Modern database development often feels like a fragmented exercise in context switching. A developer typically starts by sketching a structure in ERwin, jumps over to DBeaver to refine queries, and then relies on a separate generator to produce the boilerplate code. This three-window shuffle requires manual synchronization of schema changes across different tools, creating a friction-filled workflow where a single missed update can lead to catastrophic data accidents in a production environment.
The Unified Database Workbench
NeoSQL addresses this fragmentation by consolidating ERD modeling, a SQL editor, and code generation into a single integrated workbench. The platform provides broad compatibility, supporting MySQL, MariaDB, PostgreSQL, Oracle, SQL Server, SQLite, and H2. At the modeling stage, it offers Forward and Reverse Engineering capabilities alongside automatic foreign key recognition. When a schema changes, the tool tracks the modifications and automatically generates the corresponding ALTER DDL to ensure the physical database remains in sync with the visual model.
The SQL editor is designed for precision and safety, featuring autocomplete and execution plan visualization. To mitigate the risk of accidental data loss, NeoSQL sets Manual Commit as the default behavior, forcing developers to consciously confirm changes before they are finalized. Table editing is handled through an inline, spreadsheet-like interface where any modification is instantly converted into an ALTER DDL statement. This is complemented by a Schema Diff feature that allows developers to compare differences between development, staging, and production databases, enabling selective application of changes.
For the application layer, NeoSQL utilizes the Velocity template engine to generate code across multiple languages and frameworks, including Java, Kotlin, TypeScript, and Python, as well as Spring, JPA, MyBatis, and Vue. Security is integrated into the core architecture using the AES-256-GCM encryption standard to protect 17 sensitive fields, including URLs, hosts, ports, and passwords. These encryption keys are managed externally via AWS KMS to ensure a strict separation of concerns. Users can access the tool through three distinct modes: a browser-based Web App, a Desktop Online version for bypassing corporate firewalls, and a Desktop Offline version currently under development for local-only storage.
Access to these features is tiered. New users can experience the Pro plan for free for 14 days. Following the trial, the Free plan remains available indefinitely, supporting one project, one database connection, and up to 20 ERD tables. For larger scale requirements, the platform offers Basic, Pro, and Team plans.
Redefining the AI-Database Interface
While many AI-powered database tools focus on the ability to execute DDL directly from a natural language prompt, NeoSQL introduces a fundamental shift in control. Instead of allowing an AI to modify a database autonomously, NeoSQL implements a human-in-the-loop architecture. In this workflow, the AI proposes changes to the ERD first. The human developer then reviews the visual representation of those changes and must explicitly approve them before they are applied to the schema. This creates a critical verification filter, ensuring that AI-generated designs are vetted by a human expert before they touch the data layer.
This philosophy of caution extends to credential management. Rather than embedding database passwords directly into AI configuration files, NeoSQL uses an isolation method where the AI references a project ID. This architectural choice prevents passwords from being leaked into the AI context window. Furthermore, any DML executed by the AI is prevented from auto-committing. The developer must review the results in the SQL editor and manually decide whether to commit or rollback the transaction.
To manage these complex interactions at scale, NeoSQL employs Role-Based Access Control (RBAC). By dividing users into Owner, Manager, DBA, and Member roles, organizations can restrict sensitive access. For example, a user can be granted permission to execute queries without ever seeing the underlying connection credentials. The platform also enforces a DDL approval workflow; if a user without sufficient permissions proposes a change in the ERD, the request is routed to an approver who must review and authorize the modification. This structurally prevents unauthorized schema changes in production environments.
To bridge the gap between the workbench and the broader AI ecosystem, NeoSQL includes a built-in MCP (Model Context Protocol) server. This allows the tool to communicate securely with external AI agents such as Claude, Cursor, and VS Code, enabling these tools to interact with ERDs and SQL scripts without compromising security. Looking forward, the roadmap includes the integration of a native AI assistant dedicated to query assistance and schema reviews, as well as expanding support to NoSQL databases including MongoDB, Redis, and DynamoDB.
Users can access the platform at https://neosql.unvus.com and report issues via the official repository at https://github.com/unvus/neosql/issues.
The evolution of database management is shifting away from simple query execution toward the design of rigorous workflow controls that balance AI efficiency with human oversight.
