> Documentation index: [/llms.txt](/llms.txt). Every page as one file: [/llms-full.txt](/llms-full.txt). # Tools Tools [#tools] Tools are what an agent uses to act: search the web, run code, read your files, query a connector. The agent decides when to use them; you see each step in its reply. See what the agent did [#see-what-the-agent-did] While the agent works, its steps show in a box above the reply: thinking, then each tool it uses. When it finishes, the box collapses to one line, such as **Thought for 5s · Ran code … · +2 more**. Select it to see every step, and select a step to see its input and result. The steps of a reply expanded: thoughts, reading the uploaded file, reading the xlsx skill, running code and listing sandbox files, each tool step marked done A web search step shows what the agent searched for and links to each result. Tools every agent has [#tools-every-agent-has] | Tool | What it does | Needs on the deployment | | --------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------- | | Run code | Runs Python in the agent's own sandbox, a private computer for this chat. Charts, documents and other files it saves come back as files in the reply. | The code sandbox | | Shell and files | Runs shell commands, and reads, writes, edits and searches files in the sandbox. | The code sandbox | | Ask you | Asks you multiple-choice questions. See [Plan mode, questions and approvals](/docs/platform/chat/plan-mode-and-approvals). | | | Schedule a task | Sets up a scheduled task after you approve it. See [Schedule an agent](/docs/platform/scheduler/scheduled-tasks). | The scheduler | Every agent can also open the files you attach, and an agent with skills reads them when it needs them. See [Work with files in a chat](/docs/platform/chat/files-and-attachments) and [Skills](/docs/platform/skills). Tools an agent can be given [#tools-an-agent-can-be-given] A custom agent has the tools its builder turned on; see [Configure an agent](/docs/platform/agents/configure). The editor's tool list offers the four web tools, **Generate Image**, **Analyze Video**, **Progress Checklist** and **Extract Tables**. **Call Agent** comes with sub-agents, the connector tools with the agent's connections, and the other Library tools with its knowledge. | Tool | What it does | Needs on the deployment | | ------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------- | ---------------------------------- | | **Web Search** | Searches the web and returns titles, links and snippets. | A web search service | | **Web Extract** | Reads the full text of pages it has links to. | A web search service | | **Web Crawl** | Reads the pages of a site, starting from one address. | A web search service | | **Web Map** | Lists the pages of a site without reading them. | A web search service | | **Generate Image** | Creates, edits or combines images. | Google Gemini and the code sandbox | | **Analyze Video** | Summarizes or answers questions about a YouTube video; long videos are analysed in parts. | Google Gemini | | **Call Agent** | Hands a self-contained task to another agent and gets its report back. See [Sub-agents](/docs/platform/agents/sub-agents). | | | **Progress Checklist** | Shows a checklist of the steps the agent is working through, updated in place. | | | **Connector Metadata**, **Query Data Source** | Reads a connection's structure, and runs queries or actions on it. | Connectors | | **Semantic Search**, **List Files**, **Get File URL**, **Extract Tables** | Search, list and open Library files, and pull tables out of documents. | The Library | What Alo uses in a chat [#what-alo-uses-in-a-chat] Alo, the default agent, comes with run code, the sandbox tools, **Web Search**, **Generate Image**, **Analyze Video** and **Call Agent**. The composer adds tools as you go: | You | Alo also gets | | ---------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------- | | Send any message | **Web Extract** | | Turn on a connector in **+** > **Connectors**, or use one that comes with the agent | **Connector Metadata** and **Query Data Source** | | Select **Search in files**, or mention a Library file or folder with **@** | **Semantic Search**, **List Files**, **Get File URL** and **Extract Tables** | A custom agent keeps the tools it was built with, plus the connector and Library tools the same way. Limits [#limits] * An action that changes data in a connector, sends something or creates a scheduled task asks for your approval first. See [Approval cards](/docs/platform/chat/plan-mode-and-approvals#approval-cards). * In Plan mode the agent's built-in tools are limited to the ones that read; tools from MCP servers stay available. See [Plan and Act modes](/docs/platform/chat/plan-mode-and-approvals#plan-and-act-modes). * When your deployment lacks a service a tool needs, the tool tells the agent it is unavailable instead of running. Ask your administrator. * Apps are built by the App Builder agent with its own tools. See [Build an app](/docs/platform/apps/build-an-app).