ACAgentic Craft

FoundationHazelJS

Agent

Software that repeatedly chooses and executes tool-mediated actions toward a goal under budgets and policies.

Authors
editorial-team
Published
Last reviewed

Direct answer

Software that repeatedly chooses and executes tool-mediated actions toward a goal under budgets and policies.

Plain language

An agent doesn’t just answer once—it tries things, sees results, and continues until it’s done or stopped.

Technical explanation

In HazelJS, a runtime-mediated control loop inside AgentRuntime: assemble context → model proposes action/final → validate/execute @Tool calls via ToolRegistry → append observations → checkpoint → stop on success, RunBudget, or PolicyEngine deny.

Example

An OpsInvestigator @Agent that calls get_incident and search_changes via AgentRuntime.execute, then returns a structured summary within six maxSteps.

Related

← All glossary terms