Skip to main content

Building & running agents

From a one-line description to a deployed, scheduled agent with the right tools and integrations. These scenarios combine the agent, integration, and schedule functions.

Create an agent and refine it​

Goal — Stand up a new agent, then sharpen its summary and behavior.

Example prompt

"Create an agent weekly_report_writer that drafts our Monday status update. … Now tweak its description to say it pulls from the project tables."

Functions used

  1. agent-create — creates the agent with name, shortDescription, and systemPrompt.
  2. folder-list — finds the agent id for the follow-up edit.
  3. agent-update — updates the shortDescription (commits a draft version).

What you get — An agent whose edit is staged as a draft version. Higento asks you to deploy it in chat before it goes live.

Connect an integration, then build the agent​

Goal — Build an agent that needs a third-party service, making sure the connection is in place first.

Example prompt

"I want an agent that reads my Outlook inbox and drafts replies. Connect Outlook first, then create the agent and attach Outlook to it."

Functions used

  1. integration-connect — shows the Outlook card so you can connect it; reports whether it's configured/active.
  2. agent-ask-create — (optional) confirms scope/options once the connection works.
  3. agent-create — creates the agent.
  4. agent-link-integration — attaches Outlook. If it isn't configured yet, the call returns requiresIntegrationConfig and Higento prompts you to connect it.

What you get — An agent wired to a connected integration, ready to deploy.

Attach (and later remove) a workspace tool​

Goal — Give an agent a reusable capability you built as a tool, then unlink it when it's no longer needed.

Example prompt

"Link the ticket-summarizer tool to my support_triage agent. … On second thought, unlink it again."

Functions used

  1. folder-list — resolves the agent and tool ids.
  2. agent-link-tool — links the workspace tool (tool_…) to the agent.
  3. agent-unlink-tool — removes it.

What you get — The agent's draft version gains (then loses) the tool. Remember: workspace tools use agent-link-tool; catalog connectors use agent-link-integration.

Schedule an agent to run automatically​

Goal — Run an agent on a recurring cron schedule, then pause it.

Example prompt

"Have weekly_report_writer run every Monday at 9am with the prompt 'compile last week's status'. … List its schedules, then pause that one."

Functions used

  1. folder-list — finds the agent id.
  2. schedule-create — creates a trigger with cronExpression: "0 9 * * 1" and the run prompt.
  3. schedule-list — shows the agent's schedules.
  4. schedule-update — sets the schedule active: false.

What you get — A recurring trigger that runs the agent, which you can enable/disable or retune (cronExpression, prompt, executeAsUser) at any time.

Run an agent on demand​

Goal — Execute an agent once with an ad-hoc input to see its output.

Example prompt

"Run weekly_report_writer now with input 'draft this week's update'."

Functions used

  1. folder-list — finds the agent id.
  2. agent-run — executes the agent with your input; org/user/session ids are set automatically.

What you get — The agent runs and returns its result. (Execution requires the agent backend to be configured.)