NVIDIA Auto Ontology on Railway
NVIDIA's open-source semantic layer that turns database questions into SQL, with MCP, on Railway.
Auto Ontology is an open-source project from NVIDIA that builds an ontology, a map of tables, columns and business terms, over relational databases. It then answers natural-language questions by generating SQL, and shows the SQL it ran. A business glossary maps terms such as active customer onto real columns.
It is aimed at teams that want AI agents to query company data in a governed, read-only way. Agents connect through an MCP server, and scripts can use an authenticated chat completions endpoint. Supported sources include Postgres, Snowflake, Databricks, MySQL, ClickHouse and Trino.
Auto Ontology is licensed under Apache-2.0. The Railway template is a community template and is not affiliated with NVIDIA.
What NVIDIA Auto Ontology can do
- Ontology built over relational databases
- Natural-language questions answered as SQL, shown to the user
- MCP server for Cursor, Claude and other MCP clients
- Business glossary mapped to real columns
- Web UI with login and API tokens
- Scheduled semantic compilation
What people use it for
- Giving agents governed, read-only access to company data over MCP
- Natural-language analytics over Postgres and other warehouses
- A shared business glossary for data terms
- Scheduled jobs that ask questions through the API
Videos
Running NVIDIA Auto Ontology on Railway
The template deploys Auto Ontology 1.1.0 as six services: a frontend with login and an authenticated API gateway, an MCP server, private backend and ingestion services, Postgres 18 with pgvector, and a read-only Chinook sample database. Only the frontend and MCP server have public domains. You paste an NVIDIA API key at deploy, or point the model settings at another OpenAI-compatible provider. Model calls are billed by your provider; the six services run continuously.
NVIDIA Auto Ontology templates on Railway
Tracklayer templates are listed first. Other NVIDIA Auto Ontology templates follow, ordered by total deployments on Railway. How this works.
Deployment numbers as of 2026-11-20.
Questions about NVIDIA Auto Ontology
What do I need to deploy?
An NVIDIA API key for DEFAULT_MODELS_API_KEY, created at build.nvidia.com, or another OpenAI-compatible provider configured in the model settings.
How do I log in?
Open the frontend URL and sign in with AUTO_ONTOLOGY_ADMIN_EMAIL and the generated AUTO_ONTOLOGY_ADMIN_PASSWORD from the frontend variables. Sign-up stays closed.
How are database credentials stored?
Without Vault in this template, connection credentials are stored as plaintext JSON in the Postgres catalog. Use read-only database users, keep that database private and treat backups as secrets.
Can I use NVIDIA's hosted endpoints in production?
The template notes say free access on build.nvidia.com is governed by NVIDIA's API Trial Terms, which limit use to internal testing and evaluation without a paid subscription. Check the terms or use another provider. This is not legal advice.
What drives the cost?
Six always-on services, plus model calls. Semantic compilation runs every 24 hours over connected databases and calls the model for new tables, so large warehouses can mean many calls.
Similar apps
Tracklayer maintains the templates marked as such; NVIDIA Auto Ontology is a trademark of its owners and this site is not affiliated with them.