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

# Agents

> An agent is the core unit in AskTable—connect it to data sources and skills, and it answers data questions in natural language. This page covers what makes up an agent and how it relates to skills, channels, and scheduled tasks.

## What is an agent

An agent is a virtual data analysis assistant you configure in AskTable. Connect it to data sources and attach skills, and it can answer data questions in natural language — understanding your question, querying the data, and giving you a conclusion.

The agent is the core unit you interact with. **Skills**, **channels**, and **scheduled tasks** — covered next — all revolve around the agent.

## What an agent can do

* Understand natural-language questions and locate the relevant tables and fields
* Generate and run SQL to query data
* Run Python for complex calculations and data processing
* Turn results into charts
* Ask you a follow-up question when something is unclear
* Search the web for public data (when enabled)
* Remember key information across conversations (when enabled)

Which abilities are on is controlled by the agent's **Capabilities**, described below.

## What makes up an agent

| Part             | Description                                                                                                                                      |
| ---------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Data sources** | Which databases and tables the agent can access. Without a data source, there's nothing to query.                                                |
| **Skills**       | Reusable analysis methods that teach the agent how to analyze (such as anomaly detection or comparison). See [Skills](/en/concepts/agent/skill). |
| **Capabilities** | A set of toggles: Python, charts, clarification, and web search. The first three are on by default; web search is off by default.                |
| **Memory**       | When enabled, the agent remembers preferences and conclusions across conversations and brings them into later analysis.                          |
| **Permissions**  | When enabled, only allowlisted members can use the agent, with optional row-level data isolation by role.                                        |
| **Avatar**       | The agent's display icon.                                                                                                                        |

## Create from a template

AskTable ships with 9 agent templates, each preloaded with a set of skills suited to a scenario — for example: Retail Store Analyst, E-commerce Monitor, Financial Analyst, Market Insights Analyst, User Growth Analyst, and Executive Assistant.

Creating from a template prefills the name, description, and skills — you just pick your own data sources. You can also start from **Blank**. Templates are read-only; the agent you create from one is fully editable.

## How skills, channels, scheduled tasks, and embedding relate

Four companion concepts revolve around the agent:

<Columns cols={2}>
  <Card title="Skills" icon="puzzle" href="/en/concepts/agent/skill">
    Give the agent reusable analysis methods.
  </Card>

  <Card title="Channels" icon="message-square" href="/en/concepts/agent/channel">
    Let users talk to the agent straight from IM tools like Feishu.
  </Card>

  <Card title="Scheduled tasks" icon="clock" href="/en/concepts/agent/schedule">
    Run the agent on a schedule and push the results.
  </Card>

  <Card title="Embedding" icon="code" href="/en/concepts/agent/embed">
    Drop the agent into your own website as an iframe for anonymous visitors.
  </Card>
</Columns>

In short: **skills** decide what the agent knows, **channels** and **embedding** decide where users reach it, and **scheduled tasks** make it work on its own.

## Frequently asked questions

**Q: Can an agent use multiple data sources?**
A: Yes. An agent can access all the data sources you configure for it.

**Q: Can I use an agent without skills?**
A: Yes. Skills are an enhancement. Without them the agent can still query data, draw charts, and run Python — it just lacks the preset analysis methods.

**Q: Can I edit a template directly?**
A: Templates are a read-only starting point. Once you create your own agent from one, the name, data sources, skills, and capabilities are all adjustable.