Configuration describes the data sources — the framework discovers the rest at runtime. Connectors, screens, menus and dashboards are built in the Settings editors — or in one pass with the Screen Creation Assistant; the React UI then builds a typed grid, a filter row, a modal form and lookups from the columns the query actually returns. No schema duplication, no per-screen code.
How it works
Define a connector in the editors, expose its queries to a screen or a dashboard, and the framework renders everything else — grid, dialog, filters, auth, role pruning, export buttons, audit columns, the assistant and scheduled jobs.
The grammar
Everything the framework does composes from five primitives. Once you’ve read these five, the framework’s surface is essentially mapped.
A connection to a database — a SQLAlchemy URL plus credentials.
“How do I reach this data?”
A named set of queries or REST endpoints on top of a pool.
“What questions do I want to ask this source?”
A grid + edit dialog over a connector’s queries.
“How do I let a person look at this and edit it?”
A layout of KPIs and charts over the same queries.
“How do I summarise this for the headline view?”
The sidebar tree that organises screens and dashboards into an app.
“How do I make this navigable?”
Around the five sit the supporting layers: the dictionary for labels and formats, auth & roles for who-sees-what, Nomaflow for scheduled work, the AI assistant and Python plugins.
Schema-discovered, not schema-declared
Nothing about your data is duplicated in framework configuration. The connector knows the query; the database knows the columns. When a column changes, the UI reflects it on the next reload.
Use cases
CRUD over existing tables, role-gated, with a sensible default UX. From “I have a database” to “users are using it” in about a day.
KPIs and charts over named SQL queries. Shared filter bar, drill-down to grids, PDF export — no separate BI tool for the 80% case.
A modern UI over JD Edwards, SAP or NetSuite — users, security, BIP queue, master data. Gives operators a faster way to live in the ERP.
A UI on top of webhook receivers, scheduled syncs and ETL pipelines. Nomaflow runs the jobs; the framework renders the runs.
A scoped slice of an internal app exposed to outside users via OIDC. Row-level access, audit trail and file uploads, all framework-native.
Approve / reject / route — multi-step forms, status workflows, scheduled escalations. Form conditions handle the per-state UX.
The right tool when most of your screens are list-and-edit on top of a database, with sprinklings of dashboards, jobs and integrations — which describes the vast majority of internal applications.
Built-in AI
Liberty Next ships a Claude assistant with tool use, opened from any page. It runs the read queries of the connectors the signed-in user may use, computes totals, rankings and trends on the Data Explorer models, follows the playbooks and tools an application ships, and knows the page you are on. Ask “which suppliers were paid late last month?” and it picks the query, runs it with your permissions and answers with the figures.
The same tools answer Claude Desktop, claude.ai and Claude Code through the MCP endpoint of the server: Claude acts as a Liberty Next user, with that user’s permissions and data scope. Nomasx-1, Nomajde and Nomaflow get both without any extra development. Without an API key the assistant is not configured, and the MCP endpoint is off until an administrator turns it on — every other feature keeps working.
Permission-scoped
Each tool checks what the screens check for the signed-in user, and the data scope filters the rows. A tool the user cannot use is not listed.
Read-only by default
Adding or changing a record, running a screen action or a job goes through the screen, and needs three locks: the Write tools setting, the ai:write permission and the permission of the screen or the job. In the chat, nothing happens until the user clicks Apply on a card.
Claude over MCP
Connect Claude Desktop, claude.ai or Claude Code by sign-in, with a Claude Desktop extension when the server is not on the internet, or with a personal access token. Changes over MCP sit behind their own switch, off by default.
Sensitive data and usage record
Flag a dictionary item or a column as sensitive to keep it from the assistant, in the chat and over MCP. Monitoring → AI usage lists who asked, the estimated cost, the tools called and every change — never a row returned nor a value written.
Vendor applications
The framework itself is free and open source. A license key (an RS256-signed JWT) unlocks the bundled vendor products that ship on top of it.
Enterprise security, Object Usage Tracking, Oracle & JD Edwards licence compliance and Segregation of Duties — one audit-ready platform.
Master data, security maintenance, transactions and BIP reporting on one searchable grid. SQL + AIS API, no fat-client ping-pong.
Cron jobs, ETL pipelines, LDAP sync and HTTP fan-out with retry, run history and a live log tail — in-process, on the same connectors and permissions.
Under the hood
React 19 + Vite + TypeScript, built once and served static by the backend on the same port. Dark default with a light theme swap, react-i18next EN / FR, @tanstack/react-table for the grid, react-markdown for the assistant, Monaco for in-app editing. You don’t write any of it — the framework renders everything off the configuration.
Python 3.12, FastAPI, SQLAlchemy 2.0 async with asyncpg (PostgreSQL), oracledb (Oracle, thin) and aiosqlite (SQLite), the Anthropic SDK for the AI assistant, authlib for OIDC, Argon2 for hashing, AES-256-GCM field encryption, APScheduler for cron jobs, Socket.IO for live updates. One process, one port, no companion daemons.
FAQ
Liberty Next is an open-source, connector-driven low-code framework published by Nomana-IT. You connect a SQL or API source and build connectors, screens, menus and dashboards in the built-in Settings editors and the Screen Creation Assistant; the React UI then builds the typed grid, filter row, edit dialog, lookups and dashboards from the columns the query actually returns, with no per-screen code. The configuration is stored as plain TOML files, so it can be versioned and promoted between environments.
Yes. The framework is free and open source. Nomasx-1, Nomajde and Nomaflow are applications built with and for it; a license key (an RS256-signed JWT) unlocks the licensed vendor applications.
Liberty Next connects to PostgreSQL, Oracle Database and SQLite through SQLAlchemy async pools, and to REST services through HTTP and API connectors. The schema is discovered at runtime from the live query.
The built-in assistant uses Anthropic Claude with tool use. It runs the read queries of the connectors the signed-in user may use, computes on the Data Explorer models and follows the playbooks an application ships; each tool checks the user’s permissions and data scope. It reads only unless the write tools are turned on and the role holds the ai:write permission — a change then goes through the screen and is applied by the user from a card.
Yes. Liberty Next has an MCP (Model Context Protocol) endpoint that offers the assistant’s tools to an external Claude client. The client connects by sign-in, through a Claude Desktop extension when the server is not on the internet, or with a personal access token, and acts as a Liberty Next user with that user’s permissions and data scope. It reads only unless an administrator allows changes over MCP, and every call is recorded. The applications built on Liberty Next — Nomasx-1, Nomajde, Nomaflow — are reached the same way.
Yes. ERP administration is one of its main use cases: Nomajde (the JD Edwards companion) and Nomasx-1 (security, licences and Segregation of Duties) are both built on Liberty Next, with Nomaflow for scheduled jobs.
One Python process serves the API and the React UI on one port, deployed with systemd, Docker or Kubernetes on your own servers or in your cloud — or hosted by Nomana-IT on Oracle Cloud Infrastructure (OCI) or AWS.
Liberty Next is free and open source. Try the live demo, read the docs, or deploy your own instance.