From business requirements  to production-ready data architecture 

offdataai takes you from business requirements to production-ready data architecture and AI-ready enterprise context, in six stages: Design, Generate, Define, Connect, Govern and Activate. You can start from a plain-English interview, one of 174 templates, the drag-and-drop Model Builder, or the Database Modeler (import an existing database). It generates ERDs, platform-native DDL, complete dbt projects, seed data and documentation for Snowflake, BigQuery, Databricks, Redshift, Postgres, Synapse, Microsoft Fabric and DuckDB, plus a business ontology your AI agents can use.

Data modeling (or data modelling) tools fall into four groups: desktop modelers such as erwin Data Modeler, ER/Studio and PowerDesigner; diagram tools such as dbdiagram, Lucidchart and draw.io; code-first tools such as dbt and SQLMesh; and AI-native platforms such as offdataai. offdataai designs the model from a conversation, a template or an existing database and generates the ERD, platform-native DDL, a dbt project and a semantic layer from it, so it covers modeling, code and meaning in one place.

Yes. The offdataai ontology is a semantic layer: business concepts, relations and governed metrics (for example Revenue = sum of order amount) bound to your tables. Questions are written in those business terms and compiled to SQL for each source by walking the relations, never by letting a model write SQL freehand, and results from different systems can be combined. On Databricks it can also publish a Unity Catalog metric view. See https://offdataai.com/semantic-layer/.

A context layer gives AI agents the business meaning they need to use your data correctly: what tables mean, how they join, how metrics are defined and what each agent may see. offdataai provides it through a built-in MCP server, so Claude and other MCP clients can search concepts, get query context and join paths, and ask questions in plain English under your access rules and PII masking. See https://offdataai.com/context-layer/.

Yes. The offdataai data catalog lists every system, table and column in your connected sources (PostgreSQL, MySQL, Snowflake, BigQuery, Redshift, Databricks Unity Catalog and HubSpot), shows which ontology concepts use each one, and finds where the same data lives twice. It reads metadata only, never rows. See https://offdataai.com/data-catalog/.

offdataai is a purpose-built pipeline: an interview agent gathers grain, cardinality, and SCD requirements; a synthesis agent compiles your answers into a validated Intermediate Representation (IR); validators check referential integrity and type coercions before any code is generated; and platform-specific generators emit DDL, dbt scaffolds, and seed CSVs. ChatGPT can sketch a schema, but offdataai ships warehouse-ready artifacts that are tested and consistent across paradigms (Kimball, Data Vault 2.0, 3NF).

The IR is a validated JSON document that captures your entire data model — entities, attributes, relationships, grain, and platform hints. It is the single source of truth that every generator consumes. You can inspect, edit, and version it like any other artifact.

offdataai uses OpenAI's GPT-4o, served through Microsoft Azure OpenAI, for the interview, model synthesis, the Refine assistant and ontology enrichment, and OpenAI's text-embedding-3-small (also on Azure) to search the documents in your knowledge bases.

Yes. Ask the Refine assistant in plain English (for example "add a promotions dimension"), or edit the model directly in the app. Changes flow through the validators and every downstream asset regenerates automatically.

offdataai ships native DDL and dbt projects for Snowflake, Google BigQuery, Databricks (Delta), Amazon Redshift, PostgreSQL, Microsoft Synapse, Microsoft Fabric and DuckDB. Each generator respects platform-specific syntax, clustering, partitioning, and data type rules.

Yes. offdataai supports three modeling paradigms out of the box: Kimball dimensional modeling (star and snowflake schemas), Data Vault 2.0 (hubs, links, and satellites), and Third Normal Form (3NF). You choose the paradigm — the synthesis agent shapes the IR accordingly.

A complete dbt project: dbt_project.yml, source YAML, staging models that read the bundled seeds, mart models (incremental facts, snapshots for SCD2 dimensions), schema tests and seed CSVs. On Pro and above the seeds carry placeholder rows that keep foreign keys valid. The project, seeds included, passes `dbt build` on the first run.

Your descriptions, interview answers and the schemas and documents you attach are sent to Microsoft Azure OpenAI for processing; Azure OpenAI does not use API data to train models. The Database Modeler reads metadata only (tables, columns, keys) from connected databases, never rows. Rows are read only when you or an agent query your data, for example by asking an ontology a question (for agents, under its access rules and masking). For a plain-English question, up to 25 rows of the result go to the model to phrase a one-sentence answer. Your projects and uploaded files are stored in offdataai's PostgreSQL database and Cloudflare R2 object storage.

Yes. The Database Modeler connects to PostgreSQL, MySQL, Snowflake, BigQuery, Amazon Redshift and Databricks (Unity Catalog). It reads only the catalog (tables, columns, keys, indexes and comments, never your rows), suggests undeclared relationships for you to accept, classifies tables as facts, dimensions and more, and builds a project from it: mirrored 1:1, or remodeled as a star schema or Data Vault. On Databricks it can also learn joins from query history, apply the keys, descriptions and PII tags you approved back to Unity Catalog, and publish a metric view so Genie stops double-counting.

Yes. Connect GitHub, GitLab or Bitbucket and push the generated DDL, dbt project, seed SQL, documentation and the canonical IR to a branch in one click, optionally opening a pull request, so the output is reviewed and version-controlled in your real workflow.

Yes. Every column carries a classification and an optional masking strategy (hash, partial or full). offdataai inventories PII, reports masking coverage and compliance tags (e.g. HIPAA, GDPR), and generates masking SQL for your platform: Snowflake masking policies, Databricks column masks, and Dynamic Data Masking on Synapse and Fabric. Partial masking keeps the first two characters for every role. On Postgres, Redshift and BigQuery, which have no native column masking, it records the intent as column comments and suggests a secure view.

Yes. Every edit creates a new IR version. offdataai diffs any two versions and generates the exact ALTER TABLE / CREATE TABLE migration SQL for your target platform, so you can evolve a live warehouse safely instead of hand-writing DDL.

Yes. Invite teammates as viewers or editors, share a read-only link with stakeholders who don't have an account, and discuss the model in comments. @mention someone and they get an email, plus a Slack message if they have connected Slack. Each person chooses their notifications: mentions are on by default, a note on every comment on your own projects is opt-in. On Team, people you add to the organization get an invitation email.

Yes. Create an API key and use the public REST API to create projects, fetch the generated IR and pull platform-native DDL, which is ideal for CI pipelines and internal tooling. Keys with the ontology:query scope can also ask an ontology questions, including in plain English (POST /api/v1/ontologies/{id}/ask). The same keys connect AI agents to the built-in MCP server.

Yes, 174 of them, included on every plan. The template gallery offers complete, validated domain models across 18+ industries (banking, healthcare, insurance, SaaS, retail, SAP, Salesforce, telecom and more) in Kimball, Data Vault 2.0 and 3NF. Opening one creates a project with the ERD, DDL and dbt project already generated, with no interview needed.

Synthesis runs in the background. offdataai can notify you by email and/or Slack the moment your model is ready — or if a run needs attention — so you don't have to watch a progress bar.

Plans start at $1/month. Explorer ($1/month) includes 3 projects, all 174 templates, the Model Builder canvas, the ERD, SQL for every platform, dbt and docs, with no AI credits. AI features use credits that come with the higher plans: Starter ($29/month, 60 credits) adds the AI interview, the Refine assistant, uploading your schemas, dictionaries and sample files, ontologies and the data catalog, and one knowledge base for your documents (10 documents); Pro ($99/month, 250 credits) adds live connections (the Database Modeler, questions on your data, HubSpot), governance and lineage, seed data, migrations, AI ontology enrichment, the public API and MCP server, comments with @mentions and share links, publishing dbt to GitHub or GitLab, and 5 knowledge bases (100 documents); Team ($399/month, 1,200 credits) adds an organization workspace with members, roles and shared credits, unlimited projects and unlimited knowledge bases (1,000 documents). Enterprise is custom.

An ontology is the business meaning layer on top of your data model: concepts (Customer, Order), relations between them, and metrics (Revenue = sum of order amount). offdataai drafts it from your model, puts every AI suggestion in a review queue, and lets you publish it with resolvable IRIs and export it as OWL Turtle, JSON-LD, RDF/XML or JSON. You can also start a blank ontology and import a project, a database model or a drawn model into it later, or create one from an existing OWL file (Turtle, RDF/XML, JSON-LD or N-Triples).

offdataai includes an MCP (Model Context Protocol) server. Connect Claude or any MCP client with an API key and it can search concepts, get the query context for a question, find join paths, ask the ontology a question in plain English (ask_ontology, one credit per question), run read-only queries against your connections and propose new concepts for review. Questions asked through the ontology follow its access rules and row filters, and personal data stays masked unless the key has the pii:read scope.

The Model Builder is a drag-and-drop canvas for designing entities and relationships by hand, with a full-screen mode and one-click auto-arrange. It offers Kimball, Data Vault 2.0 and 3NF building blocks (facts, dimensions, hubs, links, satellites, bridges and more), saves automatically, exports to JSON or PNG, and can be the source of an ontology. A JSON export imports back as a new model, with the same tables, columns and layout. It needs no AI credits.