What is APEX Document Management?

APEX Document Management (ADM) is a document management system that runs inside your Oracle database, with an Oracle APEX front end. It stores versioned files in folders, with users, groups, shares and full-text search. RAG collections are part of its optional AI Pack. See the ADM product page .

Is the AI Pack included in my ADM license?

Yes. The AI Pack comes with every ADM license, and it is optional to install. The license tiers differ only in the number of servers. Cloud model and OCR usage is billed separately by your provider. See ADM pricing .

Do I need a separate vector database?

Not on Oracle 23ai or 26ai. There, a collection stores its vectors in Oracle AI Vector Search, in the same database as your documents. On Oracle 19c, a collection needs a Qdrant server.

Which AI providers can I use?

ADM calls the models through UC AI , our open-source PL/SQL library. OpenAI, OCI Generative AI and a local Ollama server work, and so do the other providers that UC AI supports. You can use one provider for the embeddings and another for the answers.

Can a user get an answer from a document they are not allowed to open?

The search runs as the ADM user of the session. It returns only chunks from documents that user may open, before anything goes to the answer model. Folder permissions, group membership and shares all apply. In an APEX application, the page passes the signed-in user to ADM. The tutorial shows how, also for a chat region that runs in a background job.

Does my data leave the database?

The text of the chunks goes to the embedding and answer models that you configure. To build the index, the embedding model reads every document in the collection. With a local Ollama server and no cloud text extraction, nothing leaves your network. Your documents, chunks and vectors stay in your Oracle database, or in your Qdrant server if you use one.

How do I add a chat to my APEX application?

Give the collection to an AI agent as a search tool, and show the agent in the UC AI Chat region on an APEX page. The lesson Add an APEX chat page shows each step.

Can I change the settings of a collection later?

Yes, with the collection manager or update_rag_collection : the threshold, hybrid search, the answer model, text extraction and more. The embedding model and the vector store are fixed when you create the collection. To change them, create a new collection.

AI Pack for APEX Document Management

Ask your documents.
Get answers with sources.

APEX Document Management (ADM) is a document store that runs in your Oracle database. With its AI Pack, people ask questions about their documents and get answers that name their sources.

Answers use only documents the signed-in user may access.
Documents stay in your infrastructure. Choose a cloud provider or a local model to process their content.
Included with every ADM license, optional to install.

Request a demo Start the tutorial

AM Asked by an employee

How many vacation days can I carry over to next year, and when do I have to use them?

You can carry over up to 5 unused vacation days into the next year. 1

Carried-over days must be used by 31 March. After that date they expire. 1

To carry over more days, for example after a long project, your manager must approve it in writing. 2

Sources

1 vacation-policy-2026.pdf, section 4

2 hr-faq.docx, "Vacation"

An example answer. The question and the documents are made up.

Permissions

Answers use only documents the user may access

A question runs as the ADM user who asks it. The search returns only chunks from documents that this user may open, through folder permissions, group membership or a share.

So one collection can cover several departments, and each person still gets answers only from their own documents. There is no second permission model to keep in sync.

Indexing is separate from asking: to build the search index, the embedding model reads every document in the collection. The permission check applies when someone asks, before any text goes to the answer model.

Same question, two users: "How much did the repair of the Station Road burst cost?"

Customer service user

The documents do not contain this information.

Management user

The repair cost was £48,600. incident-report-IR-2026-031.pdf

Answers from our test, shortened. The incident report is in a folder that only management may open.

How it works

Choose the documents, then ask

A collection is a set of documents that people can ask questions about. You choose which documents belong to it, and ADM keeps it up to date.

Choose the documents

Point the collection at a folder, or at every document that carries a label (an annotation). A label on a folder covers everything in it, so one collection can span several departments.

ADM prepares them

ADM prepares your documents for search and updates the index when files change. You write no indexing code.

Ask

Ask in the collection manager, or in a chat on your own APEX page. Developers can also call it from PL/SQL, or give the collection to an AI agent as a search tool.

Diagram: the discovery job finds documents, the processing job extracts, chunks and embeds them, and a question is answered from the matching chunks the user may see

Collection manager

Try a question without writing code

The collection manager in the ADM app shows every file, its chunks, and which reader produced its text: the built-in filter, a vision model or OCR. Test search shows the answer, the model that wrote it and the documents retrieved for it.

Test search in the RAG collection manager: the generated answer, the model that wrote it and the five documents retrieved for it — A test search in the collection manager, with a small local model. "Sources" lists the documents retrieved for the model, not only the ones the answer cites. Recreated and shortened from a test search in the collection manager, with a small local model. The list shows the documents retrieved for the model.

For developers

The same in PL/SQL

Everything the collection manager does is also a PL/SQL call, so you can build answers into your own APEX pages and processes. The configuration is one JSON document: which documents, which embedding model, which answer model.

Create the collection

declare
  l_id number;
begin
  l_id := adm_ai_rag_api.create_rag_collection(
    p_collection_name => 'company_kb',
    p_config_json     => '{
      "source_file_annotation_key": "kb",
      "content_type": "text",
      "embedding": {
        "provider": "openai",
        "model": "text-embedding-3-small",
        "dimensions": 1536,
        "config": { "g_apex_web_credential": "OPENAI" }
      },
      "answer_model": {
        "provider": "openai",
        "model": "gpt-6.1-sol",
        "config": { "g_apex_web_credential": "OPENAI" }
      }
    }'
  );
end;

Scroll sideways to see the full code.

Ask a question

l_answer := adm_ai_rag_api.generate_answer(
  p_rag_collection_id => l_id,
  p_query             =>
    'How many vacation days carry over?'
);

What you get back

A Markdown answer with source references, from the documents that the current ADM user may open. The scheduled jobs keep the collection current, so there is no indexing code to write.

Features

What a collection does

Office files, PDFs, data files and e-mail

PDF, Word, Excel, PowerPoint, OpenDocument, RTF, HTML, Markdown, CSV, text, JSON, XML and e-mail are read out of the box. For scans, a collection sends selected files to a vision model or an OCR service (ADM 26.2 or later).

Stays up to date

Background jobs index new and changed files, and remove files moved outside the collection. Run a sync when users need newly uploaded content at once. Trashed documents are excluded immediately.

Oracle AI Vector Search or Qdrant

On Oracle 23ai or 26ai, the vectors stay in the same database, with nothing extra to run. On Oracle 19c, a collection uses a Qdrant server. You choose when you create the collection.

Cloud or local models

ADM calls the models through UC AI, our open-source PL/SQL library. OpenAI, OCI Generative AI, a local Ollama server and other providers work. API keys stay in APEX web credentials.

Agents and APEX chat

Give a collection to an AI agent as a search tool. The agent can split a long question into several searches. Show it in the UC AI Chat region on any APEX page.

Every answer can be traced

A debug call stores the question, every matched chunk with its score, the sources and the prompt. Read it in the collection manager, or as a Markdown report from PL/SQL.

What you need

Requirements and deployment options

APEX Document Management

The document store that the collections read from, on Oracle APEX 24.2 or later. The AI Pack comes with every ADM license, is optional to install, and installs the UC AI library with it.

An Oracle database

Oracle 23ai or 26ai keeps the vectors in the database, with Oracle AI Vector Search. On Oracle 19c, a collection stores its vectors in a Qdrant server.

A model provider

A cloud provider, with its API key stored as an APEX web credential, or a local Ollama server. Cloud model and OCR usage is billed by the provider.

Outbound network access

The database must reach the model provider and, if you use it, Qdrant. Configure a network ACL, and a proxy where your environment requires one.

Details are in the AI Pack installation guide. For licensing, see ADM pricing or ask us.

Learn more

Tutorial · 7 lessons

Frequently Asked Questions

Common questions about RAG collections in APEX Document Management.

Still have questions?

Contact Our Team

See it in a live demo

We show a RAG collection with users from different groups: the answers, the sources, and what each user may see. Bring your questions.

Request a demo Start the tutorial

About APEX Document Management

Ask your documents.
Get answers with sources.

Answers use only documents the user may access