Build a real AI agent — a streaming chatbot with memory and real tool-calling, not just a wrapper around a chat API — using Claude, Codex, or Lovable.
7 prompts: streaming chat UI → real tool-calling → RAG (pgvector) → guardrails → deploy.
Unlock the rest with purchaseVercel AI SDK, Claude API, Claude Agent SDK/MCP, Supabase pgvector — and why each one, not just what.
Unlock the rest with purchase5 attacks to run on your own agent: direct + indirect prompt injection, instruction leakage, tool-call abuse, output XSS.
Unlock the rest with purchase