# AI Knowledge \[Outdated]

{% hint style="warning" %}
This is a BETA feature. 
{% endhint %}

## What is AI Knowledge?

AI Knowledge is Jimo’s built‑in large‑language‑model layer that transforms the documentation, files, and FAQs you feed it into a conversational help agent. It turns Jimo into an AI‑powered help desk. Users will interact with it through the **Smart Search** block you add to any Resource Center.

### What it’s used for

* **Instant answers** – Users ask questions in plain language and get precise, context‑rich replies.
* **Content discovery** – Replies can surface related documentation and link or launch Jimo Experiences, guiding users deeper without leaving the app.
* **Self‑serve support** – Reduces repetitive tickets by letting the AI handle common “how‑do‑I?” queries 24 / 7.

<figure><img src="/files/NfUU2vDZXMGEIchsIkNk" alt=""><figcaption></figcaption></figure>

### How it works

1. **You supply knowledge** in the Settings page via crawled URLs, uploaded files, pasted text, or Custom Answers.
2. **Jimo trains a private model** that indexes this corpus and retrains on the cadence you choose.
3. **End‑users engage** through the **Smart Search – AI Knowledge** block you add to any [Resource Center](/docs/experiences/resource-center.md). The field looks like a search bar; behind the scenes it queries your trained model and returns answers plus optional links/CTAs.

<figure><img src="/files/URcnxyBigwtAdHsd53Or" alt=""><figcaption></figcaption></figure>

> Everything the AI can say is governed by the content you manage in this Settings page—giving you full control over accuracy and tone.

***

## 1. Opening the AI Knowledge panel & tabs overview

1. From your project settings sidebar, select **⚙** [**Settings** › **AI Knowledge**](https://testing.i.usejimo.com/settings/ai-knowledge/general) ([wand icon](#user-content-fn-1)[^1]).
2. Use **Test Knowledge** (top‑right) at any time to [open a private chat and preview responses](#user-content-fn-2)[^2].

<figure><img src="/files/L8XFt1mpHUhUwZonzxvA" alt=""><figcaption></figcaption></figure>

| Tab                | What you manage here                                        |
| ------------------ | ----------------------------------------------------------- |
| **General**        | Retraining cadence · Summary tables · Global persona & tone |
| **Contents**       | All data sources that feed the AI (URLs, text, files)       |
| **Custom Answers** | FAQ‑style Q\&A pairs with optional CTA actions              |

***

## 2. General 🔧

This tab is your **control panel**: choose how often the model retrains, review content & FAQ summaries, and shape the AI’s persona.

### 2.1 Auto‑retraining frequency

<figure><img src="/files/q1MtILchWU9SS0gDZas1" alt=""><figcaption></figcaption></figure>

Choose **Never · Daily · Weekly**. Jimo will re‑index every content source on this schedule so answers stay up to date. You can still trigger a one‑off retrain at any time from the Contents list.

### 2.2 Content sources summary

<figure><img src="/files/ctYZ7B53ZdUMYEB3VQJQ" alt=""><figcaption></figcaption></figure>

A compact table showing every item you added in **Contents**:

* **Content** – Name or URL/file.
* **Type** – Site Crawl · URLs · Text · File.
* **Status** – *Trained* ✓ or *Training* ⏳.
* **Last updated** – Timestamp of the last successful crawl/upload.

Actions:

* **⟳ Refresh** – Retrain only this source.
* **🗑 Delete** – Remove the source entirely.
* Click the row for **Content Details** (see § 3.3).

### 2.3 Custom answers summary

<figure><img src="/files/JKsarCrPDmStG7yZdBzG" alt=""><figcaption></figcaption></figure>

Mirrors the structure above but lists each curated Q\&A pair (see § 4). Use **✏️ / 🗑** for quick edits.

### 2.4 General Instructions

<figure><img src="/files/3AiffdptfV2wlztDU2bd" alt=""><figcaption></figcaption></figure>

Markdown editor that defines the AI’s personality. Default template:

```
### Role
...
### Persona
...
### Constraints
...
```

Presets: **AI Agent · In‑App Coach · Onboarding Specialist · Support · Expert Explainer**. Edit or replace freely—changes apply to every answer.

***

## 3. Contents 📚

This tab is your **knowledge base**: add, retrain, and inspect every URL, file, or text snippet that fuels the AI.

### 3.1 Adding a source

Click **+ Add content** and choose a type:

{% tabs %}
{% tab title="URL" %}

<figure><img src="/files/PDizykiKpvXhjrHE54MB" alt=""><figcaption></figcaption></figure>

* **Site crawl** – Enter a starting URL. Jimo follows internal links, extracts the text of every reachable page, and splits long pages into manageable chunks. Add **Path filters** (e.g. `/docs/`, `/kb/`) to limit the crawl.
* **Single page** – Paste a specific URL when you only need one page indexed. Give the entry an optional **Source name**, then press **Train** to kick off indexing.
  {% endtab %}

{% tab title="Text" %}

<figure><img src="/files/dHA7TcqN3xHtdZYsui9S" alt=""><figcaption></figcaption></figure>

* Paste any plain‑text or Markdown snippet such as release notes or code comments.
* Keep each snippet under \~25 k characters for fastest processing.&#x20;
  {% endtab %}

{% tab title="File" %}

<figure><img src="/files/Y5BqkNEFp1zVWA4iLtck" alt=""><figcaption></figcaption></figure>

* Upload `.pdf, .doc/.docx, .txt, .csv, .md, .xls/.xlsx, .json, .html, .xml` files.
* Large files (> 10 MB) are automatically chunked so the model can process them efficiently.&#x20;
  {% endtab %}
  {% endtabs %}

⇒ Give the source an optional name and click **Train**. \
New sources appear with *Training* ⏳ status; they flip to *Trained* ✓ when ready.

### 3.2 Understanding status

| Status        | Meaning                                                      |
| ------------- | ------------------------------------------------------------ |
| **Training**  | Being processed—answers won’t use it yet.                    |
| **Trained** ✓ | Ready and searchable.                                        |
| **Failed** ⚠️ | Something went wrong—hover the icon for the error and retry. |

### 3.3 Content Details drawer

<figure><img src="/files/O7SYpLkLWCjg2yA5weug" alt=""><figcaption></figcaption></figure>

Click any source to inspect:

* **Header** – Name · Type · Starting URL / File type · Status · Added at · Added by.
* **Data Chunks** – For crawls, every path with extracted text; for files, each section/page. Expand a chunk to preview the indexed text.

> **Tip:** Large crawls? Break them into logical folders (e.g. */docs/onboarding*, */docs/api*). Smaller chunks retrain faster and isolate failures.

***

## 4. Custom Answers 💬

This tab lets you craft curated **question → answer** pairs the AI prioritises and optionally attach CTA buttons that open posts, launch experiences, or direct users to URLs.

### 4.1 Creating a Q\&A

<figure><img src="/files/aB9jDjbJRPoTPKVTpeNF" alt=""><figcaption></figcaption></figure>

1. Click **+ Add custom answer**.
2. **Question** – The phrase or intent users will type.
3. **Answer** – Exact reply you want returned.
4. **Action** (optional) – *Open Post · Launch Experience · Navigate To URL · None*. The AI renders a CTA button when this answer is served.
5. **Save**. Jimo auto‑trains the new pair in seconds.

### 4.2 Managing Q\&As

<figure><img src="/files/CNF1uV03zhDB4INFtooa" alt=""><figcaption></figcaption></figure>

List columns: *Question · Type (Q\&A) · Status · Last updated*. Use **✏️** to edit wording or change the action; **🗑** to delete.

### 4.3 Content Details drawer

<figure><img src="/files/fe4M0BoN5MBsMXLdLk4r" alt=""><figcaption></figcaption></figure>

Click any Q\&A row to open its detail panel:

* **Header** – Question · Created at · Added by · Status.

### 4.4 How matching works

If a user query is semantically similar to the **Question** text, the AI returns your curated **Answer** first, then augments with any relevant content sources.

<figure><img src="/files/ET4q1zQhGCVdHWr53PmX" alt=""><figcaption></figcaption></figure>

***

## 5. Best practices

{% stepper %}
{% step %}
**Start broad, refine later**

Begin with a full doc site crawl; prune irrelevant sections as you analyse feedback.
{% endstep %}

{% step %}
**Avoid noisy sources**

Focus on technical and descriptive product content; leave out blogs, or customer stories  that can muddy answers.
{% endstep %}

{% step %}
**Use Custom Answers for critical compliance wording**

E.g., security disclaimers, legal statements.
{% endstep %}

{% step %}
**Set Daily retrain during rapid product changes**

Switch to Weekly once docs stabilise.
{% endstep %}

{% step %}
**Keep persona succinct**

The longer the instruction, the slower the response.
{% endstep %}
{% endstepper %}

\
\&#xNAN;*Last updated {{DATE}}*

[^1]: ![](/files/cucXfnRaOBk9PQ6LHflB)

[^2]: ![](/files/Ygfo4ObnQWtrXQcFunuE)


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://help.usejimo.com/docs/settings/ai-knowledge-outdated.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.