Architecture: What is the verified technical spec for evolving Vitto into ...

Documentation for Sessions

🏗️ Architecture: What is the verified technical spec for evolving Vitto into ...

Date: 2026-03-17
Time: 19:17:21 UTC
Session: gspace1-229.14
Agent: documentation
Confidence: 92%
Status: ✅ Documented

📋 Decision Context

Question:

What is the verified technical spec for evolving Vitto into a Jarvis-like agent?

Task: Write verified Vitto Jarvis Evolution spec

🎯 Decision Made

Chosen:

4-phase spec written at docs/superpowers/specs/2026-03-17-vitto-jarvis-evolution-spec.md. Phase 1 (fix foundations + ship daemon): wire embedding search (4 files, ~50 lines — LearningEmbeddingAdapterService bridging DecisionEmbeddingService to LearningService), fix 5 cost tracker bugs (missing cost field in response, spendingHistory not rehydrated, static provider/model, teach intent untracked, memory-only stats), health check ping, ship Tauri daemon (1-line visible:false + infinite reconnect + heartbeat). Phase 2: VM snapshot auth persistence + CU action memory + OAuth delegation (Composio) + human takeover VNC. Phase 3: voice pipeline (Whisper + TTS) + browser extension relay + OpenTelemetry + eval framework (50 golden pairs + Haiku judge). Phase 4: closed-loop skills + persona adaptation + platform intelligence aggregation.

Options Considered:

💭 Reasoning

All 4 Phase 1 items code-verified by dedicated agents: embedding search confirmed disconnected (DecisionSearchService.vectorSearch exists but not exported, LearningService.setEmbeddingClient never called), cost tracker has 5 stacked bugs (not 1), chat SSE already works for web UI (polling only affects proxy path), Tauri app is buildable and closer to daemon mode than expected (1-line config + reconnection fix). Spec grounded in line-level code evidence, not assumptions.

📁 Files Involved

🔗 Related

Auto-generated by MCP Smart Consolidation