Skip to main content
TripleBits Apprentice - Software for Agents

Menu

Sign In Register

Configure Agent Capabilities

Use Apprentice agent capabilities to enable or block broad tool families such as filesystem, shell, browser, network, memory, tasks, messaging, scheduler, notifications, and external MCP tools.

20 views

Configure Agent Capabilities

Capabilities are the broad tool-family switches for an Apprentice agent. Use them when you want to say "this agent may never use this class of tool," even if a permission mode or saved approval would otherwise allow a tool call.

Capabilities are separate from folder access and permission prompts:

  • Directories decide what host folders the agent can see.
  • Permissions decide when Apprentice should ask, allow, or deny individual tool requests.
  • Capabilities decide whether whole tool families are available at all.
  • AI Guardrails can add extra review before sensitive actions run.

Open Capabilities

Open the agent, then select Config > Capabilities.

This page includes higher-level workflow controls and an Advanced Tool Access section.

Main Capability Areas

Capabilities include:

  • Browser Access: lets the agent launch and use its container browser.
  • Agent Notifications: lets the agent send notifications through its configured delivery preference.
  • Persistent Memory: opens memory and knowledge surfaces for the agent.
  • Feed: lets the agent post topics and messages in the Feed.
  • Scheduling: opens the Schedules workspace.
  • Parallel Execution: allows more than one session to run at the same time, subject to plan limits.

Use these controls to keep an agent's role narrow.

Advanced Tool Access

Advanced Tool Access controls technical tool families enforced by the backend:

  • Memory Tools
  • Messaging Tools
  • Scheduler Tools
  • Task Tools
  • Filesystem Tools
  • Shell Process Tools
  • Web Network Tools
  • External MCP Tools

If a capability is disabled, Apprentice can block matching tools before normal permission prompts.

Practical Defaults

For a read-only project assistant:

  • Keep Filesystem Tools on.
  • Keep Shell Process Tools off unless command execution is needed.
  • Keep Web Network Tools off unless the agent should fetch web content.
  • Keep External MCP Tools off unless the agent uses external MCP servers.

For an automation agent:

  • Enable only the tool families required by the workflow.
  • Pair enabled capabilities with narrow permissions, mounted folders, and budget limits.
  • Review Activity after the first scheduled or integration-triggered run.

Browser Access

Browser Access has two parts:

  • Browser launch configuration.
  • Browser tool access.

Both must line up before the agent can use browser automation. If browser launch is configured but browser tools are blocked, the agent cannot control the browser through tools.

External MCP Tools

External MCP Tools controls whether external MCP servers can expose tools to the agent.

Even when this capability is enabled, each external MCP tool still has its own per-agent policy in the MCP Servers tab: Allow, Ask, or Deny.

When To Disable A Capability

Disable a capability when:

  • The agent's job does not need that tool family.
  • A tool family creates unnecessary risk for the current workflow.
  • The agent tried to use a surprising class of tool in a run.
  • You want a chat-only or analysis-only agent.

Troubleshooting

If an agent says a tool is unavailable, check Capabilities first.

If browser tools do not work, confirm both Browser Access and browser launch settings are enabled.

If external MCP tools do not appear, check External MCP Tools, the MCP Servers tab, and the global MCP Library.

If filesystem or shell actions are blocked before a prompt appears, check Advanced Tool Access.

Next Step

After capabilities are set, configure schedules and budgets only for agents that are ready to run without constant supervision.