Build your first data application

This tutorial follows one fixed path. You will load the two files in the tutorial data package into MOI, process them with a workflow, add the results to a knowledge base, and use an agent to produce a downloadable Markdown file.

Before you start, complete the five checks in Before you begin. The tutorial creates the following objects. Use these names throughout:

Object

Fixed name or location

Source files

moi_quickstart / data / source

Processed files

moi_quickstart / data / processed

Workflow

moi_quickstart_pipeline

Knowledge base

moi_quickstart_knowledge

Each step ends with a Completion check. Continue only after you see the expected result.

1. Open the Create Import Task page

Start by opening the page where you will load local files into MOI.

  1. Sign in to MOI and confirm that the workspace prepared for this tutorial appears in the top-left corner.

  2. In the top navigation, click Data.

  3. Under Data Connection in the left navigation, click Import Tasks.

  4. In the upper-right corner of the Import Tasks page, click Create Task.

MOI opens the import task creation page. No files have been uploaded and no task has been created yet.

Completion check: The page title is Create Import Task, and settings for the load type and data source are visible.

2. Load the tutorial data

Upload the two tutorial files to the source volume and wait for the import task to finish.

  1. For Load Type, select Unstructured.

  2. For Data Source, select Local Upload.

  3. Under the target Data Volume, create the catalog moi_quickstart, the database data, and the volume source. Select the resulting path moi_quickstart / data / source.

  4. In the same database, create another volume named processed. Do not upload files to this volume; the workflow will write its output here later.

  5. Add the two extracted files from the tutorial data package. Confirm that only these two files appear in the list.

  6. Leave the remaining settings unchanged and click Create & Start Loading.

  7. Wait until the status on the task details page changes from processing to Finished.

The two source files are now in the source volume. The processed volume is still empty.

Completion check: The task status is Finished, with two successful files and zero failed files.

3. Verify the files in Catalog

Confirm that Catalog has registered the location of the two files you just loaded.

  1. Under Resource Center in the left navigation, click Catalog.

  2. Open moi_quickstart, then data, then source.

  3. Check the current location at the top of the page. It must be moi_quickstart / data / source.

  4. Confirm that both files from the tutorial data package appear in the file list.

These are the same files loaded in step 2. Do not upload them again. The workflow will read them from this Catalog location.

Completion check: The page shows the correct Catalog path and both files.

4. Create a workflow from the template

Create the workflow that will parse the files, split their content, and build a vector index.

  1. Under Data Processing in the left navigation, click Workflow.

  2. In the upper-right corner of the Workflow page, click Create Workflow.

  3. In the Create Workflow window, find RAG Data Preparation for Mixed Documents and click Use Template.

  4. Enter moi_quickstart_pipeline as the workflow name and confirm.

MOI opens the workflow editor. The template already contains the nodes and connections required to read the files, parse the documents, split the content, build the vector index, and save the results. Do not add, remove, or reconnect nodes in this tutorial.

Completion check: The workflow editor shows the name moi_quickstart_pipeline, with the template nodes connected on the canvas.

5. Set the input and create the knowledge base

Point the workflow to the source files verified in step 3, then create the knowledge base that will receive the processed data.

  1. Open the Data Read node on the canvas.

  2. Under Catalog Source, select moi_quickstart / data / source, confirm the selection, and return to the canvas.

  3. Open the Build Vector Index node.

  4. Under Knowledge Base, create a knowledge base named moi_quickstart_knowledge, then select it.

  5. Leave the document-processing and embedding models at the workspace defaults. Confirm the settings and return to the canvas.

This step creates an empty knowledge base. The parsed, split, and indexed content will enter it only after the workflow completes successfully.

Completion check: The Data Read node shows moi_quickstart / data / source, and the Build Vector Index node shows moi_quickstart_knowledge.

6. Set the output and submit the workflow

Choose where the processed files will be saved, then save the workflow and submit its first execution.

  1. Open the Save to Catalog node on the canvas.

  2. Set Target location to moi_quickstart / data / processed, confirm the setting, and return to the canvas.

  3. Check the three key values: the input is source, the knowledge base is moi_quickstart_knowledge, and the output is processed.

  4. In the upper-right corner, click Complete.

MOI saves the workflow and immediately submits an execution for the two input files. You do not need to click a separate run button.

Completion check: MOI displays Execution submitted, and Executions is available at the top of the workflow page.

7. Wait for the executions to complete

The workflow creates one execution record for each file. Because you uploaded two files, both records must finish successfully.

  1. At the top of the workflow page, open Executions.

  2. Find the two execution records you just submitted.

  3. If either record is Pending or Running, stay on the page and wait for the status to update.

  4. Confirm that both records reach Success.

The workflow has now parsed the documents, split their content, built the vector index, and written the results. The next step only checks that the knowledge base received those results; it does not process the files again.

Completion check: Both execution records show Success, with no failed record.

8. Confirm that the knowledge base received the data

The knowledge base created in step 5 was initially empty. The completed workflow executions in step 7 added the two processed files to it.

  1. Under Resource Center in the left navigation, click Knowledge Base.

  2. Search for moi_quickstart_knowledge.

  3. Find the matching knowledge base card and confirm that it shows 2 files.

  4. Open the knowledge base and confirm that its data is ready.

Do not upload a ZIP file or create another knowledge base. The two files in this knowledge base came from the workflow executions you just completed.

Completion check: moi_quickstart_knowledge contains two files, and its data is ready.

9. Verify retrieval and sources with a chat

Before using the knowledge base with an agent, run one fixed query to verify that MOI can retrieve its content and cite the sources.

  1. On the moi_quickstart_knowledge card, click Chat.

  2. On the chat page, click New Session.

  3. Paste the following prompt into the input box and send it:

Using only the current knowledge base, summarize the main points in these materials.
Cite the source for each conclusion.
List anything the materials do not state explicitly under Items to verify.

  1. Wait for the response to finish.

  2. Check the main conclusions and confirm that each one includes a source you can open.

  3. Check the end of the response and confirm that information not stated explicitly in the materials appears under Items to verify instead of being presented as fact.

This chat verifies document parsing, retrieval, and source citations. You will create the application result in the next step.

Completion check: The response includes source citations and lists unsupported information under Items to verify.

10. Generate an application result with an agent

Bind the verified knowledge base to an agent and use the same data to produce a downloadable Markdown file.

  1. In the top navigation, click Agent.

  2. Click New Chat, then select Knowledge Explore.

  3. In the model field, use the workspace’s default language model.

  4. Open the Knowledge Base panel and search for moi_quickstart_knowledge.

  5. Select the knowledge base and click Bind. Confirm that the panel shows 2 files total and Ready.

  6. Close the Knowledge Base panel. Paste the following prompt into the chat input and send it:

Use only the knowledge base bound to this conversation.
Create a Markdown report that summarizes the main points.
Cite the source for each conclusion.
List anything the materials do not state explicitly under Items to verify.
When the report is complete, place the file in Outputs so it can be downloaded.

  1. Stay in the current chat and wait for the agent to finish.

  2. In Outputs, find the generated Markdown file and click Download. Confirm that the file can be saved locally.

The agent uses the knowledge base verified in step 9. It does not upload or copy the source data again. Once the downloadable file appears, you have completed the path from data ingestion to an application result.

Completion check: A downloadable Markdown file appears in Outputs for the current chat.

Tutorial complete

You have loaded data, verified its Catalog registration, processed it with a workflow, validated the resulting knowledge base, and produced an application result with an agent. Continue to What you just did to see how this path turns a set of one-off actions into reusable data capabilities.

Last updated on