# Data & Knowledge (Reference: https://docs.iqra.bot/build/knowledge)

An AI Agent is only as good as the information it has access to. In Iqra AI, we separate data into distinct categories to optimize for **Accuracy**, **Latency**, and **Token Usage**.

<Callout type="info" title="Global vs. Session Data">
  This section covers **Global Data** (Facts about your business that rarely change).

  If you are looking to store data specific to a *single phone call* (like the user's name or booking ID), use **[Script Variables](/build/script/variables)** instead.
</Callout>

Data Layers [#data-layers]

1. Business Context (Structured) [#1-business-context-structured]

**The Hard Facts.** Highly structured entities defined in the dashboard.

* **Examples:** Branch Locations, Opening Hours, Product Catalog.
* **Strategy:** "Summary Injection." The agent gets a list of product names in its prompt (low cost), but uses a tool to fetch the full specs only when asked (high detail).

2. Knowledge Base (Unstructured) [#2-knowledge-base-unstructured]

**The Library.** Documents ingested via a **RAG (Retrieval Augmented Generation)** pipeline.

* **Examples:** PDF Manuals, Policy Docs, FAQs.
* **Strategy:** "Vector Search." The system chunks your documents and retrieves only the relevant paragraphs based on the user's query.

3. Cache (Performance) [#3-cache-performance]

**The Short-Term Memory.** Storing generated outputs to save time and money.

* **Examples:** TTS Audio files, LLM Responses, Vector Embeddings.
* **Strategy:** "Exact Match." If the system has seen this input before, it skips the AI generation entirely.

***

Workspace Modules [#workspace-modules]

<Cards>
  <Card icon={<Store />} title="Context Manager" href="/build/knowledge/context">
    **Structured Data.** Define your Branding, Branches, Services, and Products. Control the injection strategy (Prompt vs. Tool).
  </Card>

  <Card icon={<Database />} title="Knowledge Base (RAG)" href="/build/knowledge/rag">
    **Unstructured Data.** Configure vector stores, advanced chunking (Parent-Child), and search triggers.
  </Card>

  <Card icon={<Zap />} title="Caching" href="/build/knowledge/cache">
    **Performance.** Manage cache groups to reduce latency and API costs.
  </Card>
</Cards>

Data Flow Architecture [#data-flow-architecture]

How data moves from storage to the agent's brain:

<Mermaid
  chart="graph TD
    subgraph &#x22;Persistent Storage&#x22;
        Ctx[&#x22;Context (Products/Branches)&#x22;]
        KB[&#x22;Knowledge Base (Docs)&#x22;]
    end

    subgraph &#x22;Cache Layer (Redis/S3)&#x22;
        EmbedCache[Embedding Cache]
        MsgCache[Message Cache]
        AudioCache[&#x22;Audio Cache (TTS)&#x22;]
    end

    subgraph &#x22;Agent Runtime&#x22;
        Prompt[System Prompt]
        Tool[Internal Tools]
        LLM[LLM Inference]
        Speaker[Audio Output]
    end

    %% Context Flow
    Ctx -->|Summary Injection| Prompt
    Ctx -->|Detailed Lookup| Tool

    %% RAG Flow
    KB -->|Vector Search| Tool
    Tool -.->|Check| EmbedCache

    %% Inference Flow
    Prompt --> LLM
    LLM -.->|Check| MsgCache
    
    %% Audio Flow
    LLM -->|Text Stream| Speaker
    Speaker -.->|Check| AudioCache
    
    style Ctx fill:#2563eb,stroke:#333,stroke-width:2px,color:#fff
    style KB fill:#2563eb,stroke:#333,stroke-width:2px,color:#fff
    style AudioCache fill:#16a34a,stroke:#333,stroke-width:2px,color:#fff"
/>