BxBaseStudioAgents.php:116 Stack trace: #0 /home/XXXXX/public_html/studio/template/scripts/BxBaseStudioWidget.php(121): BxBaseStudioAgents->getHelpers() #1 /home/XXXXX/public_html/studio/agents.php(26): BxBaseStudioWidget->getPageCode() #2 {main} thrown in /home/XXXXX/public_html/studio/template/scripts/BxBaseStudioAgents.php on line 116
Automators : Seem to link Providers to Helpers. So at this stage is limited to Shopify and a gpt with a prompt ?
Note : Even with OpenAI there is no place to insert your API Key?