[Syed.Atyab@Codeline.rihal.om] Voice-to-Chat: One-Tap Personal Agent Creator

[CodelineAtyab]

🎯 User Story

As a brand-new, non-technical user,
I want to send a short voice note saying
“Build me an agent that can answer questions about my files”
and instantly receive a chat URL plus my API key,
so that I can begin chatting with my own file-aware agent without any manual setup.

---

✅ Acceptance Criteria

#	Requirement	Key Details
1	Voice Note Transcription	• WAV/MP3 ≤ 15 s • ≥ 95 % transcription accuracy
2	LLM-Generated n8n Workflow	Must return a valid, importable JSON with:   • Webhook IN node (expects input_text, api_key)   • AI-Agent node (5-message short-term memory)   • HTTP/RAG node     – Queries user-specific Supabase table first     – Falls back to shared table if no match   • Webhook / Chat UI OUT node   • Structured Output node → returns { chat_url, api_key }
3	Auto-Import & Activation	Wrapper app auto-imports and activates the workflow in the local n8n instance within 30 s of voice upload
4	API Response	Returns a working HTTPS chat URL + the user’s unique API key
5	Chat UI	Visiting the URL opens a simple UI where the user can: • Send a test message • Receive an AI-generated answer from the new workflow
6	Local-Only Deployment	All components run locally via a provided Docker-compose file — no paid cloud dependencies

---

🗺️ End-to-End Flow

1. User uploads voice note (≤ 15 s, WAV/MP3).
2. Transcription service converts audio → text (≥ 95 % accuracy).
3. LLM consumes text, outputs fully-formed n8n JSON workflow (see criteria [syed.atyab@codeline.rihal.om] Edit-My-Feedback Kiosk MVP #2).
4. Wrapper app:
   • Imports workflow into local n8n instance.
   • Activates it within 30 s.
5. API response returns { chat_url, api_key }.
6. User opens chat URL, sends a test query.
7. AI-Agent fetches relevant data (Supabase RAG) → responds via Chat UI.

---

🛠️ Tech Stack

• Audio → Text: Open-source speech-to-text (e.g., Whisper)
• LLM: Local or self-hosted model capable of JSON code-gen
• Workflow Engine: n8n (Docker)
• Database / RAG: Supabase (self-hosted)
• Containerization: Docker-compose (single-command setup)

---

⏱️ Performance Targets

• Voice-to-workflow latency: ≤ 30 s
• Chat round-trip time: ≤ 2 s (local network)

---

Kick off with a 15-second voice note → get a chat link & key → start asking your files anything, all in under a minute, 100 % locally.