Add Phase 4: Staff AI Assistant to self-hosted AI stack plan

docs/core/tasks.md

@@ -714,6 +714,78 @@ docker exec ollama ollama pull llama3.2-vision:11b # Images (8GB RAM)
 
 ---
 
+### **Phase 4: Staff AI Assistant (Staff Wiki Integration)**
+
+**Purpose:** Embedded AI assistant for staff to answer questions about Firefrost procedures, policies, and server management without bothering Michael/Meg.
+
+**Use Cases:**
+- Server-specific questions ("How do I use WorldEdit on creative server?")
+- Policy lookups ("What's our griefing response protocol?")
+- Quick troubleshooting ("Player can't connect, what do I check?")
+- Learning Firefrost procedures and culture
+- Onboarding tool for new staff (builders, moderators, helpers)
+
+**Implementation Options:**
+
+**Option A - Embedded Chat Widget (Recommended):**
+- Add chat interface to staff.firefrostgaming.com
+- Staff logs in via existing auth
+- AI responds using Firefrost knowledge base
+- Clean, simple, where staff already work
+
+**Option B - Dedicated Staff Portal:**
+- Separate section of Open WebUI on TX1
+- Staff-specific login credentials
+- Pre-loaded with Firefrost context
+- Usage tracking and conversation logs
+
+**Option C - Discord Bot:**
+- AI responds in staff-only Discord channels
+- Staff ask questions where they already hang out
+- Logs conversations for review
+
+**Knowledge Base (What It Knows):**
+- Staff wiki content (procedures, SOPs)
+- Server-specific guides (WorldEdit, permissions, etc.)
+- Firefrost culture and philosophy
+- Common troubleshooting steps
+- **NOT operations-manual repo** (keeps private stuff private)
+
+**Access Control:**
+- Read-only access (won't edit docs)
+- Only sees approved staff documentation
+- All conversations logged (optional oversight)
+- You control exactly what it can access
+
+**Benefits:**
+- ✅ 24/7 staff support (no waiting for Michael to wake up)
+- ✅ Reduces "Michael, how do I...?" interruptions
+- ✅ Consistent answers based on documented procedures
+- ✅ Onboarding tool for recruitment (supports builder/helper expansion)
+- ✅ Self-hosted, free, private, controlled
+- ✅ Pairs with recruitment strategy (staff can self-serve answers)
+
+**Technical Implementation:**
+- Uses same TX1 Ollama backend (already deployed in Phases 1-2)
+- Add staff-wiki docs to knowledge base
+- Configure access permissions
+- Deploy chat interface or bot integration
+- Test with Michael/Meg before rolling out to staff
+
+**Timeline:** 2-3 hours after Phases 1-3 complete
+
+**Dependencies:**
+- Phases 1-2 complete (Ollama + models deployed)
+- Staff wiki content documented
+- Decision on implementation option (A, B, or C)
+
+**Future Enhancement:**
+- Can expand to public-facing "Firefrost Helper" for players (FAQ, basic support)
+- Multi-language support for international community
+- Voice interface for accessibility
+
+---
+
 ### **⚠️ NOT RECOMMENDED: Migrate TX1 Games to NC1**
 
 **Analysis (Feb 14, 2026):** Migration provides NO meaningful benefit and introduces unnecessary risk.
@@ -741,6 +813,7 @@ docker exec ollama ollama pull llama3.2-vision:11b # Images (8GB RAM)
 
 ### **End Result:**
 
+**Personal/DERP Use:**
 - ✅ Self-hosted AI (80-90% Claude capability)
 - ✅ $0 monthly cost (vs $20-40 for cloud AI)
 - ✅ Complete privacy and control
@@ -748,6 +821,13 @@ docker exec ollama ollama pull llama3.2-vision:11b # Images (8GB RAM)
 - ✅ DERP backup ready (partnership survives provider failure)
 - ✅ Can run multiple models simultaneously
 
+**Staff Support:**
+- ✅ 24/7 AI assistant for staff questions
+- ✅ Reduces Michael/Meg interruptions
+- ✅ Accelerates staff onboarding
+- ✅ Consistent policy/procedure answers
+- ✅ Supports recruitment and scaling
+
 ---
 
 **Related:**