Level 3 Complete
Level 3 Complete!
Section titled “Level 3 Complete!”You’ve built your first autonomous agent — a research agent that independently researches, summarizes, and only saves after your approval. That’s a big step: from chatbot to actionable agent.
What You Learned
Section titled “What You Learned”- Tool Calling: Define tools with
tool()— Zod inputSchema,.describe()for parameters,executefor execution. The LLM autonomously decides when and with which parameters a tool is called. - Tools in the Frontend: Message parts (
text,tool-call,tool-result) for transparent UIs.fullStreamdelivers all events for loading states and tool-specific rendering. - Tool Loop Agent: Multi-step agents with
stopWhen: stepCountIs(n). The LLM iterates autonomously through the agentic loop — tool call, execute, result, next step — until the task is done. - MCP (Model Context Protocol): Standardized protocol for external tool servers.
createMCPClientwith HTTP or stdio transport,client.tools()loads all available tools automatically. - Tool Approval: Human-in-the-loop with
needsApproval. Static (true) or dynamic (async function).toolCallApprovalhandler for controlled execution of critical operations.
Updated Skill Tree
Section titled “Updated Skill Tree”
Next Level
Section titled “Next Level”Level 4: Persistence — Your agent can now act, but it forgets everything after each request. How do you store chat histories, restore conversations, and build persistent agents? You’ll learn database integration, message history, and session management.