# Agents

> Agents are AI assistants that use tools.

**URL:** https://inference.sh/docs/concepts/agents
**Last updated:** 2026-10-11

---

Agents are AI assistants that use tools.

---

## What makes an agent?

| Component | Purpose |
|-----------|---------|
| **LLM** | The AI brain (GPT-4, Claude, etc.) |
| **System Prompt** | Personality and instructions |
| **Tools** | Apps and flows it can use |
| **Sub-Agents** | Other agents it can delegate to |

---

## How agents work

1. You send a message
2. Agent decides what to do
3. Agent calls tools if needed
4. Agent responds with results

---

## Built-in capabilities

Every agent has these (not editable):

| Tool | Purpose |
|------|---------|
| **memory** | Scratch notes for the current chat only (use knowledge for durable facts) |
| **plan** | Break down complex tasks |
| **finish** | Mark task as complete |

---

## Example

```
You: "Transcribe this audio and summarize it"

Agent: 
  1. Calls whisper tool → gets text
  2. Calls summarize tool → gets summary
  3. Returns both to you
```

---

## Learn more

| Topic | Guide |
|-------|--------|
| Create and configure agents | [Creating an Agent](/docs/agents/creating) |
| Delegate to specialists | [Sub-Agents](/docs/agents/sub-agents) |
| Parallel tool calls in one turn | [Sub-Agents: parallel execution](/docs/agents/sub-agents#parallel-execution) |
| Typed JSON from `finish` | [Structured Output](/docs/api/agent/structured-output) |
| Chat via API | [Agents REST API](/docs/api/rest/agents) |

---

## Next

→ [Workers](/docs/concepts/workers)