Use cases

Each use case says what problem it solves and links to the page that explains how it works. Most also show example prompts or code.

The prompts work in the Manta chat, either in Manta itself or in the embedded chat you mount in your own app. The agent answers from the datastore the user has access to, so the answers are about your data, not general knowledge.

Ask questions about your business data

Your data sits in dashboards and exports, and every new question means a new report. The agent queries the knowledge graph directly and answers in plain language, with a table or chart when it helps.

• “Which campaigns spent the most last week, and how many conversions did each one bring?”
• “Show me a chart of daily spend for the last 30 days.”
• “Which ads have spend but no clicks this month?”

Learn more: Agents and chat

Bring your own data in

Your data lives in a system Manta does not integrate with, or in a spreadsheet. Upload it as CSV or Parquet and it is ingested into the knowledge graph, next to the data from connected sources.

pnpm manta-upload orders.csv --datastore your-datastore

Once the batch is ingested, ask about it like any other data:

• “How many orders came in from each country this month?”
• “Did my last upload ingest successfully?”

You can also attach a file to a message in the chat, and the agent reads it as part of the conversation.

Learn more: Getting data in

Keep derived values up to date with rules

Totals, statuses and scores computed in your app drift out of date as the data changes. Describe them once as a rule, and the graph derives them for every entity and keeps them current.

• “Add a rule that calculates the cost per click for every ad.”
• “Explain how the cost per click for this ad was calculated.”
• “Which rules in this datastore deal with budgets?”

The agent writes the rule and stages it. Nothing goes live until the staged rules are validated and committed.

Learn more: Rules

Turn data into actions, with approval

You find a problem in the data and then have to fix it by hand in another platform. The agent proposes the fix as an action plan: the changes it would make, which you review and run.

• “Pause the ads that spent more than €50 this week without a conversion.”
• “Propose a plan to move budget from the weakest campaign to the strongest one.”

A plan needs approval until it becomes trusted, and every run is recorded. Changes that cannot be applied automatically become manual actions for a person to resolve.

Learn more: Actions

A copilot that knows what the user is looking at

Users should not have to name the campaign or order they are looking at. Pass the entities on screen as pageContext, and the agent answers about “this campaign” or “these orders” directly.

<EmbeddedChat {...props} pageContext={{ iris: [campaignIri], typenames: ["Campaign"] }} />

• “Why did spend on this campaign drop yesterday?”
• “Summarize how these ads performed compared with last week.”

Learn more: Tell the agent what the user sees

Read the graph from React

You want your own screens on top of the same data the agent works with, without writing queries by hand. manta-codegen generates typed LDKit schemas from your datastore, and the SDK’s hooks read entities through them.

const options = useSparqlOptions();
const { data = [] } = useLensFind({ schema: YourSchema, options, queryKey: ["your-entities"] });

Learn more: Schemas and LDKit

Automate from a backend

A scheduled job or another service needs to manage rules or plans, with no user and no React. Create a custom app with machine access, mint a token for it with @manta-eu/auth, and call the same GraphQL and SPARQL APIs the SDK uses.

const accessToken = await mintWithClientCredentials({ clientId, clientSecret });

Learn more: Calling Manta from a backend

Dashboards that render on the server

You want the same dashboards over your data in every app that shows them. Manta Horns runs their queries on the server as the viewer, and draws them in any React host.

Learn more: Manta Horns

Recipes

Recipes chain several steps, and each one builds on the answer before it.

From a spreadsheet to an approved change

1. Upload ads.csv with manta-upload and wait for the batch to be ingested.
2. “Which of these ads have the highest cost per conversion?”
3. “Add a rule that flags an ad when its cost per conversion is more than twice the campaign average.”
4. “Propose a plan to pause the flagged ads.”
5. Review the plan and run it.

Explain a number

1. “What was our total spend last month?”
2. “Break that down by campaign.”
3. “How was the spend for the top campaign calculated, and where does the data come from?”
4. “Chart it by week.”

Teach the agent your conventions

1. “When I say ‘last week’, I mean Monday to Sunday.”
2. “Remember that we report all amounts in euros.”
3. In a new conversation: “What was the spend last week?”