{"slug": "nvidia-open-source-auto-ontology", "title": "Nvidia – open-source auto-ontology", "summary": "NVIDIA released Auto Ontology, an open-source library under the Apache License 2.0 that builds a context layer from enterprise metadata so AI agents can discover, ground, validate and expose business definitions across systems via MCP, CLI and API. The library stores its semantic catalog in Postgres with pgvector, uses NVIDIA NIM endpoints for inference with defaults of nvidia/nemotron-3-nano-30b-a3b for the LLM, nvidia/nemotron-3-embed-1b for embeddings and nvidia/llama-nemotron-rerank-vl-1b-v2 for reranking, and ships a Helm chart for Kubernetes or Docker Compose for local deployment. NVIDIA said the project is not currently accepting external contributions, and it points developers to a separate ontology-sql-eval repository for running text-to-SQL benchmarks such as BIRD.", "body_md": "Agents can retrieve documents and query databases, but reliability gates agentic analytics: a revenue metric answered faster is worth nothing if the number is wrong. Agents need a context layer that tells them where data lives, what it means, whether to use it, how to use it, where it came from, and what is similar. NVIDIA Auto Ontology is an open library that builds that layer from enterprise metadata, reconciles definitions across systems, and exposes it to any agent harness through MCP, CLI, and API. It complements the semantic models customers already own.\n\nIt enables developers to:\n\n- Discover: generate enterprise ontologies from metadata; connect existing ontologies and semantic models; build AI-ready knowledge graphs.\n- Ground: give agents trusted business meaning for metrics, entities, and relationships; reconcile inconsistent definitions across systems.\n- Validate: check generated queries against certified definitions and lineage; return provenance with every answer.\n- Expose: serve context to any agent harness through MCP, CLI, and API. Learn: fold certified corrections and usage back into the ontology, so shared context improves with every question.\n\nTo run a text-to-SQL benchmark such as BIRD, refer to the\n[ontology SQL evaluation repository](https://github.com/NVIDIA/ontology-sql-eval/).\n\n**Licensing & contributions.** Auto Ontology is distributed under the\n[Apache License 2.0](https://github.com/NVIDIA/auto-ontology/blob/main/LICENSE). Third-party open-source components\nbundled, linked, or otherwise used by this project are listed in\n[`THIRD_PARTY_NOTICES.md`](https://github.com/NVIDIA/auto-ontology/blob/main/THIRD_PARTY_NOTICES.md). **This project\nis currently not accepting external contributions.**\n\n- **Natural-language querying** of structured data — questions are translated to\nSQL and executed against connected databases, powered by NVIDIA NIM.\n- **Semantic catalog / ontology layer** over relational sources, stored in\nPostgres alongside the catalog it describes.\n- **Authentication & RBAC** —using[Better Auth](https://github.com/better-auth/better-auth) .\n- **Web UI** for chat, analysis, data catalog, analytics, and settings.\n- **Flexible deployment** — Helm chart for Kubernetes or Docker Compose for a\nlocal stack.\n\n| Component | Technology | Role | Default port | \n|---|---|---|---|\n| Frontend | Next.js 16 (React 19, TypeScript, Tailwind CSS 4), Better Auth, Prisma | Web UI, authentication, API gateway | 3000 | \n| Backend | FastAPI (Python 3.12+), LangChain | Chat / NL-to-SQL, catalog, datasource APIs | 3001 | \n| Ingestion worker | FastAPI (Python 3.12+) | Ingests tabular data and writes embeddings | 3002 | \n| Postgres + pgvector | Relational Database | App metadata, the catalog and ontology, vector store | 5432 | \n| HashiCorp Vault | Optional | Secure storage of connection credentials | — | \n\nAuto Ontology uses NVIDIA NIM endpoints for inference — either the hosted\n[NVIDIA API catalog](https://build.nvidia.com) (`integrate.api.nvidia.com`) or\nself-hosted NIMs. The defaults in `.env.example` are:\n\n- **LLM:**`nvidia/nemotron-3-nano-30b-a3b`\n- **Embeddings:**`nvidia/nemotron-3-embed-1b`\n- **Reranking:**`nvidia/llama-nemotron-rerank-vl-1b-v2`\n\nThe endpoints and models are configured via the `DEFAULT_MODELS_ENDPOINT`, `DEFAULT_MODELS_MODEL`,\n`EMBED_ENDPOINT`, and `EMBED_MODEL` environment variables and require a\n`DEFAULT_MODELS_API_KEY`.\n\nEach model role is configured by a **triplet** of environment variables —\n`<PREFIX>_ENDPOINT` (URL), `<PREFIX>_API_KEY`, and `<PREFIX>_MODEL`:\n\n| Prefix | Role | \n|---|---|\n| `DEFAULT_MODELS` | Shared default; every field below falls back to it. | \n| `REASONING` | Main chat / NL-to-SQL model. | \n| `NON_REASONING` | Lighter model used for entity extraction. | \n| `EMBED` | Text-embedding model (must match between ingest and query). | \n| `RERANK` | Reranker used by the retrieval flow. | \n\nYou can set a full triplet, only some of its fields, or none at all — any field\nleft unset falls back to the matching `DEFAULT_MODELS_<FIELD>`. So the simplest\nsetup is to fill in `DEFAULT_MODELS_*` and override per-triplet fields only where\nthey differ (e.g. `EMBED_MODEL`, `NON_REASONING_MODEL`). `*_API_KEY` also falls\nback to the legacy `NVIDIA_API_KEY` for backward compatibility.\n\n- **Docker** with**Docker Compose v2** (e.g. Docker Desktop on macOS/Windows,\nDocker Engine on Linux) for the local stack. For Kubernetes deployments see[`DEPLOYMENT.md`](https://github.com/NVIDIA/auto-ontology/blob/main/DEPLOYMENT.md) .\n- An **NVIDIA API key** (`nvapi-...` ) for NVIDIA NIM (chat and ingestion) from\nthe NVIDIA API catalog at[https://build.nvidia.com](https://build.nvidia.com) .\n- Connection details for the source database(s) you want to query (Databricks, Postgres, Snowflake, or DuckDB).\n\n**Kubernetes**\n\nTo deploy Auto Ontology on a Kubernetes cluster, see [`DEPLOYMENT.md`](https://github.com/NVIDIA/auto-ontology/blob/main/DEPLOYMENT.md).\n\n**Local (Docker Compose)**\n\n1. \nClone the repository: \n\n```\ngit clone <repo-url> auto_ontology && cd auto_ontology\n```\n\n2. \nCreate your environment file (.env) from the template and fill in the values (Postgres credentials, `DEFAULT_MODELS_API_KEY` ,`CONNECTION_STRINGS` , etc.).\nSee[`.env.example`](https://github.com/NVIDIA/auto-ontology/blob/main/.env.example) for the full list of variables:\n\n```\ncp .env.example .env\n# edit .env\n```\n\n Note: \n  - `AUTH_SECRET` and`APP_URL` are**required** — without them every page\nfails with a Better Auth \"default secret\" error.\n  - `POSTGRES_PORT` only changes the**host-side** port mapping (set it if\n5432 is already taken on your machine); containers always reach Postgres\non the internal network port.\n3. \nBuild the images and start the stack: \n\n```\ndocker compose up -d --build\n```\n\n This builds the backend ( `auto_ontology` ) and frontend (`auto-ontology-frontend` ) images, brings\nup Postgres and pgAdmin, runs the one-shot`frontend-migrate` job to\nsync the database schema, and starts the app (backend, ingestion service,\nand frontend).\n4. \nOpen the UI at [http://localhost:3000](http://localhost:3000) (the backend API is on`:3001` ,\npgAdmin on`:5050` ).**Troubleshooting:** if the UI loads without the left navigation panel\n(or otherwise looks broken after a restart), your browser is holding a\nstale session cookie — this happens when`AUTH_SECRET` changes or the\ndatabase is reset. Clear the site data for`localhost:3000` (or delete\nthe`better-auth.session_token` cookie) and sign in again.\n\nAuto Ontology resolves the source databases it connects to from two sources:\n\n1. \n**Added through the UI** — connections created from within the app. With this\noption the connection credentials are**stored in plaintext in Postgres unless\nVault is configured** :\n  - **Without Vault:** the full connection object (including the password) is\nJSON-serialized and stored, unencrypted, on the catalog's database row.\n  - **With Vault:** the credentials are written to Vault, keyed by the database\nname, and the catalog row stores no credentials (the database name is the\nlookup key — there is no separate reference field).\n Vault is enabled only when all of `VAULT_ADDR` ,`VAULT_NAMESPACE` ,`VAULT_ROLE_ID` , and`VAULT_SECRET_ID` are set (`VAULT_AUTH_MOUNT` and`VAULT_KV_MOUNT` are optional overrides). It is off by default; a partial\nconfiguration is ignored with a warning and falls back to plaintext storage.\n2. \n**`CONNECTION_STRINGS` environment variable** — a comma-separated list of\nconnection strings supplied at deploy time (e.g. the Helm`--set connectionStrings=<CONNECTION-STRINGS>` flag). This is a fallback: it\nis used only when there are no UI-added connections.\n\nAuto Ontology Supports SSO for authentication. The Redirect URI should be configured in the IdP as: APP_URL/api/auth/sso/callback\n\nA browser signs in and gets a session cookie, which a script cannot obtain. For\nscripts, notebooks, and scheduled jobs, mint an **API token** instead:\n\n1. Open the user menu (top right) → **API Tokens** →**New token** .\n2. Name it, optionally pick an expiry (default: never), and **copy the token** .\nAuto Ontology stores only a SHA-256 hash of it, so it is shown exactly once. Losing it\nmeans minting a new one.\n\nSend it as `x-api-key` on any `/api/...` call — `Authorization: Bearer <token>`\nworks too:\n\n```\ncurl -H \"x-api-key: $AUTO_ONTOLOGY_API_TOKEN\" https://auto_ontology.example.com/api/terms\npython\nimport os\nimport requests\n\nsession = requests.Session()\nsession.headers[\"x-api-key\"] = os.environ[\"AUTO_ONTOLOGY_API_TOKEN\"]\n\nterms = session.get(\"https://auto_ontology.example.com/api/terms\").json()\n\nanswer = session.post(\n    \"https://auto_ontology.example.com/api/chat/completions\",\n    json={\"question\": \"How many orders shipped last week?\"},\n).json()\n```\n\nThings worth knowing:\n\n- **A token acts as its owner.** It carries no permissions of its own — every\ncall is authorized against the owner's role, exactly as it would be in the\nbrowser. A viewer's token cannot do admin things.\n- **The whole API surface accepts it.** Authentication is resolved in one place\n(`frontend/auth/resolve-user.ts` ) for every route, so any endpoint in[`docs/openapi/auto-ontology-api.json`](https://github.com/NVIDIA/auto-ontology/blob/main/docs/openapi/auto-ontology-api.json) that is not`withPublic` works with a token.\n- **Revocation is immediate.** Delete the token in the UI, or delete/ban the\nowning user, and the next request with it gets a 401.\n- **Tokens cannot manage tokens.** Creating and revoking requires a signed-in\nsession, so a leaked token cannot issue itself successors.\n\nPublic clients call the authenticated Next.js gateway at\n`POST /api/chat/completions`. The response is an SSE stream containing `step`,\n`result`, or `error` events, followed by `[DONE]`.\n\n`conversation_id` is optional:\n\n- Omit it for a stateless, one-shot request.\n- Supply a client-generated UUID to create or continue a conversation.\n- Reuse the same UUID for every turn in the thread. Wait for `[DONE]` before\nsending the next turn; overlapping requests for one conversation return`409 Conversation in progress` .\n\n```\n{\n  \"question\": \"What about last month?\",\n  \"conversation_id\": \"d61d8aa3-d496-4fa7-97ce-4f831c162e7f\"\n}\n```\n\nFor conversational requests, FastAPI loads up to the five most recent completed turns (bounded to 12,000 characters) and rewrites contextual input such as “What about August?” into a standalone question before running the existing text-to-SQL graph. Independent questions are passed through unchanged. Internal agent thoughts, chart payloads, errors, and raw SQL result sets are not reused as context.\n\nConversation history is scoped to the authenticated Auto Ontology user. Browser and\nservice callers should use the Next.js gateway, which resolves the user and\nforwards the internal `x-auto-ontology-user-id` header to FastAPI. The FastAPI port is a\ntrusted internal service boundary: it must not be exposed publicly because that\nheader is not independently verified by FastAPI. A direct FastAPI request\nwithout `conversation_id` remains stateless; a direct request with\n`conversation_id` requires the trusted identity header.\n\nThe web UI manages the UUID automatically. An API-created conversation can be\nopened in the UI at `/chat?focus=<conversation_id>` when it belongs to the\nsigned-in user.\n\nCoding agents: start at [`AGENTS.md`](https://github.com/NVIDIA/auto-ontology/blob/main/AGENTS.md). Skills live in\n[`skills/`](https://github.com/NVIDIA/auto-ontology/blob/main/skills) (` nvidia-ontology-setup`, `nvidia-ontology-management`, `nvidia-ontology-query`).\nMaintainers keep using [`CLAUDE.md`](https://github.com/NVIDIA/auto-ontology/blob/main/CLAUDE.md).\n\nAgent harnesses such as Cursor and Claude Desktop can query your data through\nAuto Ontology's [MCP](https://modelcontextprotocol.io) server. See\n[`mcp/`](https://github.com/NVIDIA/auto-ontology/blob/main/mcp) to run it and [` docs/mcp.md`](https://github.com/NVIDIA/auto-ontology/blob/main/docs/mcp.md) for the details.\n\nAuto Ontology is licensed under the [Apache License, Version 2.0](https://github.com/NVIDIA/auto-ontology/blob/main/LICENSE).\nSPDX identifier: `Apache-2.0`.\n\nEach NVIDIA-authored source file in this repository carries an SPDX header of the form:\n\n```\nSPDX-FileCopyrightText: Copyright (c) 2024-2026, NVIDIA CORPORATION & AFFILIATES.\nAll rights reserved.\nSPDX-License-Identifier: Apache-2.0\n```\n\nThird-party open-source components used by Auto Ontology are enumerated, with their\nupstream licenses and project URLs, in\n[`THIRD_PARTY_NOTICES.md`](https://github.com/NVIDIA/auto-ontology/blob/main/THIRD_PARTY_NOTICES.md). Components added by the\ncontainer images are listed separately in\n[`CONTAINER_THIRD_PARTY_NOTICES.md`](https://github.com/NVIDIA/auto-ontology/blob/main/CONTAINER_THIRD_PARTY_NOTICES.md).\n\nNOTICE AND DISCLAIMER: This software automatically retrieves, accesses or interacts with external materials. Those retrieved materials are not distributed with this software and are governed solely by separate terms, conditions and licenses. You are solely responsible for finding, reviewing and complying with all applicable terms, conditions, and licenses, and for verifying the security, integrity and suitability of any retrieved materials for your specific use case. This software is provided \"AS IS\", without warranty of any kind. The author makes no representations or warranties regarding any retrieved materials, and assumes no liability for any losses, damages, liabilities or legal consequences from your use or inability to use this software or any retrieved materials. Use this software and the retrieved materials at your own risk.\n\nExternal materials Auto Ontology interacts with at run time:\n\n| Material | When | Terms | \n|---|---|---|\n| NVIDIA-hosted model APIs ( [build.nvidia.com](https://build.nvidia.com) ,`integrate.api.nvidia.com` ) for LLM, embedding, and reranking | Chat, ingestion, and semantic compilation, with the default model configuration | Terms of the NVIDIA API catalog and of each model | \n| Any other OpenAI-compatible model endpoint you configure ( `*_ENDPOINT` /`*_MODEL` ) | Same | Terms of that provider and model | \n| KumoRFM predictive-query service | Only when `KUMO_RFM_API_KEY` is set | Kumo's terms | \n| Your source databases | Ingestion, profiling, and query execution | Your agreements for those systems | \n\n**This project is currently not accepting contributions.** Issues, pull\nrequests, and patches submitted from outside the Auto Ontology maintainer team will\nnot be reviewed or merged. Security-relevant reports should follow the\nprocess described in [`SECURITY.md`](https://github.com/NVIDIA/auto-ontology/blob/main/SECURITY.md). Maintainer workflow is in\n[`CONTRIBUTING.md`](https://github.com/NVIDIA/auto-ontology/blob/main/CONTRIBUTING.md).", "url": "https://wpnews.pro/news/nvidia-open-source-auto-ontology", "canonical_source": "https://github.com/NVIDIA/auto-ontology", "published_at": "2026-11-03 07:04:50+00:00", "updated_at": "2026-11-03 07:21:14.061634+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "agent-protocols", "developer-tools", "artificial-intelligence"], "entities": ["NVIDIA", "Auto Ontology", "NVIDIA NIM", "Postgres", "pgvector", "Better Auth", "LangChain", "BIRD"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/nvidia-open-source-auto-ontology", "markdown": "https://wpnews.pro/news/nvidia-open-source-auto-ontology.md", "text": "https://wpnews.pro/news/nvidia-open-source-auto-ontology.txt", "jsonld": "https://wpnews.pro/news/nvidia-open-source-auto-ontology.jsonld"}}