firefrost-operations-manual/docs/tasks/firefrost-codex/MIGRATION-TO-OPEN-WEBUI.md

Firefrost Codex - Migration to Open WebUI + Repomix

Document ID: CODEX-MIGRATION-001
Created: February 21, 2026
Status: 📋 PLANNED - Ready to Execute
Estimated Time: 1 hour
Risk Level: LOW (can rollback to AnythingLLM if needed)

---

🎯 OBJECTIVE

Replace AnythingLLM with Open WebUI + Repomix workflow to solve document retrieval quality issues.

Current Problem:

• AnythingLLM indexing 319 files creates poor search relevance
• AI finds old/archived docs instead of current operational info
• Vector search overwhelmed by document volume and similarity

Solution:

• Open WebUI: More efficient RAG engine, better performance
• Repomix: Packages entire operations manual into single, clean file
• Combined: Fast + accurate retrieval without noise

---

📊 COMPARISON

Feature	AnythingLLM (Current)	Open WebUI + Repomix
Documents	319 individual files	1 packaged digest
Search Quality	Poor (finds old docs)	Good (clean context)
Performance	Slow indexing	Fast responses
Updates	Re-sync 319 files	Re-run packager
Maintenance	High (vector DB)	Low (single file)
Cost	$0/month	$0/month

---

⚠️ PREREQUISITES

Before starting:

• SSH access to TX1 (38.68.14.26)
• Docker running on TX1
• Ollama running on TX1 (port 11434)
• Git repo cloned at /root/firefrost-operations-manual
• ~30 minutes of focused time
• Backup plan (can restart AnythingLLM if needed)

---

🛠️ MIGRATION STEPS

Step 1: Install Repomix (5 minutes)

SSH to TX1:

ssh root@38.68.14.26

Install Repomix globally:

npm install -g repomix

Verify installation:

repomix --version

Expected output: Version number (e.g., repomix v2.x.x)

---

Step 2: Generate Operations Manual Digest (5 minutes)

Navigate to operations manual:

cd /root/firefrost-operations-manual

Run Repomix to package entire repo:

repomix --output firefrost-ops-digest.md

What this does:

• Scans entire repository
• Automatically excludes .git, node_modules, binaries
• Creates single markdown file with intelligent structure
• Optimized for LLM consumption

Verify the digest:

ls -lh firefrost-ops-digest.md
wc -l firefrost-ops-digest.md

Expected: File should be several thousand lines, representing all docs

---

Step 3: Stop AnythingLLM (2 minutes)

Navigate to AnythingLLM directory:

cd /opt/anythingllm

Stop the containers:

docker-compose down

Verify stopped:

docker ps | grep anythingllm

Expected: No output (containers stopped)

Optional - Remove containers completely:

# Only if you want to free up space
# docker-compose down -v  # This removes volumes too

Note: We're keeping the data volumes in case we need to rollback.

---

Step 4: Install Open WebUI (10 minutes)

Pull and run Open WebUI container:

docker run -d \
  -p 3001:8080 \
  --name open-webui \
  --add-host=host.docker.internal:host-gateway \
  -v open-webui:/app/backend/data \
  -e OLLAMA_BASE_URL=http://host.docker.internal:11434 \
  --restart always \
  ghcr.io/open-webui/open-webui:main

Breakdown:

• -p 3001:8080 - Same port as old Codex (http://38.68.14.26:3001)
• --add-host - Allows container to reach host's Ollama
• -v open-webui:/app/backend/data - Persistent storage
• -e OLLAMA_BASE_URL - Points to existing Ollama instance
• --restart always - Auto-start on server reboot

Verify container running:

docker ps | grep open-webui

Expected: Container should show as "Up"

Check logs:

docker logs open-webui

Expected: Should see startup messages, no errors

---

Step 5: Initial Setup via Web UI (10 minutes)

Access Open WebUI:

1. Open browser to: http://38.68.14.26:3001
2. First visit will prompt to create admin account
3. Create account:
   • Username: mkrause612
   • Email: (your email)
   • Password: (secure password)

Verify Ollama connection:

1. Click Settings (gear icon)
2. Go to "Connections"
3. Should show Ollama at http://host.docker.internal:11434
4. Should list available models (qwen2.5-coder:7b, llama3.3:70b)

If models don't show:

# Verify Ollama is accessible from container
docker exec open-webui curl -s http://host.docker.internal:11434/api/tags

---

Step 6: Upload Operations Manual Digest (5 minutes)

In Open WebUI:

1. Click "Documents" in sidebar
2. Click "Upload Documents"
3. Select /root/firefrost-operations-manual/firefrost-ops-digest.md
4. Upload completes
5. Document should appear in list

Alternative - Upload via command line:

# Get file size to verify
ls -lh /root/firefrost-operations-manual/firefrost-ops-digest.md

# File is uploaded via web UI (easier than API for first time)

---

Step 7: Create Workspace and Test (10 minutes)

Create Operations workspace:

1. Click "Workspaces" (or similar)
2. Create new workspace: "Operations"
3. Attach firefrost-ops-digest.md document
4. Select model: qwen2.5-coder:7b

Test queries:

Query 1: Current tasks

What are the current Tier 0 tasks according to tasks.md?

Expected: Should list actual current tasks (Whitelist Manager, NC1 Cleanup, etc.)

Query 2: Infrastructure

What servers does Firefrost Gaming operate and what are their IP addresses?

Expected: Should list TX1, NC1, Command Center, etc. with correct IPs

Query 3: Recent work

What was accomplished in the most recent Codex session?

Expected: Should reference Phase 2 work, workspaces created, etc.

If responses are accurate: ✅ Migration successful!

---

Step 8: Create Additional Workspaces (10 minutes)

Replicate the workspace structure:

1. Operations (already created)

   • Document: firefrost-ops-digest.md
   • Model: qwen2.5-coder:7b
   • Users: Admins only

2. Brainstorming

   • Document: firefrost-ops-digest.md
   • Model: llama3.3:70b (for deeper thinking)
   • Users: Admins only

3. Public KB

   • Document: (none yet - future public docs)
   • Model: qwen2.5-coder:7b
   • Users: Public access (future)

4. Subscriber KB

   • Document: (none yet - future subscriber docs)
   • Model: qwen2.5-coder:7b
   • Users: Subscriber access (future)

5. Relationship

   • Document: firefrost-ops-digest.md (has relationship docs in it)
   • Model: qwen2.5-coder:7b
   • Users: Admins only

6. Pokerole Project

   • Document: (none yet - needs pokerole repo)
   • Model: qwen2.5-coder:7b
   • Users: Admins + Holly

---

Step 9: Create User Accounts (5 minutes)

In Open WebUI Settings → Users:

1. gingerfury (Meg)

   • Role: Admin
   • Email: (Meg's email)
   • Temporary password (she can change)

2. Unicorn20089 (Holly)

   • Role: User (not Admin)
   • Email: (Holly's email)
   • Access: Pokerole Project workspace only

Note: Open WebUI has different permission model than AnythingLLM. May need to configure workspace access differently.

---

Step 10: Document Sync Automation (10 minutes)

Update the sync workflow:

Instead of syncing individual files, we now just re-run Repomix.

Create new sync script:

cat > /root/codex-sync-openwebui.sh << 'EOFSCRIPT'
#!/bin/bash
# Firefrost Codex - Open WebUI Sync Script
# Updates the operations manual digest

set -e

REPO_DIR="/root/firefrost-operations-manual"
DIGEST_FILE="$REPO_DIR/firefrost-ops-digest.md"

echo "=== Firefrost Codex Sync (Open WebUI) ==="
echo "Started: $(date)"
echo ""

# Pull latest from Git
cd "$REPO_DIR"
echo "📥 Pulling latest from Git..."
git pull
echo ""

# Regenerate digest
echo "📦 Regenerating operations manual digest..."
repomix --output firefrost-ops-digest.md
echo ""

# Show file info
echo "✅ Digest updated:"
ls -lh "$DIGEST_FILE"
echo ""

echo "📌 Next step: Upload firefrost-ops-digest.md to Open WebUI"
echo "   Location: http://38.68.14.26:3001"
echo ""

echo "✅ Sync complete: $(date)"
EOFSCRIPT

chmod +x /root/codex-sync-openwebui.sh

Test the new sync script:

/root/codex-sync-openwebui.sh

Then manually upload the new digest via web UI

---

🔄 ROLLBACK PLAN

If Open WebUI doesn't work or has issues:

Step 1: Stop Open WebUI

docker stop open-webui
docker rm open-webui

Step 2: Restart AnythingLLM

cd /opt/anythingllm
docker-compose up -d

Step 3: Access old Codex

• URL: http://38.68.14.26:3001
• All data still intact (we didn't remove volumes)

Time to rollback: 2 minutes

---

📋 POST-MIGRATION CHECKLIST

After migration is complete:

• Open WebUI accessible at http://38.68.14.26:3001
• Ollama connection working (models listed)
• Operations workspace created with digest uploaded
• Test queries return accurate, current information
• Additional workspaces created (Brainstorming, etc.)
• User accounts created (gingerfury, Unicorn20089)
• Sync script updated and tested
• Old AnythingLLM containers stopped (but not deleted)
• Documentation updated in Git

Optional cleanup (after 1 week of successful use):

• Remove AnythingLLM Docker volumes (frees ~500MB)
• Remove old sync script /root/codex-sync.sh
• Update PROJECT-SCOPE with new architecture

---

📊 EXPECTED IMPROVEMENTS

Performance:

• Faster responses (Open WebUI more efficient)
• No indexing lag (single file vs 319 files)
• Lower memory usage

Accuracy:

• Better retrieval (clean context vs noisy corpus)
• Current info prioritized (packaged digest vs mixed archives)
• Fewer "hallucinations" (clearer document structure)

Maintenance:

• Simpler updates (re-run packager vs re-sync 319 files)
• Clear versioning (digest is dated file)
• Easier to troubleshoot (one file to check)

---

🆘 TROUBLESHOOTING

If Open WebUI won't start:

# Check logs
docker logs open-webui

# Common issues:
# - Port 3001 already in use (make sure AnythingLLM is stopped)
# - Ollama not accessible (check host.docker.internal)

If Ollama connection fails:

# Test from inside container
docker exec open-webui curl http://host.docker.internal:11434/api/tags

# If that fails, Ollama might not be running
systemctl status ollama

If upload fails:

• Check file size: ls -lh firefrost-ops-digest.md
• File too large? Repomix should auto-limit, but check
• Try uploading via web UI instead of API

If queries return poor results:

• Check which model is being used
• Verify digest file uploaded correctly
• Try more specific queries
• May need to adjust retrieval settings

---

📝 NOTES

Why Repomix instead of manual curation?

• Automatic exclusion of .git, node_modules, binaries
• Intelligent structure preservation
• Consistent formatting for LLMs
• Easy to regenerate when repo updates
• Industry-standard tool (maintained, documented)

Why Open WebUI instead of alternatives?

• Most mature AnythingLLM alternative
• Active development community
• Better performance than AnythingLLM
• More features (function calling, tools)
• Docker-based (consistent with our stack)

Can we keep AnythingLLM running alongside?

• Yes, but they'd need different ports
• Not recommended (confusing to have both)
• Better to fully commit to one approach

---

🔗 RESOURCES

• Open WebUI Docs: https://docs.openwebui.com/
• Repomix GitHub: https://github.com/yamadashy/repomix
• Ollama API: http://38.68.14.26:11434/api/tags
• New Codex URL: http://38.68.14.26:3001

---

✅ SUCCESS CRITERIA

Migration is successful when:

1. ✅ Open WebUI accessible and responsive
2. ✅ Operations manual digest uploaded
3. ✅ Test queries return current, accurate information
4. ✅ Response time under 10 seconds
5. ✅ Multiple workspaces functional
6. ✅ User accounts working with correct permissions
7. ✅ No errors in Docker logs

At that point: Phase 2 can be considered COMPLETE ✅

---

Fire + Frost + Foundation + Better Tools = Where Love Builds Legacy 💙🔥❄️

Document Status: Ready to execute
Next Action: Run Step 1 when ready
Estimated Completion: 1 hour from start to finish