> ## Documentation Index > Fetch the complete documentation index at: https://developers.nlpearl.ai/llms.txt > Use this file to discover all available pages before exploring further. # Agent Knowledge Base Before your Pearl can answer questions accurately, it needs context. This step helps define the essential background information your agent will rely on during conversations. **Company Name** Give your Pearl the name of the company it represents. This name may be used when Pearl introduces itself or provides company-related information.

Send Email action interface on light mode

Send Email action interface on dark mode

*** **Company Description** Add a brief company description including your industry, mission, or key services. This allows Pearl to answer questions with more contextual relevance.

*** **Pearl Knowledge Base** This is where you teach Pearl what it needs to know to respond with relevance and accuracy. Add detailed information about your services, pricing, opening hours, locations, and any key operational data. You can either write this content manually or upload a reference file.\ Multiple documents can be added if needed. Supported formats: `.doc`, `.docx`, `.txt`, `.pdf`. Whether you're describing how your subscription plans work, listing available services, or clarifying your business hours, this section allows Pearl to access structured information that supports more precise answers.

*** **Memory** When enabled, Pearl remembers the user or lead, allowing it to seamlessly continue conversations from where they left off. This feature is particularly useful for: * Multi-step processes * Follow-ups * Any scenario where continuity matters ### Example use case: Hospitality Let's say a guest checks into a hotel. Pearl can: * Remember their room number * Greet them by name * Recall their preferences (like "extra towels" or "no wake-up call") * Offer a smooth, personalized experience every time they call **Resetting Memory** Need a fresh start? Pearl's memory can be reset via API. This is especially important in contexts like hospitality, where once a guest checks out, Pearl clears the slate, ensuring the next visitor starts with a clean experience.

*** **Speech Recognition Keywords** Some words are just… not in the dictionary. Whether you're using brand-specific terms, creative product names, or industry jargon, teaching Pearl how to recognize them makes all the difference. Use this feature to define: * **Keywords**: The unique word or name you want Pearl to catch * **Pronunciations**: Variations of how users might say it (spoken form) ### Example Let's say your bakery sells a pastry called `FluffoBun`. Here's how you'd define it: * **Keyword:** FluffoBun * **Pronunciations:** fluh-foh-bun, floof-oh-bun, fluff-bunb\ *(Separate each pronunciation with a comma)* This helps Pearl match even creatively named terms to real user input, no matter how it's pronounced.

*** **Speech Output Pronunciations** Speech Output Pronunciations let you fine-tune how Pearl *pronounces* specific words or short phrases out loud. You provide the text as it appears (**Pearl says**) and a simple phonetic respelling of how it should actually sound (**Caller hears**). **What it does** Whenever Pearl is about to speak a word or phrase listed in **Pearl says**, it pronounces it using the **Caller hears** respelling instead. It changes the **audio only**: your scripts, variables, and node text stay exactly as you wrote them. **When to use it** Reach for it whenever text-to-speech mispronounces something important, such as: * **Brand and product names**: `Nike` → `ny-kee` * **Acronyms** you want spelled out: `NLPearl` → `N-L-Pearl` * **Industry terms**: `SQL` → `sequel` * **Abbreviations** you want expanded: `Dr.` → `Doctor` * **Foreign or uncommon words** that need a specific phonetic reading **How to configure** In the Flow Editor, open the **Knowledge Base** and scroll to **Speech Output Pronunciations**. In the empty row at the bottom, type the text in **Pearl says**, then the phonetic respelling in **Caller hears**. Use plain, readable syllables separated by hyphens (e.g. `ny-kee`, `kwa-san`). No special phonetic alphabet is required, just write it the way it should sound. Use the trash icon on a row to delete it. Write **Caller hears** the way it should be *spoken*, not spelled, e.g. `2024` → `twenty twenty-four`, or `NLPearl` → `N L Pearl`. Keep it simple and phonetic. **Each row must have both fields filled**: a **Pearl says** value *and* a **Caller hears** value. Otherwise the row is flagged as incomplete. Empty rows are ignored and dropped when you save. **Limits:** up to **25** entries, and each field is capped at **50 characters**. **Examples** | Pearl says | Caller hears | Why | | ---------- | ------------------ | ----------------------- | | Nike | ny-kee | Brand name | | NLPearl | N-L-Pearl | Spell out the name | | SQL | sequel | Preferred pronunciation | | Dr. | Doctor | Expand the abbreviation | | 2024 | twenty twenty-four | Read the year naturally | **Speech Output Pronunciations vs. Speech Recognition Keywords** These two Knowledge Base tools are complementary, so don't mix them up: * **Speech Output Pronunciations** control how Pearl **speaks** (text-to-speech / output). Fields: **Pearl says** → **Caller hears**. * **Speech Recognition Keywords** help Pearl **understand** what the caller says (speech-to-text / input). Fields: a **keyword** plus one or more **pronunciations** it might be heard as.