Agentic AI Suite

What’s included

The Agentic AI Suite is composed of the following major components:

• Ada: The AI digital assistant, for natural language interaction and development.
• AutoGraph and AutoRAG: AutoGraph organizes enterprise data into a Context Graph, assigning each domain the right processing depth. AutoRAG is the retrieval layer on top: it deploys the retrievers that answer questions from that Context Graph. Both stages are driven from the AutoGraph Studio view of the web interface. See GraphRAG Concepts for the approach behind them.
• Natural Language to AQL/AQLizer: Generate AQL queries from natural language to explore your data and gain insights without having to learn the query language first.
• Reasoner: Automatically analyze and optimize AQL queries using AI-powered reasoning, with validated performance improvements.
• GraphML: Apply machine learning to graphs for link prediction, classification, and computing embeddings.
• Graph Analytics: Run graph algorithms such as PageRank on dedicated compute resources to discover influential nodes and patterns.

Most components have an intuitive graphical user interface integrated into the Arango Contextual Data Platform web interface, guiding you through the process. In the left-hand sidebar, Agentic AI Suite groups AutoGraph Studio, Graph Analytics, and GraphML.

Alongside these components, you also get the following additional features:

• Jupyter notebooks: Run a Jupyter kernel in the Contextual Data Platform for hosting interactive notebooks for experimentation and development of applications that use ArangoDB as their backend.
• Public and private LLM support: Use public large language models (LLMs) such as OpenAI or private LLMs with Triton Inference Server.
• MLflow integration: Use the popular MLflow as a model registry for private LLMs or to run machine learning experiments.
• Application Programming Interfaces (APIs): Use the underlying APIs of the Agentic AI Suite and build your own integrations. See the API Reference  for more details.

Where your data lives

The Arango Contextual Data Platform deploys and integrates multiple services, but the data itself lives in the ArangoDB core database system. Everything the Agentic AI Suite produces (knowledge graphs, embeddings, analytics results, query history) is persisted as collections and documents in ArangoDB databases, alongside your existing application data.

The exception is raw files (PDFs, images, office documents, and other binaries) that you upload for Agentic AI processing, such as the documents you feed into AutoGraph. These are stored in object storage (S3, MinIO, or another blob store) and managed through the File Manager service. The same File Manager also holds the code packages uploaded through the Container Manager’s Bring Your Own Code flow, so its contents are not exclusive to the Agentic AI Suite. Any structured data extracted from uploaded files (entities, relationships, embeddings) is written back into ArangoDB.

Sample datasets

If you want to try out ArangoDB’s data science features, you may use the arango-datasets Python package to load sample datasets into a deployment.

Supported LLM and embedding models

The services of the Agentic AI Suite work with OpenAI-compatible APIs as well as self-hosted models served through Triton Inference Server. The recommended setup is the openai provider with the OpenAI models listed below: that is the combination ArangoDB tests, and other endpoints can differ in behavior such as latency.

You can still use any other OpenAI-compatible endpoint — OpenRouter, Google Gemini, Anthropic, Azure, or a corporate LLM — and run a model that is not on the list. In the Importer, AutoGraph, and AutoRAG, use the custom provider for these: it is the intended way to point a service at an OpenAI-compatible endpoint that is not the OpenAI API itself, and you should always set chat_api_url / embedding_api_url explicitly with it. In those three services, pointing the openai provider at a non-OpenAI URL is not supported. Natural Language to AQL has no custom provider and reaches such endpoints with openai plus a chat_api_url — see Natural Language to AQL setup. Models beyond the ones listed below are outside ArangoDB’s testing, so validate them in your own environment.

“OpenAI-compatible” here has a specific meaning: the suite talks to providers through the OpenAI Chat Completions client, so an endpoint must implement the /v1/chat/completions contract that client expects (and /v1/embeddings for embedding models). An endpoint that exposes only a different API surface is not supported, even if it is marketed as OpenAI-compatible. Some newer OpenAI models require the Responses API (/v1/responses) instead; the Importer and AutoGraph detect this and fall back automatically.

A model is listed as supported by the suite only if it works seamlessly across the Importer, AutoRAG, and AutoGraph services. Individual services may also work with additional models — for the full list available to a specific service, see that service’s own documentation (for example, Importer LLM Configuration).

A model marked Yes under Default is the one applied automatically when the model name is not set with the openai provider. Whether the custom provider falls back to the same model varies by service — check the parameter reference for the service you are configuring.