Tools
Tools are the external services your agents can use when executing responsibilities. They represent the actions an agent can take — sending emails, posting Slack messages, creating Linear issues, searching the web, and more.
Browsing tools
Navigate to Tools from the sidebar to see all available tools in your workspace. Tools are displayed in two groups:
- Connected — Tools that are authenticated and ready to use. These show a checkmark and display which agents and responsibilities use them.
- Available — Tools that require a connection before they can be used. Click Connect to set up authentication.
Tool properties
Each tool has:
- Name and description — What the tool does.
- Authentication — Whether it requires a personal connection (per-user OAuth) or an organization-level connection (shared).
- Capabilities — Whether it supports local processing, cloud processing, or both.
- Event triggers — Whether it can emit events that trigger responsibilities.
- Risk level — A badge indicating the potential impact of the tool’s actions (e.g., read-only vs. write operations).
- Connection status — Whether the tool is currently connected and authenticated.
- Related skills — Skills that reference this tool in their allowed tools list, so you can see what skills depend on it.
Assigning tools to steps
Tools are enabled or disabled at the step level within each responsibility. When configuring a step, select which tools the agent can use for that specific step. You don’t need to enable all tools — only the ones relevant to what the step does.
Each agent also has a Security tab (in its settings) showing a read-only audit of all tool permissions aggregated across all its responsibilities, grouped by toolkit with approval level indicators.
Approval gates
Tool approval is configured per step within each responsibility. When editing a step, you can enable or disable specific tools and set an approval gate for each one.
For each tool, the approval options are:
- Auto-approve — The agent uses the tool freely.
- Any team member — Any member of the workspace can approve.
- Specific users — Only designated people can approve.
This lets you give agents read access to services while requiring approval for write actions. For example, an agent might read Gmail freely but require approval before sending a message.
Tool merging by skills
When a skill is referenced in a step’s instructions, its allowed tools are automatically merged into that step’s tool configuration. This means the tools the skill depends on become available to the step without needing to enable them manually. If you remove the skill reference from the step, those tools revert to their original step configuration.
Tool usage
The tool detail page shows which agents and responsibilities depend on that tool, making it easy to understand the impact before disconnecting an integration. It also cross-references which skills include this tool in their allowed tools configuration.