Has anyone set up their own AI agent to be able to query the database and reference the Epicor Help file to answer deep questions and to find future exceptions?
Use case: Say we sell a product and based on changing usage our min on hand quantity might not be adequate (just one hypothetical for demonstration). Wondering if you could ask the prompt something like: “Based on this customer’s recent purchases check the min on hands for all the raw materials for the products they purchase and suggest if any min on hand quantities should be adjusted.”
I’m thinking some hypothetical agent could have access to the database schema and/or a copy of the database and then the agent would also have been trained on the Epicor Application help file so it understands the relationships between the tables. Then you can just interact with it as you might with a typical AI agent.
Has anyone tried something like that or have any recommendations on where to start?
Before OpenClaw, one would recommend writing an MCP Server and build tools that do common tasks: Open orders, Inventory Inquiry, etc. As it turns out, agents like OpenClaw, etc. really, really like command line interfaces. They are able to do the same tool selection while using less tokens than the context that MCP servers dump.
There was a thoughtful podcast about this talking about the security implications (which MCP servers are good at, BTW).
I like the new HI Model. Their slogan “HI. What was old is now new.”
It’s a very powerful paradigm.
They say under the covers processing engine uses an organic framework with extensive reasoning, cognative, and statistical capabilities. Works really well integrating with other AI engines, models and Agents.
Sounds a bit like Ash from Alien.
I don’t know much more about it other than that it’s been around for a long time.
Oh did I mention it has a special common sense capability, it can be a bit spotty at the moment.
For transparency, this post was made using the HI Agentic agent M.E. which I have been working on for quite a while now.
