Table of contents

Just for you

Building AI chatbots with RAG: a step-by-step guide

5 benefits of retrieval-augmented generation (RAG)

Agentic RAG: definition, benefits, and real-world examples

Jon Gitlin

Senior Content Marketing Manager

@Merge

Retrieval-augmented generation (RAG) allows you to receive accurate answers to business-specific questions.

You can ask an AI assistant about your team’s upcoming onsite, for example, and the AI assistant can leverage a document on the onsite in your file storage system to generate an accurate response.

Screenshot of RAG use case for finding location of team onsite — *Guru's enterprise AI search platform can use RAG to generate the output above*

But what if you also wanted to take an action, like declining or accepting the offsite invitation based on the office it takes place in?

That’s where agentic RAG comes in.

Read on to learn how agentic RAG works, why it’s becoming increasingly valuable, and how real-world platforms support it.

Agentic RAG overview

Agentic RAG is when an AI agent uses a RAG pipeline to perform a specific action or set of actions on your behalf.

For example, a traditional RAG pipeline works as follows:

1. A user asks a question (e.g., “Give me the marketing team’s first names and addresses?”).

2. This question gets embedded (or converted into a vector representation) and a large language model (LLM) uses it to search for semantically-similar embeddings in a vector database.

3. The LLM can then use the embeddings it identified as context for generating an output.

*How a RAG pipeline can work when a user wants to find colleagues in the marketing team*

Agentic RAG will go a step further: Based on the user’s role and department, the AI agent can offer to take specific actions, like asking marketing for an update on a project that’s relevant to the user or adding these marketing colleagues to an upcoming meeting that requires their support.

Agentic RAG can also involve asking an AI agent to take specific actions directly.

For instance, a user can ask the AI agent to share all the tickets marked as “Urgent” in a specific Slack channel. This requires the agent to—using RAG—identify these tickets in the project management system before going on to share them in the designated Slack channel.

https://www.merge.dev/blog/rag-vs-mcp/?blog-related=image

Benefits of agentic RAG

As you can probably tell from our examples, agentic RAG offers several advantages over a traditional RAG workflow.

Automates tedious and mundane tasks

Agentic RAG lets your employees avoid time consuming and unpleasant work, such as sharing updates with colleagues, setting reminders to perform certain tasks, assigning a ticket to a team member, etc. This frees up your employees and allows them to perform the work they actually enjoy and that’s more impactful for the business.

Provides secure data access

While it depends on your implementation, you can take measures to ensure the AI agent keeps sensitive data secure and private. For example, the AI agent can request data in 3rd-party systems via their API endpoints to ensure users have the proper permissions to access and interact with the data.

Supports complex workflows

Agentic RAG doesn’t just support the boring and simple parts of your employees’ jobs. They can also handle multi-step, multi-application processes that rely on business logic.

You can even use agentic RAG to determine workflows for specific scenarios.

For example, you can set up a workflow where, depending on a ticket’s level of importance, the customer(s) that's impacted, and the category it falls under, the AI agent can determine a tailor-made remediation process and go on to execute it.

Uses several data sources

AI agents can access your internal or customer data in a variety of ways, whether that’s via API endpoints, tools exposed from a Model Content Protocol (MCP) server, flat files downloaded from an application, and so on.

Since data access isn’t a blocker for AI agents, they can support a wide range of RAG scenarios.

Examples of agentic RAG

Several 3rd-party solutions already let you implement internal or customer-facing agentic RAG use cases relatively easily.

Here are some examples.

Ema’s Employee Assistant AI agent

Ema, a universal AI employee solution, offers an “Employee Assistant AI” agent that lets users make all kinds of requests, such as submitting a vacation request.

Once the AI agent receives a vacation request, it'd retrieve that employee’s vacation balance by making the relevant API request in the customer’s HR software.

Assuming the employee has enough days remaining in their PTO balance, the AI agent can make the request directly in the HR software via a separate API request. And once that's successful, the agent would send the employee a confirmation message.

https://www.merge.dev/blog/rag-tools?blog-related=image

Peoplelogic’s Team Growth Coordinator AI agent (“Noah”)

Peoplelogic, a leader in AI for HR, offers an AI agent, Noah, that can take admin tasks off of recruiters’ plates.

For example, a recruiter can ask the AI agent within Slack to share information on candidates with interviewers on the day of their interviews. These messages can include details like the time the interviews are taking place, the roles candidates are interviewing for, summaries on the candidates’ backgrounds, and more.

Noah would then retrieve the candidates’ interview schedules, profile information, attachments, and more from the integrated ATSs to generate timely messages for interviewers.

Screenshot of Peoplelogic's Noah agent sending interviewers messages

Juicebox’s “Juicebox Agents”

Juicebox, an AI recruiting platform, offers Juicebox Agents to, among other things, help recruiters identify relevant candidates, quickly.

These agents can use job descriptions from the open roles in customers’ integrated ATSs and public sources—professional profiles, websites, published papers, etc.—as context for generating thousands of relevant candidates for given role.

How to kickstart agentic RAG use case — *Once a user clicks on a role, the Juicebox Agent initiates its agentic RAG workflow for sourcing high-fit candidates*

Leverage agentic RAG in your product with ease through Merge

Merge, the leading unified API solution, lets your AI agent access hundreds of customer-facing applications (via tools) through Merge’s Model Context Protocol (MCP) server.

Merge also:

Supports enterprise-grade security by using access control lists (ACLs) and encrypting data at rest and in transit
Provides observability tooling to help your customer-facing teams manage integration issues with ease
Lets you use Merge MCP by writing just a few lines of code (as shown below)

{
  "mcpServers": {
    "merge-mcp": {
      "command": "uvx",
      "args": ["merge-mcp"],
      "env": {
        "MERGE_API_KEY": "xxxxxxx",
        "MERGE_ACCOUNT_TOKEN": "xxxxxxx"
      }
    }
  }
}

Learn more about Merge MCP by scheduling a demo with one of our integration experts.

Jon Gitlin

Senior Content Marketing Manager

@Merge

Jon Gitlin is the Managing Editor of Merge's blog. He has several years of experience in the integration and automation space; before Merge, he worked at Workato, an integration platform as a service (iPaaS) solution, where he also managed the company's blog. In his free time he loves to watch soccer matches, go on long runs in parks, and explore local restaurants.