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

# AI Overview

> How Elementum integrates AI providers, services, and agents into your workflows

Elementum embeds AI directly into your business workflows. Rather than operating as a standalone tool, AI in Elementum connects to your data cloud infrastructure and works within the same apps, automations, and processes your teams already use. This means AI actions run on live data without copying or moving it between systems.

There are three layers to AI in Elementum: **providers** supply the models, **services** make those models available to your workflows, and **capabilities** (agents, search, and automation actions) put them to work.

## AI Providers

AI providers are authenticated connections to external model services. Each provider gives your Elementum environment access to a different set of language models and capabilities.

<CardGroup cols={2}>
  <Card title="OpenAI" icon="brain" href="/ai-agents/openai-setup">
    GPT and reasoning models for language processing, classification, and generation
  </Card>

  <Card title="Anthropic" icon="bot" href="/ai-agents/anthropic-setup">
    Direct access to Claude models for agents, automations, and Studio Agents
  </Card>

  <Card title="Snowflake Cortex" icon="snowflake" href="/ai-agents/snowflake-cortex-setup">
    AI that runs natively in your Snowflake environment with LLM functions, embeddings, and search
  </Card>

  <Card title="Google Gemini" icon="sparkles" href="/ai-agents/gemini-setup">
    Multimodal models through Vertex AI for text, image, and document processing
  </Card>

  <Card title="Amazon Bedrock" icon="aws" href="/ai-agents/bedrock-setup">
    Claude models hosted in your own AWS account (Bedrock Agents are a separate setup)
  </Card>

  <Card title="Custom Provider" icon="plug" href="/ai-agents/ai-services#configure-a-custom-provider">
    Connect any OpenAI-compatible endpoint, including LLM gateways, proxies, and self-hosted models
  </Card>

  <Card title="AI Models" icon="chart-bar" href="/ai-agents/ai-models">
    Compare supported models, capabilities, and provider-specific features
  </Card>
</CardGroup>

<Info>
  You need at least one configured AI provider before you can create AI services or use AI capabilities in your workflows. You can configure multiple providers simultaneously to use different models for different tasks, implement failover strategies, or optimize costs by routing to the most efficient provider. See [AI Models](/ai-agents/ai-models) for a detailed breakdown of available models across providers.
</Info>

<Warning>
  Only verified domain users can configure AI Providers. Elementum employees cannot create or modify AI Providers or Models in any customer org.
</Warning>

<Note>
  In addition to the built-in providers, you can connect a **Custom Provider** to bring in any OpenAI-compatible endpoint—useful for internal LLM gateways, proxies, or self-hosted models. Custom providers support **API Key** and **OAuth Client Credentials** authentication, and once configured behave like any other provider in agents and automations.
</Note>

## AI Services

AI services sit between your providers and your workflows. They define which model to use, how it behaves, and where it can be applied. There are two types:

* **LLM services** -- Configure a language model for use in automation actions, agents, and other AI-driven tasks. Each service specifies the provider, model, and default parameters.
* **Embedding services** -- Configure an embedding model for [AI Search](/ai-agents/ai-search), which converts your data into vector representations for semantic querying.

<CardGroup cols={2}>
  <Card title="AI Services" icon="settings" href="/ai-agents/ai-services">
    Create and manage LLM and embedding services for your environment
  </Card>

  <Card title="AI Search" icon="search" href="/ai-agents/ai-search">
    Enable semantic search across Elements and Tables using embedding services
  </Card>
</CardGroup>

## AI in Automations

AI actions are available as steps within Elementum's automation system. They operate on record data and return structured results that your automation can act on.

* **Classification** -- Categorize records by analyzing field content and assigning labels with confidence scores
* **Summarization** -- Generate concise summaries from long-form text fields, comments, or attached documents
* **Data transformation** -- Restructure, normalize, or enrich field values using language model processing
* **File analysis** -- Extract structured data from PDFs, invoices, contracts, and other uploaded documents

Each AI action returns results that downstream automation steps can use for routing, field updates, notifications, or further processing.

<CardGroup cols={2}>
  <Card title="AI in Automations" icon="wand-sparkles" href="/ai-agents/ai-automations">
    Learn how to configure AI-driven automation actions
  </Card>

  <Card title="Automation System" icon="bolt" href="/workflows/automation-system">
    Understand the triggers, conditions, and actions that power Elementum automations
  </Card>
</CardGroup>

## AI Agents

Agents are conversational AI components that operate within your apps. Unlike one-shot AI actions in automations, agents maintain context across a conversation, use tools to read and write data, and follow policies you define.

<CardGroup cols={2}>
  <Card title="Agents Overview" icon="robot" href="/ai-agents/agents-overview">
    What agents are, what they can do, and where they fit in your processes
  </Card>

  <Card title="Building Agents" icon="hammer" href="/ai-agents/agents-experience">
    Create, configure, and test agents in your apps
  </Card>
</CardGroup>

## Data Cloud Architecture

AI processing in Elementum runs within your data cloud environment. This architecture has several practical implications:

* **No data movement** -- AI actions query and process data where it already lives. Records stay in your data warehouse; only prompts and results move between systems.
* **Inherited security** -- AI capabilities respect your existing access controls, role-based permissions, and audit logging. No separate security layer is required.
* **Real-time data** -- AI operates on current record states, not cached snapshots. When an agent queries an Element, it reads the latest data.

## Getting Started

Setting up AI in Elementum follows a consistent sequence. Complete each layer before moving to the next.

1. **Connect a provider** -- Configure at least one AI provider ([OpenAI](/ai-agents/openai-setup), [Anthropic](/ai-agents/anthropic-setup), [Snowflake Cortex](/ai-agents/snowflake-cortex-setup), [Google Gemini](/ai-agents/gemini-setup), [Amazon Bedrock](/ai-agents/bedrock-setup), or a [Custom OpenAI-compatible endpoint](/ai-agents/ai-services#configure-a-custom-provider)) with your API credentials.
2. **Create AI services** -- Set up [LLM services](/ai-agents/ai-services) for the models you want to use in automations and agents. If you plan to use semantic search, create an [embedding service](/ai-agents/ai-services) as well.
3. **Enable AI search** *(optional)* -- Configure [AI Search](/ai-agents/ai-search) on Elements and Tables where you want natural language querying.
4. **Add AI actions to automations** -- Use [AI automation actions](/ai-agents/ai-automations) to classify, summarize, transform, or analyze data within your existing workflows.
5. **Build and deploy agents** -- Start with the [Agents Overview](/ai-agents/agents-overview), then create [agents](/ai-agents/agents-experience) with tools, skills, and policies tailored to your business processes. Deploy them in-app or across [Teams](/ai-agents/agents-teams-integration), [Slack](/ai-agents/agents-slack-integration), or [phone](/ai-agents/agents-phone-integration).

<Tip>
  Start with a single use case -- like classifying incoming requests or summarizing documents -- before building multi-step agent workflows. This lets you validate your provider configuration and service setup with minimal complexity.
</Tip>