I bought an RTX 6000 Pro Blackwell edition. Nothing interesting going on over here.
4 Likes
Oooh! I see an Insights Presentation in your future!!! (If Epicor doesnât feel threatened)
1 Like
Does your organization have administrative overhead for any of these document tasks? If so, this can free up those resources to do other things and allow the workload to grow without adding more labor resources. (Not saying it replaces people entirely!)
At our parent organization, we have plenty of these use cases and as business growth continues, there has been a workload increase in document related areas. Instead of an entire new hire, you get a professional device at a one-time cost.
If you donât have AP automation, but workload is increasing there, these devices can solve the same problem with only a recurring power cost that is surely much less than a much larger one-time PLUS recurring subscription (maintenance).
Just the document automation piece optimizing administrative resources is worth it alone. The rest is bonus.
Iâd recommend going the number route. Pitch the ROI and low risk, as it is an easy calculation in this instance. Itâs a model which doesnât need updates if the current acceptance rate meets company standards. Itâs entirely within your control, including the model. Finally, it is a really decent server if you donât use it for anything else.
2 Likes
Iâm jealous. Though you could get two DGX Spark devices joined together for the same cost as just the card. The performance difference, though 
Yeah I had considered DGX or similar, but I just couldnt justify the performance vs cost. I am running 27B coding model + 100K context at 90t\s. Not too shabby.
2 Likes
Thanks to some studying on 3D rendering methods, I was able to utilize AI for a for-reals 3d rendering engine, shaders and all. This is rendered in WebGPU. It is well capable of running at 240fps on my video card/screen, although the room gets quite warm. It even supports 4x MSAA for S&G. This version doesnât need to know about your data ahead of time. It has heuristics to detect the most likely dataview, columns, and graph type to suit the dataset on load, and all of that is dynamically adjustable at runtime.
It has touch/mobile support, and runs well on the phone.
This version is deterministic (it is not built on the fly by the AI, rather coded ahead of time as a traditional app would be. Less flexibility but more predictable 
)
I demoed this build at a presentation at PA Epicor Users Group today, it seemed popular 
âshow me all open invoices from 2021 for the top 10 customers by invoiced amountâ
Dynamic Tooltips are provided when user mouses over or âtouches overâ
6 Likes
I designed some logic and code on the back end for attachments, and I asked AI to select an icon for me and embed it in my textboxâŚ
It legit selected this icon all on itâs own.
Previously, I feared for the future, but I think weâre going to be okay, guysâŚ
6 Likes
There is an Epicor Idea for Application Studio to have these Dynamic Tooltips/Cards/Detached Slideouts.
Enhance Slide Out Panel Functionality to Float/Pop-Up anywhere on page like a tool tip: KIN-I-6230
1 Like
Starting in August I was tasked with building an automation pipeline for AP utilizing open-weights models. In November I finally started using Claude Code, and the following few months wrote several custom slash commands and skills files to turn any AI/automation project idea related to Epicor into a matter of copying and pasting a project plan and having Claude generate a working and tested MVP in a matter of minutes. I also made a skills file for making C# BPMs, making automation a little more accessible at my company (currently I am the only coder).
A few weeks ago, I used said skills files to create a full read/write MCP server connected to Epicor via the Claude Desktop app in two days. Users authenticate with Microsoft Azure and then people can âtalkâ to the Epicor database by querying BAQs, creating BAQs, executing dashboards, send emails, create their own dashboards, and so much more. Using Cowork, my users can have Claude interact with their files and Epicor at the same time, and now AI is spreading like a wildfire over here. Some spirits might be crushed when we switch to FedRAMP High and lose access to Cowork, but we all gotta make sacrifices.
If only I could convince my VP to let me open source my invaluable skills files and my MCP server⌠More on that later.
7 Likes
Certain people here would have a stroke if we did that 
I actually agree our data probably wouldnât be safe if we did that, but I am a bit more skeptical of the value of our data..
1 Like
It would be no different than using Prism, which routes to these same providers in the same ways. AI data processing is in your control, as a user. You donât have to give the AI every bit of data to get a result. For a lot of problems, AI needs context and the goal, creating code to solve the problem against your data. Or providing parameters to a lookup function. Generating the query⌠etc
AI processes the data you give it. The same as the rest of the internet.
â
Cool stuff:
No idea what Iâm actually calling this thing, but itâs a tool that automatically pulls the service lists from Epicor and gives them to an AI agent (and its subagents!). It can explore the endpoints and generate documentation and code from the results. In the current version, it is being expanded to maintain BPMs. Mini-Epicor-IDE!
The next one is a local agent like Claude Cowork or Codex. This one is a major WIP but works surprisingly well and it already improves itself. This is purely for learning harness engineering, as a model is just one piece of an AI engineering process.
All of my tools focus on the following two principles: Ultimate observability and user control. You should get an idea of how the products you purchase are being utilized, otherwise $30x100 adds up with no real gain in value or insight. You should also get total control of the context, documents, and tools that the AI has. On top of that, each of these tools also has an MCP so you can just work with the backend from your favorite frontend (ChatGPT, Claude, Gemini, etc).
4 Likes
Anybody tried goose? As often the case, jackâs early and underrated. It was open-claw-plus-plus before open-claw. Mutli-agent, multi-provider, multi-model, x-plat cli and GUI, extensions are MCP inherently so list is endless, vscode ext experimental, etc, etc.
All OSs
Kinetic OpenAPI-to-MCP (goose extention) anyone?
//goose-extension-config.json
{
"mcpServers": {
"multi-api": {
"command": "openapi-mcp-server",
"env": {
"API_REGISTRY_FILE": "/openapi-registry.json"
}
}
}
}
//openapi-registry.json
{
"version": 1,
"description": "Registry of Epicor ERP Business Object OpenAPI specifications. Each entry represents a single Epicor BO REST service exposed via Epicor REST v2.",
"defaults": {
"auth": {
"type": "bearer",
"header": "Authorization",
"token_env": "EPICOR_API_TOKEN"
},
"base_headers": {
"Accept": "application/json",
"Content-Type": "application/json"
},
"tool_naming": {
"strategy": "service_operation",
"separator": "_"
}
},
"services": [
{
"id": "erp-sales-order",
"name": "Epicor Sales Order BO",
"epicor": {
"module": "Erp",
"bo": "SalesOrder",
"service": "Erp.BO.SalesOrderSvc"
},
"openapi": {
"url": "https://mydomain.epicorsaas.com/Server/api/v2/odata/Erp.BO.SalesOrderSvc/$metadata/openapi"
},
"base_url": "https://mydomain.epicorsaas.com/Server/api/v2/odata",
"tags": ["sales", "order_management"],
"tool_namespace": "salesorder",
"enabled": true,
"rate_limits": {
"per_minute": 120
}
},
{
"id": "erp-customer",
"name": "Epicor Customer BO",
"epicor": {
"module": "Erp",
"bo": "Customer",
"service": "Erp.BO.CustomerSvc"
},
"openapi": {
"url": "https://mydomain.epicorsaas.com/Server/api/v2/odata/Erp.BO.CustomerSvc/$metadata/openapi"
},
"base_url": "https://mydomain.epicorsaas.com/Server/api/v2/odata",
"tags": ["customers", "crm"],
"tool_namespace": "customer",
"enabled": true
},
{
"id": "erp-part",
"name": "Epicor Part BO",
"epicor": {
"module": "Erp",
"bo": "Part",
"service": "Erp.BO.PartSvc"
},
"openapi": {
"url": "https://mydomain.epicorsaas.com/Server/api/v2/odata/Erp.BO.PartSvc/$metadata/openapi"
},
"base_url": "https://mydomain.epicorsaas.com/Server/api/v2/odata",
"tags": ["inventory", "engineering"],
"tool_namespace": "part",
"enabled": true
},
{
"id": "erp-job",
"name": "Epicor Job Entry BO",
"epicor": {
"module": "Erp",
"bo": "JobEntry",
"service": "Erp.BO.JobEntrySvc"
},
"openapi": {
"url": "https://mydomain.epicorsaas.com/Server/api/v2/odata/Erp.BO.JobEntrySvc/$metadata/openapi"
},
"base_url": "https://mydomain.epicorsaas.com/Server/api/v2/odata",
"tags": ["manufacturing", "jobs"],
"tool_namespace": "job",
"enabled": true
},
{
"id": "erp-ud",
"name": "Epicor UD Tables BO",
"epicor": {
"module": "Erp",
"bo": "UD",
"service": "Erp.BO.UD01Svc"
},
"openapi": {
"url": "https://mydomain.epicorsaas.com/Server/api/v2/odata/Erp.BO.UD01Svc/$metadata/openapi"
},
"base_url": "https://mydomain.epicorsaas.com/Server/api/v2/odata",
"tags": ["customization", "user_defined"],
"tool_namespace": "ud",
"enabled": false
}
]
}