> ## Documentation Index
> Fetch the complete documentation index at: https://docs.aroramcp.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Pull exactly the records you need from big app data

> Big app results get cut off, error out, or flood your chat. Arora saves them whole, and this tool pulls out exactly the answer you need.

Some questions sound small and come back huge. "Show me my open ClickUp tasks" or "pull this customer's invoices" seems simple. But the app sends back every field on every record: custom fields, metadata, history. A short list can arrive as tens of thousands of characters.

## What goes wrong without it

An AI app has three bad options when a result is that big:

* **Cut it off.** The AI only sees part of the data and doesn't know the rest is missing. A 10,000-row spreadsheet comes back as 132 rows, and the total it gives you is off by a factor of 72. No warning.
* **Fail.** The result is too big to pass through, and the request errors out.
* **Flood the chat.** It fits, barely, and buries your AI. Answers get worse and the conversation runs out of room sooner.

## What Arora does instead

**[Heimdall](https://docs.aroramcp.com/guardians/heimdall.md)**, one of Arora's [Guardians](https://docs.aroramcp.com/guardians/overview.md), catches oversized results before they reach your AI and saves them whole on your [workspace](https://docs.aroramcp.com/tools/workspace.md). Your AI works from the full result, not a cut-off sample, and your chat doesn't fill up. If a result is ever too big even for that, Arora says so. Nothing gets trimmed without you knowing.

Then this tool takes over. Your AI asks the saved data a precise question and gets back only the answer: the 12 orders you asked about, not all 1,000.

## Why it matters

* **No confidently wrong answers.** When data gets cut off quietly, your AI has no idea anything is missing. It answers from part of the picture with full confidence, and you can't tell. With Arora, your AI has the whole result, or it tells you it doesn't.
* **Complete data.** The whole result is kept and every record is checked. Nothing gets skipped.
* **Room to keep working.** Your chat stays clear, so long conversations stay sharp. Fewer tokens, faster replies.

## What you can ask

You never write the filter. Ask in plain English and your AI does the rest.

**Pick out the records that match**

```text theme={null}
From those orders, show me only the ones over $500.
```

**Keep only the fields you need**

```text theme={null}
For each of those, just give me the order number, customer email, and total.
```

**Group and count**

```text theme={null}
Group those ClickUp tasks by assignee and tell me how many each person has open.
```

**Find the needle**

```text theme={null}
Which HubSpot deals in that list have no next step and haven't been touched in 30 days?
```

**Reshape for the next step**

```text theme={null}
Turn those contacts into a list of name and email I can paste into a spreadsheet.
```

## How it fits together

1. You ask for data from one of your apps.
2. If the result is big, Heimdall saves it on your workspace and tells your AI it's there.
3. You ask your question. Your AI uses this tool to pull out just the answer.

It only reads. It never changes the data on your workspace or in your apps.

## Good to know

* It works on structured app data, which is what most apps send back. For text like transcripts or documents, your AI uses workspace search instead.
* For totals and averages with a built-in audit, your AI uses [exact math](https://docs.aroramcp.com/tools/math-and-charts.md). This tool is for picking, trimming, and regrouping records.
* Very heavy requests stop after 5 seconds. If that happens, ask your AI to narrow it down first.

> [accordion] Technical details
>
> Tool name: `bridge_gungnir` (shown as "Bridge: Filter & Reshape Data"). Runs a read-only filter over a staged JSON file on the workspace. If the result is itself large, it's staged too, so it can be filtered again.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.