AutoGraph Quick Start

Turn a pile of documents into a knowledge base you can chat with, with answers cited back to the source

Prerequisites

An Arango Contextual Data Platform deployment.
A project in your target database (keeps datasets isolated). See Projects.
LLM and embedding API access (OpenAI-compatible).
A valid JWT for the API (Authorization: Bearer ...).

Build your knowledge base

Create a project

In the sidebar, open Agentic AI Suite > AutoGraph Studio and create a project. A project is an isolated workspace: you keep one set of related documents per project and their data stays separate from everything else.

Upload the documents you want answers from

Add your files (PDF, Office, text, and more). This is the raw material AutoGraph reads.

Connect an AI model / LLM

Enter your model provider and key (for example, an OpenAI key covers both the chat and embedding settings). AutoGraph uses the model to read your documents now and to answer your questions later. Leave the import settings at their defaults for your first run.

Let AutoGraph analyze your documents

Deploy the service and start the build (shown as the Corpus Graph in the UI). AutoGraph reads every document, groups them by the topics they actually cover, and picks the right processing depth for each group, so that dense technical material and simple FAQs are each handled appropriately. This runs on its own; wait for it to finish.

Build the knowledge base

Click Generate strategies. Pick how much of the corpus should lean on graph extraction with the Complexity slider — the balanced default is a good first run — then review the strategies AutoGraph picked per topic group and click Build Knowledge Graph. AutoGraph turns each topic group into a part of a searchable knowledge base. Wait for the Knowledge Graph built confirmation.

Deploy a retriever

Deploy a retriever against your Context Graph. It is what your questions, agents, and applications talk to. Wait for it to go live.

Ask your first question

Ask a question in plain language. Answers come back with citations linking to the source documents they came from.

To add, remove, or replace individual documents later, use the API. The web interface does not offer these operations in Arango Contextual Data Platform 4.1.0. See Incremental Graph Updates.

First, install the AutoGraph service through the Arango Control Plane:

POST https://<EXTERNAL_ENDPOINT>:8529/_platform/acp/v1/autograph

Then run the pipeline. Every call uses an Authorization: Bearer <token> header. Each step below links to the endpoint’s full reference, and the Reference section lists every endpoint in one table.

Check the service is ready

GET https://<EXTERNAL_ENDPOINT>:8529/autograph/v1/health

Confirm AutoGraph is up before you start. It is a liveness and token check, so a 200 here does not mean the model providers or the File Manager are reachable. See Endpoints.

Upload your files

POST https://<EXTERNAL_ENDPOINT>:8529/_platform/filemanager/_db/:database/rag-input

Upload your documents to the File Manager under the scope [project, category], one category per knowledge domain, for example legal and engineering.

Build the corpus

POST https://<EXTERNAL_ENDPOINT>:8529/autograph/v1/corpus/builds

List your category labels in categories. Builds the Corpus Graph and discovers domain clusters. Returns 202 with a corpus_build_id. See Create Corpus Build.

Poll GET /v1/corpus/builds/{corpus_build_id} until status is completed before moving on, and check error_code even then, see Build error codes.

Generate RAG strategies

POST https://<EXTERNAL_ENDPOINT>:8529/autograph/v1/rag-strategizer/analyze

Assigns each domain a strategy (FullGraphRAG or VectorRAG) and an ontology. project and complexity are required; send complexity: "very_high" if every domain has to answer entity-based questions, see How complexity is applied. Run this only after the corpus build has completed. See Trigger RAG Strategizer.

Poll GET /v1/rag-strategizer/jobs/{strategize_job_id} until status is completed.

Orchestrate the import

POST https://<EXTERNAL_ENDPOINT>:8529/autograph/v1/orchestrate

Spawns the per-domain Importer workers to build the knowledge graph. Run this only after the strategizer has finished. Returns 202 with an orchestration_id. See Trigger Orchestration.

Poll GET /v1/orchestrate/{orchestration_id} until it reports a terminal status. Only completed means the knowledge graph is ready.

Keep the graph up-to-date (optional)

POST https://<EXTERNAL_ENDPOINT>:8529/autograph/v1/graph/update

Once the graph is built, you can add, remove, or replace individual documents with /v1/graph/insert, /v1/graph/delete, and /v1/graph/update instead of rebuilding the corpus. After an insert or update, call /v1/orchestrate again with the file_ids of the changed documents so that the change reaches the knowledge graph. See Incremental Graph Updates.

Insert and update identify every document by its File Manager file_id, which is why step 2 uploads through the File Manager. The deprecated POST /v1/import-multiple path uploads files directly to the service, but a corpus built that way cannot be changed document by document. See Identifying documents for Layer 3.

At any point, GET /v1/projects/{project}/overview reports where the project stands: the document and cluster counts, and which categories still need a build, strategies, or an orchestration.

Next, query your knowledge base with AutoRAG.

You now have a knowledge base built from your documents and a chat interface that answers questions about them, with citations back to the source. To query it programmatically, see AutoRAG.

Next steps

Use Cases: Real-world enterprise scenarios.
Incremental Graph Updates: Add, remove, and replace documents in a knowledge graph that has already been built.
Architecture: The three-layer knowledge graph and resulting collections.
Design Guide: Structure data with categories, layers, and components.