> For the complete documentation index, see [llms.txt](https://docs.gaiodataos.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.gaiodataos.com/tools/tasks/analytics/agent.md).

# Agent

The **Flow Agent** is an AI agent designed to operate **inside task and workflow contexts**, rather than direct user conversations. In this scenario the agent acts as a **Task-based Agent**, executed as part of an automated or analytical flow.

<figure><img src="/files/x6z6x5mo6eLltkj8ItqQ" alt=""><figcaption></figcaption></figure>

## How to Access the Flow Agent

The Flow Agent is accessed through the **Tasks** panel in the left sidebar.

1. Open the left navigation menu
2. Go to **Tasks**
3. Select a task category (e.g. Analytics)
4. Click on the agent after selecting a table or access it by right-clicking on a table and searching for Agent.

A Flow Agent is used when you need AI to:

* Process data as part of a workflow
* Analyze, classify, or transform inputs
* Operate without direct end-user interaction
* Produce structured outputs for downstream steps

Unlike Chat Agents, Flow Agents are typically:

* Triggered by tasks
* Driven by prompts and inputs
* Integrated into pipelines

## How to configure your Flow Agent

<figure><img src="/files/e6dQs8uUJwjZbrLYPZ4c" alt=""><figcaption></figcaption></figure>

The **General** tab shown in this panel uses the **same configuration model** as a Chat Agent:

* Name
* Description
* Role
* Goal
* Backstory
* Tools
* RAG access

Since these concepts are already covered in the[ **Chat Agent documentation**](https://docs.gaiodataos.com/~/revisions/S2YQM2SJIf46cxt3Uy9R/tools/agents), they are **not redefined here**.

### Tools and RAG in Flow Agents

Flow Agents can access:

* Tools (Discovery, Diagrams, HTTP Request, MCP, RAG)
* Knowledge bases (RAGs)
* Schemas and diagrams

Access control follows the same rules:

* Only explicitly enabled tools can be used
* Only attached RAGs can be consulted

This ensures predictable and secure execution inside workflows.

#### Key Difference from Chat Agents

| Aspect      | Chat Agent       | Flow Agent                |
| ----------- | ---------------- | ------------------------- |
| Interaction | Conversational   | Task / pipeline-based     |
| Trigger     | User message     | Task Execution            |
| Output      | Natural language | Structured or operational |
| Usage       | Front-facing     | Backend/Analytical        |

### Prompt Configuration

The **Prompt** tab defines **how the Flow Agent receives instructions and input data** during task execution.This is the core area where you specify **what the agent should do with incoming data** and **how it should reason over it**.

In a Flow Agent, the prompt is not written for direct user interaction, but for **deterministic execution inside a workflow**.

#### Purpose of the Prompt Tab

Use the Prompt tab to:

* Define the agent’s execution instructions
* Describe the expected input structure
* Control how data is injected into the prompt
* Ensure consistent and repeatable behavior

#### Prompt Tab Sections

The Prompt tab is composed of four main elements:

1. System Prompt
   1. Optional global instructions applied before any execution.
   2. Acts as a **system-level context**
   3. Useful for high-level constraints or global behavior
   4. Optional in most Flow Agent use cases
   5. Typical use cases:
      1. Enforcing strict output rules
      2. Applying compliance or formatting constraints
      3. Defining global execution policies
2. Loop Table
   1. Defines whether the agent should execute once or iterate over a table.
3. Prompt (Instructions)
   1. This is the main instruction block for the Flow Agent.
   2. Here you describe:
      1. The agent’s role for this specific task
      2. What it should analyze or transform
      3. How it should reason over the input

### Output Configuration

The **Output** tab defines **how and where the Flow Agent stores its execution results**. This configuration turns the agent’s response into a **structured artifact** that can be consumed by downstream tasks, analytics, or storage layers.

### Purpose of the Output Tab

Use the Output tab to:

* Persist agent results in a table
* Enforce a strict output schema
* Enable deterministic, machine-readable outputs
* Integrate AI results into pipelines and analytics

#### Output Destination

1. Save agent result at
   1. Choose where the agent output will be stored on a table or a parameter.
2. Table name
   1. Define the table where results will be written.
3. Select how the agent output will be stored as a text or JSON.
4. Insert mode: append records to table (optional)
5. Define output schema format

```json
 {
  "type": "object",
  "properties": {
    "name": { "type": "string" },
    "email": { "type": "string" },
    "score": { "type": "integer" },
    "description": { "type": "string" },
    "date": { "type": "string" },
    "category_1": { "type": "string" },
    "category_1_score": { "type": "number" }
  }
}

```

This schema ensures that:

* Every output follows the same structure
* Data types are enforced
* AI responses are compatible with analytics and storage

{% hint style="info" %}
Click **Generate** to auto-create a schema based on context.
{% endhint %}

## When to Use a Flow Agent

Use a Flow Agent when:

* AI is part of a data or automation pipeline
* Outputs must be deterministic or structured
* The agent supports downstream tasks
* No direct chat interface is required

## Best Practices for Flow Agent Outputs

* Always define an output schema for production workflows
* Prefer JSON for complex AI results
* Align output fields with downstream consumers
* Avoid free-text outputs in automated pipelines
* Version schemas when making structural changes