Architecture: Comprehensive design for evolving Vitto into a Jarvis-level ...

Documentation for Sessions

🏗️ Architecture: Comprehensive design for evolving Vitto into a Jarvis-level ...

Date: 2026-03-18
Time: 14:09:46 UTC
Session: gspace1-232.4
Agent: architecture
Confidence: 93%
Status: ✅ Documented


📋 Decision Context

Question:

Comprehensive design for evolving Vitto into a Jarvis-level super agent

Task: Create comprehensive Vitto Jarvis evolution plan


🎯 Decision Made

Chosen:

11-section spec covering: sub-Vitto engine (parallel orchestration via AgentLoopService reuse), VM agent tier (Hetzner sandbox, SandboxedToolProxy, warm pool), expert sub-agent pragmatic creation (chat-based, marketplace, manual), OAuth credential delegation (backend-api owns user OAuth data, shared Redis cache, Pattern 7 brokered credentials, PKCE with session-bound CSRF state), dual-backend data flow contract, Vitto soul (personality drift auto-apply + visual evolution via normalized maturity score), Teams channel adapter, full-stack wiring (every backend feature visible in UI), MongoDB cleanup (drop channel-user-links + vitto-task-history ghost), build order (8 weeks phased). Spec reviewed by code-reviewer agent — 3 critical issues fixed (wrong service names, OAuth CSRF gap), 6 important issues fixed (adapter path, generateSkill claim, SubAgent refactoring scope, Redis dependency, routes gap, marketplace separation).

Options Considered:

  1. Build everything from scratch
  2. Use Composio for everything
  3. Hybrid: internal OAuth + Composio for scale + VM agents + existing infra reuse (chosen)
  4. Incremental per-feature specs instead of comprehensive plan

💭 Reasoning

Synthesized findings from: (1) deep codebase analysis of 45+ services, 53 MongoDB collections, 17 tool plugins, 60+ frontend components; (2) competitive research on OpenAI AgentKit, Anthropic Claude SDK, Manus, Devin, CrewAI, Character.AI; (3) 10 design decisions validated with product owner; (4) OAuth security research on Pattern 7 brokered credentials, PKCE, JIT tokens; (5) infrastructure analysis confirming shared Redis already deployed on Railway. Key architectural decisions: Vitto is the living entity (personality drift, visual evolution), sub-agents are tools (no soul). OAuth user data belongs in gograb-api (not vitto-mind). Build internal OAuth for Google+GitHub first, add Composio for scale later. VM agents run on existing underutilized Hetzner infrastructure. All inter-service state cached in shared Redis.


📁 Files Involved


🔗 Related


Auto-generated by MCP Smart Consolidation