🏗️ Architecture: How to transform Vitto from chatbot to agent — model upgrade...

Date: 2026-02-21
Time: 02:34:53 UTC
Session: gspace1-131.3
Agent: backend
Confidence: 92%
Status: ✅ Documented

📋 Decision Context

Question:

How to transform Vitto from chatbot to agent — model upgrade, routing, memory, tools

Task: Transform Vitto from chatbot to groundbreaking AI agent

🎯 Decision Made

Chosen:

Hybrid model tiers (Opus for complex, Sonnet for medium, Haiku for simple), fix classifier short-message trap, wire UserKnowledgeService, semantic memory summaries, register execute_code tool via E2B

Options Considered:

• No alternatives documented

💭 Reasoning

WhatsApp conversation analysis showed Vitto behaving as a chatbot because: (1) conversation mode is toolless, (2) classifier routes too many messages to conversation, (3) model is mid-tier everywhere, (4) memory summaries are truncated noise, (5) UserKnowledgeService exists but was never wired, (6) E2B sandbox exists but has no tool exposing it. All fixes target root causes with minimal code changes.

📁 Files Involved

• apis/backend-vitto-api/src/app/engine/vitto-engine.service.ts
• apis/backend-vitto-api/src/app/strategy/request-classifier.service.ts
• apis/backend-vitto-api/src/app/tools/tool-registry.service.ts
• apis/backend-vitto-api/src/app/tools/tools.module.ts
• .env.local

🔗 Related

• Decision ID: dev_1771641293902_nd5vsqa
• Session: 131
• Workspace: gspace1

Auto-generated by MCP Smart Consolidation