Multi-Agent System for Production-Grade Database Schema Generation - n8n Workflow

• Database Architects and Developers needing fast, validated schema blueprints.


• Users building AI services that require complex, production-grade logic.


• n8n users seeking advanced examples of multi-agent and flow control in an n8n workflow.


• SaaS providers offering automated technical consulting services.

Developing a robust database schema is complex, requiring expertise in normalization, indexing, and scalability. This advanced n8n workflow solves this by utilizing a coordinated team of four specialized AI agents (Architect, Reviewer, Optimizer, and SQL Generator) to transform natural language requirements into optimized, executable SQL scripts. This particular n8n workflow incorporates critical corelogicflow_control elements, including a Quality Gate and an iterative retry loop (max 3 attempts). If the design fails the rigorous AI review (earning a C, D, or F grade), the n8n node flow loops back, sending detailed feedback to the Architect agent for refinement, guaranteeing high-quality output. This automated system transforms conceptual requirements into actionable database assets using the power of n8n templates.

The process begins with the Form Submission n8n trigger, where the user inputs detailed requirements (database technology, scale, key workflows).

1. Initialization: The workflow sets variables (iterationcount, formdata) to manage the iterative process.


2. Agent 1: Database Architect: This agent uses the provided requirements to generate the initial, structured JSON schema design.


3. Agent 2: Schema Reviewer: The reviewer rigorously checks the design against best practices for normalization, performance, and scalability, producing structured feedback (CRITICAL, HIGH, MEDIUM).


4. Agent 3: Optimizer & Scorer: This agent applies final polish, adds advanced features, and provides a comprehensive score card (A-F grade).


5. Quality Gate: The Is Score A or B? n8n node checks the score. If the score is A or B, the design is approved, and the flow proceeds to step 7.


6. Iterative Retry Loop: If the score is C or below, the Can Retry? n8n node checks if the iteration_count is less than 3. If so, the system loops back to Agent 1, providing the critical feedback for immediate redesign. This powerful n8n workflow ensures continuous improvement.


7. Agent 4: SQL Generator: Once approved (or after max iterations), this agent converts the final JSON schema into a production-ready SQL migration script.


8. Conditional Execution: The If User Is Using Excel (PostgreSQL check) n8n node determines if the script should be executed automatically using the Execute SQL in PostgreSQL n8n node. If execution fails, the flow gracefully continues via the error path to provide the script and error details.


9. Final Response: A final response is returned via the Form node, providing the generated SQL script, the design schema, and the full scorecard/feedback.

To deploy this expert n8n workflow, follow these steps:

1. Import the JSON: Copy the provided JSON data and import it directly into your n8n instance as a new workflow.


2. Anthropic Credentials: Update the credentials for the four main AI Model nodes (Architect Model, Reviewer Model, Optimizer Model, SQL Generator Model). These use the LangChain Anthropic LLM n8n node, requiring an Anthropic API Key.


3. PostgreSQL Credentials (Optional): If you intend to use the automatic deployment feature, you must configure the credentials for the Execute SQL in PostgreSQL n8n node. If you do not have a PostgreSQL database configured, the flow will skip the execution step.


4. Activate the n8n Trigger: Ensure the Form Submission n8n trigger node is active to start collecting requirements via the generated webhook URL.


5. Customization: Review the system messages within the Agent n8n node configurations to adapt the expertise of the AI roles to your specific needs. This powerful n8n templates design is highly flexible.

Form Submission (n8n trigger): The starting n8n node, acting as a user interface to collect complex requirements (database type, scale, workflows) that drive the entire agent system.
Initialize Variables (Set n8n node): Essential for managing the state of the iterative loop, initializing the iteration_count and storing initial form data.
Architect Model / Reviewer Model / Optimizer Model / SQL Generator Model (Langchain Chat Anthropic n8n nodes): These nodes connect to the Claude 4.5 LLM service, providing the intelligence for the four specialized agents. Temperatures are specifically tuned for each role (e.g., Reviewer is cold/strict, Optimizer is balanced).
Agent 1: Database Architect (Langchain Agent n8n node): Uses a detailed system message to enforce the role of a senior database architect, outputting a highly structured JSON schema.
Agent 2: Schema Reviewer (Langchain Agent n8n node): Focuses solely on critique, performance, and best practices, generating structured feedback necessary for the retry loop.
Agent 3: Optimizer & Scorer (Langchain Agent n8n node): Responsible for the critical scoring mechanism that controls the flow via the quality gate.
Is Score A or B? (If n8n node): The core flow control n8n node, checking the scoring agent's output using a regex to determine if the workflow should proceed to generation or enter the refinement loop.
Can Retry? (If n8n node): Manages the maximum iteration limit (3 attempts), preventing infinite loops and controlling the cost of this complex n8n workflow.
Agent 4: Generate SQL Script (Langchain Agent n8n node): This precision n8n node converts the final, validated design into raw SQL code, crucial for immediate deployment.
Execute SQL in PostgreSQL (Postgres n8n node): Optionally attempts to run the generated SQL script directly against a configured database, handling potential deployment errors gracefully and continuing the n8n workflow.