Skip to Content
Docs are being rebuilt — start at Introduction → How it works.
IntroductionCore conceptsTool & capability

Tool & capability

These two words mean similar things — and most agent builders never need to know the difference. Here’s the difference anyway, so the rest of the docs make sense.

Tool

A tool is one specific action an agent can take. Tools have names like search_web, read_file, send_email, create_artifact, execute_command, make_agent_call. When you read in an agent’s trace “called search_web with query ‘qatar labour law leave entitlement’”, search_web is the tool.

There are around 80 tools shipped with TeamMate. They span:

  • Files & code — create_file, read_file, str_replace, execute_command, execute_code.
  • Web & research — search_web, search_tavily, scrape_webpage, browser_navigate, browser_extract.
  • Communication — send_email_tool, plus the connected-app tools (Slack, Teams, Gmail, Outlook).
  • Data — query_dataset, create_dataset, create_data_analytics_plan.
  • Knowledge — search_knowledge, write_knowledge.
  • Visualization — chart_generator_tool, pdf_generator.
  • Delegation — make_agent_call, search_agents, load_agent.
  • Discovery (SuperAgent only) — search_tools, load_tools, search_skills, load_skill.

You don’t pick tools one at a time. You pick capabilities, and TeamMate injects the right tool set.

Capability

A capability is the higher-level switch. In the agent builder’s Capabilities tab, you’ll see toggles for Web Research, Files & Code Management, Email, Knowledge, Database & Datasets, and so on. Each toggle controls a group of tools — turning on Web Research gives the agent search_web, search_tavily, scrape_webpage, and the browser tools.

Capabilities exist because:

  • Fewer choices. Most builders don’t want to opt into individual tools.
  • Capability-aware routing. When a capability is on, TeamMate picks the right underlying tool for the model. If the agent’s model can read PDFs inline, the file-attachment capability does nothing extra. If the model can’t, the same capability injects read_attachments.

When you’d ever care about specific tools

  • You’re debugging an agent’s trace and want to know what each call did.
  • You’re using the SuperAgent’s search_tools / load_tools meta-tools to expand its toolset on the fly. These work at the tool level, not the capability level.
  • You’re writing a skill that needs a specific tool to be loaded first (“call load_tools({ ids: ['create_skill'] }) before continuing”).

Tool vs Skill

People mix these up:

  • A tool is programmatic. It runs code, hits an API, reads a file. The agent calls it.
  • A skill is textual. It’s a written playbook the agent reads and follows.

A skill can use tools. The skill-creator skill, for example, tells the agent how to use the create_skill tool. The skill is the recipe; the tools are the ingredients.

How tools and capabilities show up in the UI

  • Agent builder → Capabilities tab — turn capabilities on or off for one agent.
  • Agent builder → Skills tab — attach skills (which may declare required tools).
  • Chat trace — when a reply involves a tool call, you see the tool name, its inputs, and its outputs inline.
SkillKnowledge base