# Built-in tools

> What every agent can do with no setup — and which of those capabilities you can switch off per agent.

**Built-in tools** are the capabilities an agent has before you connect anything. They are grouped, and each group can be switched off for an individual agent.

## When to use it

You rarely turn these *on* — they are on by default. You turn them *off* to narrow an agent. An agent that should only summarize what you paste into it does not need to fetch web pages; an agent that should never schedule anything does not need routine tools.

## The groups

| Group | What the agent can do |
| --- | --- |
| **Artifacts** | Save, list, read, update, and delete [artifacts](/docs/capabilities/artifacts/) |
| **Routines** | Create and manage [routines](/docs/automate/routines/) from inside a conversation |
| **Memory** | Remember, recall, update, and forget — see [Memory](/docs/capabilities/memory/) |
| **Projects** | Read and update the [project](/docs/capabilities/projects/) a conversation belongs to |
| **Skill tools** | Look through installed [skills](/docs/capabilities/skills/), use one, or author a new one |
| **Visualization** | Draw a chart or widget inline in the conversation |
| **Web fetch** | Read the contents of a URL |
| **Utility** | Precise arithmetic and the current time |

Turning off **Memory** also stops the agent's memory index being loaded at the start of a conversation — the group is the whole capability, not just the tools.

<Screenshot
  name="studio-tools"
  alt="The built-in tools section: each group as a card with a switch, the number of tools it contains, and a warning where turning it off does more than remove tools."
/>

## Always on

A few tools are infrastructure for the conversation itself and cannot be turned off:

- Asking you a [question](/docs/work/questions/) — options or free text.
- Requesting a [skill](/docs/capabilities/skills/#when-an-agent-asks-for-a-skill) it has not been given.

An agent that could not ask you anything would guess instead, which is worse in every case.

<Aside type="tip" title="Turning tools off is a tuning technique">
  Every advertised tool is one more thing the model weighs on each step. Removing
  groups an agent has no business using measurably sharpens its choices — this is
  one of the cheaper fixes on the [iteration list](/docs/build/iterating/#what-to-change-in-order).
</Aside>

## Limits

- Built-in tools are per agent. Turning a group off for one agent does not affect others.
- Turning a group off applies to the agent's next [run](/docs/work/runs/).
- Built-in tools do not reach your accounts. Anything touching Gmail, Slack, Notion, or another service comes from a [connector](/docs/connect/overview/), not from here.
- Web fetch reads pages that are publicly reachable. It is not a browser and does not sign in anywhere.

## Related

<CardGrid>
  <LinkCard
    title="Skills"
    href="/docs/capabilities/skills/"
    description="Capabilities you add, rather than ones that ship on."
  />
  <LinkCard
    title="Connectors overview"
    href="/docs/connect/overview/"
    description="Tools that reach your actual accounts."
  />
  <LinkCard
    title="Testing and iterating"
    href="/docs/build/iterating/"
    description="Where narrowing an agent's tools fits in tuning it."
  />
</CardGrid>
