We build AI into our own products and into our clients’ systems: assistants that query JD Edwards, Oracle and other databases through governed connectors, Claude skills for daily work, and AI-assisted development. Thirty years of ERP experience decide what the model may see — and what it may never touch.
Principles
The assistant sees and runs only what the signed-in user is allowed to see and run. A tool the user has no permission for is never offered to the model.
Write queries are excluded unless explicitly exposed and granted per role. An unintended read is recoverable; an unintended write in an ERP is not.
Every tool call is shown with its parameters. Answers cite the documentation page or the data they come from, so users can check them.
No API key, no AI call. Each installation is dedicated to one customer — on your servers, in your cloud, or hosted by us on OCI or AWS. Only the question and the results of the tools it calls go to the model.
Software publisher
A Claude assistant with tool use at /chat. Every SQL query and HTTP endpoint of every connector the user can run becomes a tool, with a JSON schema generated from the query parameters and a description taken from the connector and the dictionary.
A chat panel that knows the page and the invoice you have open. It answers from the NomaUBL documentation, from the generated documentation of your instance and from your data — strictly read-only.
Tool use, step by step
The user asks in plain language: “Which suppliers were paid late last month?”
The platform builds the list of tools from the connectors this user may run — nothing else.
Claude chooses the connector query and its parameters.
The query runs with the user’s own permissions, on your database or API.
The model reads the result, may refine with another call, and answers with the figures.
AI consulting
Assistants that answer questions on JD Edwards, Oracle Database or any SQL/REST source — choosing what to expose, writing tool descriptions the model understands, and fencing off writes.
We work daily in Claude Desktop with our own skills and connectors for tasks, email and attachments, and we build the same kind of tooling for your teams and your procedures.
We design and write our products ourselves, working with Claude and Claude Code in VS Code as assistants — architecture, review and decisions stay with the developer. We help development and ERP teams adopt the same way of working: project instructions, skills, reviews and guardrails.
Notion Workspace already exposes a workspace as Model Context Protocol tools for AI agents. Next: MCP servers for JD Edwards and Oracle, so that any MCP client — Claude Desktop, Claude Code, IDEs — can query ERP and database data with the same permission model as our products.
Yes, if it goes through governed connectors rather than direct database access. In Liberty Next, each connector query becomes a tool the assistant may call only when the signed-in user has the corresponding permission, and write queries are excluded unless they are explicitly exposed and granted per role.
The assistants in Liberty Next and NomaUBL use Anthropic Claude models through the Anthropic API with tool use. The administrator chooses the model; without an API key, no AI call is made and every other feature keeps working.
Each installation is single-tenant — on your own servers or cloud, or hosted by Nomana-IT on OCI or AWS. When the assistant is enabled, the user’s question and the results of the tools it calls are sent to the model provider to produce the answer; the database itself is never exposed. You decide which connectors and queries are available to the assistant.
Notion Workspace already exposes a workspace as MCP (Model Context Protocol) tools for AI agents. MCP connectors for JD Edwards, Oracle and our ERP products are in development; they will apply the same rules as the built-in assistants: permission-scoped and read-only by default.
Yes. We write skills and project instructions for Claude and Claude Code — for example ERP procedures, documentation conventions or development standards — based on the tooling we use every day.
No. JD Edwards is our core expertise, but the connector approach works with Oracle Database, PostgreSQL and REST APIs, and NomaUBL’s assistant works with any ERP feeding it invoices.
Tell us which question your teams ask every week. We will tell you what an assistant needs to answer it safely.