Skip to main content

Agentic Capabilities

Built-in Agent Tools

Portal One agents come equipped with a powerful set of built-in tools that execute directly in the agent's sandbox environment. These tools are auto-approved — no manual review needed — because they run in complete isolation.

Available Tools

Terminal

Execute shell commands in a full Linux environment. Agents can run scripts, install packages, execute builds, and perform any command-line operation.

File Operations

  • Read files — read any text file with line numbers and pagination
  • Write files — create or overwrite files with content
  • Search files — search file contents (like grep/ripgrep) or find files by name
  • Patch files — targeted find-and-replace edits with fuzzy matching

Web Tools

  • Web search — search the internet for information on any topic
  • Web extract — extract content from URLs as markdown (also works with PDFs)

Code Execution

Run Python scripts with full access to the tool ecosystem. Agents can write scripts that call other tools programmatically for complex multi-step workflows.

Enabling and Disabling Tools

Each built-in tool can be toggled on or off in your agent's configuration. Navigate to your agent's settings and use the tool toggles to customize which capabilities are available.

By default, all built-in tools are enabled for new agents.

How Tools Differ from MCP

Built-in tools and MCP tools serve different purposes:

  • Built-in tools are general-purpose capabilities (coding, files, web) that run in the agent's sandbox. They're auto-approved and fast.
  • MCP tools connect to external services and APIs (Gmail, databases, Slack, etc.). They go through your approval workflow because they affect external systems.

Most agents will use both — built-in tools for thinking and working, MCP tools for interacting with the outside world.