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

# Global AI Assistant (Beta)

> Answers from your account's data and your own regulatory documentation

The Global AI Assistant answers questions about the data in your account. Instead of filtering a table, searching for a property, or clicking around the map to find what you need, you ask the question and get an answer. It's in beta right now, so we're actively making it better with help from the accounts testing it.

It works across your whole account: records, objects, properties, files, and tasks.

Upload your ordinances, codes, and procedures to [AI knowledge](/docs/ai/knowledge-base) and it answers from those too, so the rules you work from are at your fingertips.

<Note>
  **This feature is in beta.** It's on the web app only, and not every account has it yet. If you don't see the **Ask Forerunner** button, contact your Customer Success Manager to learn more.
</Note>

## Opening the assistant

<Steps>
  <Step title="Click Ask Forerunner">
    You'll find the **Ask Forerunner** button in the top right of the page. It's available from anywhere in the web app.
  </Step>

  <Step title="Ask your question">
    Type your question the way you'd ask a colleague. You can also click one of the suggested prompts to get started.
  </Step>

  <Step title="Watch what it looks up">
    As it works, the assistant shows you which data it's pulling. This is how you check that it looked in the right place.
  </Step>

  <Step title="Follow the links">
    Records, objects, and properties in the answer are links. Click through to the record itself to verify anything before you act on it.
  </Step>
</Steps>

You can move the panel between a sidebar and a floating window with the layout toggle, and start over at any time with **New chat**. Each new chat begins fresh with no memory of earlier conversations.

## What you can ask

| What you need                       | Example                                                                                      |
| ----------------------------------- | -------------------------------------------------------------------------------------------- |
| **A count or a breakdown**          | "How many permits did we issue last month, broken down by type?"                             |
| **A list of records**               | "Which properties have an elevation certificate on file?"                                    |
| **A narrative writeup or an email** | "Write up a summary of everything we've done at 214 Oak Street this year and email it to me" |
| **Your own guidance**               | "What do our regulations require for substantial improvement?"                               |
| **Overdue work**                    | "What corrective actions are open and past their due date?"                                  |
| **A trend**                         | "Show me new records per month for the past year"                                            |

<Tip>
  Be specific about the record type and the date range. "How many inspections did we complete in July?" gets a better answer than "how many inspections."
</Tip>

### Ask for prose, not just numbers

The answer doesn't have to be a table. The assistant can write the narrative for you: a summary of a property's history, a paragraph for a council update, a plain-language explanation of what a set of records shows, or the body of an email you're about to send. Say what you want it to read like.

* "Summarize this property's permit history in a short paragraph I can paste into a letter"
* "Write a plain-language explanation of why this property needs an elevation certificate"
* "Give me a few sentences on how our inspection backlog changed this quarter"

It never sends anything on your behalf, so you have full control over what it writes. Copy it straight out of the chat, edit it, or paste it wherever you need it.

## Answers from your own regulations and procedures

This is one of the most useful things the assistant does, and for most teams it's the reason to reach for it first.

Upload your documents to [AI knowledge](/docs/ai/knowledge-base) and the assistant searches them and answers from them: your ordinances and codes, permitting and compliance procedures, technical standards, regulatory guidance, policy manuals, and staff runbooks. The more you add, the more it can answer, and you don't have to know which document holds the rule or where it sits. You ask the question.

Your documents are the authority. Where your own rules differ from the general ones, the assistant follows yours, and it quotes the passage it relied on rather than paraphrasing loosely.

### What to ask

| What you need              | Example                                                                     |
| -------------------------- | --------------------------------------------------------------------------- |
| **A requirement**          | "How much freeboard do we require above base flood elevation?"              |
| **A specific section**     | "Summarize the substantial improvement section of our floodplain ordinance" |
| **A defined term**         | "How do our regulations define substantial damage?"                         |
| **A procedure**            | "What's our process for a variance request?"                                |
| **A threshold**            | "At what point does a repair become substantial improvement for us?"        |
| **Plain-language wording** | "Explain that rule in language a homeowner would follow"                    |

You can ask by exact section or code number, or in your own words when you don't know where it's written down. The assistant handles both, so you don't have to know the structure of your own documents to get an answer out of them.

<Tip>
  If you know which document governs, name it. "In our stormwater manual, what applies to sites under one acre?" narrows the search and gets you a tighter answer.
</Tip>

### Your rules and your data in one answer

The most valuable questions combine the two. Because the assistant reads your records and your documents in the same conversation, it can look up what's actually true about a property and then tell you what your own rules require:

* "This property has 45% damage. What do our regulations require at that level?"
* "Which open permits are missing something our permitting procedure requires?"
* "Does this elevation certificate meet our current standard?"
* "Draft the paragraph I'd send this owner, citing the rule that applies"

That's the part that's hard to do by hand. Answering it yourself means pulling the record, finding the governing document, locating the section, and then writing it up.

## It knows what you're looking at

The assistant picks up the page you have open, so you can talk about "this" and "these" instead of describing things again.

| Where you are          | What that gets you                                                  |
| ---------------------- | ------------------------------------------------------------------- |
| A record you have open | Ask about it, or ask it to fill in and change field values for you  |
| A records table        | Ask about the records in view, including any filters you've applied |
| A property on the map  | Ask about that property without typing the address                  |
| An object on the map   | Ask about the object you have selected                              |
| An automation          | Ask what it does or when it last ran                                |
| A workflow             | Ask about the records currently moving through it                   |

## What it can't do yet

Being clear about the edges is part of the beta, and this list should get shorter as it goes on. Today the assistant cannot:

* **Create or delete records, or submit anything.** On a record you have open it can fill in and change field values, and you review and save them. Everywhere else it only reads. It can't create a new record, delete one, or approve or submit on your behalf.
* **Answer questions about map layers.** Anything that depends on what's inside a boundary, such as "which properties are in the evacuation zone" or "how many records fall in our repetitive loss area," isn't supported yet. Ask by a property attribute instead, like flood zone or FIRM panel.
* **Work on mobile.** The mobile app has the [Record AI Assistant](/docs/ai/ai-assistant) for filling out records.
* **Email anyone but you.** It can't send data to a resident, a colleague, or any outside address.

When you ask for something it can't reach, it will say so and offer the closest thing it can do rather than guessing.

## Permissions and privacy

Every question runs under your own login. The assistant sees exactly what you see in the product and nothing more, so a user who can't view a record type won't get answers about it. It works only from your account's data, not from the open internet.

Managers can download a record of every question your team asks and every answer they get. See [Exporting AI activity](/docs/ai/exporting-ai-activity).

For how Forerunner keeps AI answers accurate and how your data is handled, see [AI overview](/docs/ai/overview) and [Data security and privacy](/docs/administration/data-security).

## Tips for better answers

**Verify before you act.** Click into the records behind an answer, particularly for anything going into a report or a letter to a resident. It's in beta, so check its work.

**Keep one topic per chat.** Follow-up questions carry context, which helps. When you switch topics, start a **New chat**.

## Sharing feedback

The assistant is in beta, and that's the point of this phase: what you tell us decides what gets built and fixed next.

Use the thumbs up and thumbs down on any answer, especially when something is wrong, missing, or badly worded. Those go straight to the team building it.