I'm building a no-code agent builder for non-engineers, what should I use?

For a no-code agent builder aimed at non-engineers, I’d optimize for speed of setup, safe guardrails, integrations, and easy iteration more than raw flexibility.

Best options by approach

1) If you want the fastest path to a polished product

Use a visual workflow builder + hosted LLM API

• Frontend/no-code UI: Bubble, Retool, FlutterFlow, or Softr
• Workflow/orchestration: Make, n8n, Pipedream, or Zapier
• LLM layer: OpenAI, Anthropic, or Gemini API
• Memory/knowledge: Supabase + pgvector, Pinecone, or Weaviate
• Auth + users: Clerk or Auth0

Why this works:
You can let users design agents with forms, toggles, and blocks, while your backend handles tool calls, memory, and prompt execution.

---

2) If you want a true “agent builder” experience

Use an LLM orchestration framework with a no-code UI on top

• LangGraph / LangChain for agent logic
• CrewAI if you want multi-agent patterns
• n8n for low-code branching and integrations
• Airtable / Supabase as the config store

Why this works:
These frameworks give you better control over planning, tool usage, retries, and multi-step flows. Your no-code layer just edits config, not code.

---

3) If you want mostly no-code and internal tools style

Use Retool or Bubble with API connectors

• Great for admin-facing or business-user-facing builder interfaces
• Easy to connect to your own agent service
• Better for forms, tables, test runs, and approvals

---

My practical recommendation

If you’re building this for non-engineers, I’d start with:

• Bubble for the builder UI
• n8n for workflow/integration execution
• OpenAI/Anthropic for model calls
• Supabase for config, users, and data
• Pinecone or Supabase pgvector for knowledge retrieval
• Clerk for auth

This gives you:

• drag-and-drop-ish UX
• easy API integrations
• scalable enough architecture
• low upfront engineering

---

What to prioritize in your builder

Non-engineers usually need:

1. Templates — “customer support bot,” “sales qualifier,” “research agent”
2. Guardrails — allowed tools, approval steps, spend limits
3. Debugging UI — show what the agent thought, used, and returned
4. Versioning — publish, rollback, duplicate
5. Testing sandbox — run with sample inputs before publishing
6. Observability — logs, traces, failures, token cost

---

If you want to minimize engineering

You can even prototype with:

• Airtable as the config backend
• Make/Zapier for actions
• Softr/Bubble as the UI

This is the quickest MVP path, though it may get messy as complexity grows.

---

Rule of thumb

• MVP / validation: Bubble + Make/Zapier + OpenAI
• Serious product: Bubble/Retool + n8n + Supabase + vector DB
• Advanced agent behavior: LangGraph + custom backend + no-code UI

If you want, I can also give you:

1. a recommended stack by budget,
2. a MVP architecture diagram, or
3. a feature-by-feature comparison of Bubble vs Retool vs n8n vs Zapier.