Agentic Architecture Finder

Describe the problem by answering questions. The finder decides whether you need plain software, a single LLM call, structured output, tool calling, RAG, a workflow, an agent, a multi-agent system, MCP, memory, human approval, guardrails or an evaluation pipeline — and explains why the simpler options were rejected.

What agent architecture should I use?Do I need retrieval, and which kind?Which agent framework do I need?

Start from Option 0 — no framework — and only adopt one when a concrete need (state graphs, multi-agent roles, retrieval, type safety, or a TS runtime) justifies its abstraction cost.

Step 1

Do you need an agent framework at all?

Option 0 comes first on purpose: most systems start best as plain SDK calls plus a loop you own.