Architecture: How should WhatsApp/channel messages be routed in Vitto to e...

Documentation for Sessions

🏗️ Architecture: How should WhatsApp/channel messages be routed in Vitto to e...

Date: 2026-02-20
Time: 18:32:27 UTC
Session: gspace1-128.2
Agent: backend
Confidence: 92%
Status: ✅ Documented


📋 Decision Context

Question:

How should WhatsApp/channel messages be routed in Vitto to enable agent capabilities beyond chatbot?

Task: Session 129: Transform Vitto from chatbot to agent on WhatsApp/channels + close intelligence learning loop


🎯 Decision Made

Chosen:

3-path channel router: conversation (single LLM) / command (agent loop with filtered tools) / complex_task (4-phase pipeline with ack)

Options Considered:


💭 Reasoning

Session 128 added a fast conversational path for channel messages (single LLM call, no tools). This made Vitto a chatbot on WhatsApp, not an agent. OpenClaw's power comes from executing commands via chat. The 3-path router classifies intent using rule-based patterns (~1ms) and routes: slash commands (/status, /help, /notify) get instant responses; commands use the agent loop with 13 channel-safe tools (no shell, no files, no browser) and max 8 iterations; complex tasks get an ack message then enter the full 4-phase pipeline. Also closed the intelligence learning loop: added super-intelligence methods to BackendAiClientService (recallEpisodes, createReflection, recordEpisode, simulateAction), wired episode recall into ANALYZE phase, reflections after VERIFY, and auto-episode creation in OutcomeFeedbackService.


📁 Files Involved


🔗 Related


Auto-generated by MCP Smart Consolidation