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

# Available tools

> What the WednesdayAI agent can do for you: web search, code, files, browser, images, PDFs, memory, scheduling, and device actions.

# Available tools

WednesdayAI agents can use **tools** to take actions on your behalf. Tools run automatically - the agent decides when it needs one based on what you ask. You do not have to name a tool, though you can ask for one explicitly.

<Note>
  Which tools are available depends on how your administrator has configured the gateway. Some tools (shell execution, browser control, device actions) require explicit opt-in, and a few need an approval before they run.
</Note>

## What the agent can do

### Search the web

Find current information using **Brave Search**.

> "What's the current price of NVDA stock?"

### Read a web page

Fetch a URL and pull out the readable content.

> "Read this article and give me the three main points: \[paste a link]"

### Run code and shell commands

Write and run code in its workspace and show you the output.

> "Write a Python script to parse this CSV and find the top 10 values, then run it."

<Note>
  Running commands on the gateway *host* (outside the workspace) is a separate, locked-down capability that is off by default.
</Note>

### Read and write files

Work with files in its workspace - read them, edit them, create new ones.

> "Create a Markdown file with the outline we just discussed."

### Control a browser

Open pages, click and type, fill forms, take screenshots, and extract data from sites that need a real browser.

> "Go to example.com and grab the pricing table."

### Analyse an image

Look at a photo or screenshot and describe or interpret it.

> (attach a photo) "What plant is this, and is it healthy?"

### Analyse a PDF

Read one or more PDFs (up to 10 at a time) and answer questions or compare them.

> (attach a report) "Summarise this in five bullet points."

See [Sharing files](/users/files) for how to send images and PDFs on each channel.

### Remember things

Save durable facts and preferences that survive across conversations, and recall them later.

> "Remember that my project deadline is the 14th."
>
> "What did I tell you about my deadline?"

See [Memory](/users/memory).

### Schedule tasks and reminders

Set things to run later or on a repeat.

> "Remind me in two hours to follow up on the build."
>
> "Every weekday at 9am, give me a summary of overnight emails."

### Message other chats

Send a message into another conversation or channel it has access to, post polls, react, and manage threads.

> "Send a message to the team channel saying the deploy is done."

### Spawn a sub-task

Hand a self-contained job to a background sub-agent and report back when done - useful for longer pieces of work.

> "Spin up a sub-task to research three hosting options and report back."

### Control a paired phone or Mac

If a phone or Mac is paired as a **node**, the agent can take device actions on it: send a notification, run a command, capture a camera photo or a screen recording, or read the device's location.

> "Take a photo with the office Mac's camera."

<Note>
  Device actions need an explicit node to be paired and usually need the device app in the foreground. Camera and screen capture always respect your consent.
</Note>

## When a tool needs approval

If a tool is configured to ask first (for example shell execution with approvals on), the agent pauses and asks before running. Approve it with:

```
/approve <id> allow-once
```

To see the current execution policy, send:

```
/exec
```

## Seeing what a tool did

By default the agent summarises tool results in its reply. To see the full output:

```
/verbose on
```

Turn it off when you are done:

```
/verbose off
```

<Tip>
  Keep `/verbose` off in group chats - it can reveal tool output you did not mean to share.
</Tip>

## Asking for a specific approach

You can steer which tool or approach the agent uses:

> "Search the web for this - don't rely on your training data."
>
> "Use the browser to check that URL for me."

The agent will use the tool if it is available and you are permitted.