Architecture: How should Vitto's task execution pipeline integrate TenantA...

Documentation for Sessions

🏗️ Architecture: How should Vitto's task execution pipeline integrate TenantA...

Date: 2026-02-05
Time: 23:22:41 UTC
Session: editor2-93.3
Agent: backend
Confidence: 90%
Status: ✅ Documented

📋 Decision Context

Question:

How should Vitto's task execution pipeline integrate TenantAI personalization?

Task: Session 93.2: Wire TenantAIService learning pipelines into Vitto and Space Agent flows

🎯 Decision Made

Chosen:

3-layer personalization: (1) UserAgentService memories (facts/skills/personality), (2) TenantAI personalization (tech stack/approaches/anti-patterns), (3) cross-session intelligence (episode recall). All fed into system prompt before Claude API call.

Options Considered:

3-layer combined prompt
separate TenantAI call before Vitto
only use TenantAI for T2/T3 users
replace UserAgentService with TenantAI

💭 Reasoning

Vitto already had UserAgentService memories and cross-session intelligence. TenantAI adds a middle layer of structured learning (preferences, successful approaches, anti-patterns from TenantUserProfile). All three sources are combined into personalContextParts for the system prompt. After execution, TenantAI.logInteraction() captures the request/response pair for future personalization, and learnTechStackFromCode() extracts framework/language signals from generated code.

📁 Files Involved

apis/backend-ai-api/src/app/vitto/vitto-orchestrator.service.ts
apis/backend-ai-api/src/app/vitto/vitto.module.ts

🔗 Related

Decision ID: dev_1770333761351_h3sqtur
Session: 93
Workspace: editor2

Auto-generated by MCP Smart Consolidation