18 - Recommendarr (deprecated)

DEPRECATED - 2026-03-03

Recommendarr has been removed from this homelab due to security concerns. The upstream GitHub repository (qdread/recommendarr) was deleted without explanation around the same time as the Huntarr security incident (February 2026). Given the developer's disappearance and the lack of any official statement, running this service is no longer considered safe.

This document is kept as historical reference only.

Docker Compose Configuration¶

services:
  recommendarr:
    image: qdread/recommendarr:latest
    container_name: recommendarr
    ports:
      - "3003:3000"
    restart: unless-stopped
    volumes:
      - /srv/docker-data/recommendarr:/data
    networks:
      - arr_stack

networks:
  arr_stack:
    external: true

File Paths¶

Data Directory: /srv/docker-data/recommendarr/
Storage/DB: /srv/docker-data/recommendarr/storage/ (auto-created)

Initial Setup¶

Default Credentials¶

Username: admin
Password: 1234

⚠️ CRITICAL: Change password immediately after first login!

Access: http://192.168.0.110:3003

Configuration Steps¶

1. Ollama LLM Configuration¶

Ollama Service Setup (Nobara PC)¶

Prerequisites: - Ollama installed on Nobara: http://192.168.0.100:11434 - GPU: RTX 2060 SUPER (8GB VRAM)

Required: Enable CORS for Browser Access

# On Nobara PC
sudo systemctl edit ollama

Add:

[Service]
Environment="OLLAMA_ORIGINS=*"

Save and restart:

sudo systemctl daemon-reload
sudo systemctl restart ollama

Verify CORS:

# Should return Access-Control-Allow-Origin: *
curl -I -H "Origin: http://192.168.0.110:3003" http://localhost:11434/v1/models

# Test from LXC
curl http://192.168.0.100:11434/v1/models

Recommendarr LLM Settings¶

Navigate to Settings → AI Settings
API URL: http://192.168.0.100:11434/v1
API Key: Leave empty (local Ollama doesn't need key)
Click Fetch Available Models
Select Model:
qwen2.5:7b (recommended for 8GB VRAM) ✅
llama3.2:3b (faster, smaller)
mistral:7b (alternative)
❌ NOT nomic-embed-text (embedding-only, won't work for generation)

Model Selection for RTX 2060 SUPER (8GB VRAM): - Best performance: qwen2.5:7b (36-40 tokens/s, better reasoning) - Fastest: llama3.2:3b (50+ tokens/s) - Alternative: mistral:7b (~40 tokens/s) - Avoid: llama3.1:8b (too large, only 7-9 tokens/s, 80% VRAM)

Pull models on Nobara:

ollama pull qwen2.5:7b
ollama pull llama3.2:3b

# Verify
ollama list

# Test generation
ollama run qwen2.5:7b "Hello, test"

GPU Verification:

# Monitor GPU during generation
watch -n 1 nvidia-smi

2. Sonarr v4 Integration¶

Problem: Sonarr v4 forces authentication, "None" mode no longer valid.

Failed Solutions: - Setting <AuthenticationMethod>None</AuthenticationMethod> - UI overrides on restart - "Disabled for Local Addresses" - Docker IPs not considered local

Working Solution: External Authentication Mode

# Stop Sonarr
docker stop sonarr

# Edit config
nano /srv/docker-data/sonarr/config.xml

Find and change:

<AuthenticationMethod>Basic</AuthenticationMethod>

To:

<AuthenticationMethod>External</AuthenticationMethod>

Restart and verify:

docker start sonarr

# Verify
curl http://192.168.0.110:8989/api/v3/system/status \
  -H "X-Api-Key: YOUR_SONARR_API_KEY" | grep authentication

# Should return: "authentication": "external"

⚠️ Security Warning: External mode completely disables Sonarr authentication. Only safe on isolated local networks. Do NOT expose to internet!

Recommendarr Sonarr Settings¶

Navigate to Settings → Services
Sonarr URL: http://192.168.0.110:8989
API Key: Get from Sonarr Settings → General → Security
Save
Test connection - should work without auth errors

3. Radarr Integration (if used)¶

Same pattern as Sonarr:

docker stop radarr
nano /srv/docker-data/radarr/config.xml
# Change to External auth
docker start radarr

Recommendarr Radarr Settings: - URL: http://192.168.0.110:7878 - API Key: From Radarr Settings → General → Security

4. Jellyfin Integration¶

Recommendarr Jellyfin Settings: - URL: http://192.168.0.110:8096 - API Key: Jellyfin Dashboard → API Keys → Create New

Usage: - Fetches watch history for generating recommendations - Uses user activity to personalize suggestions

⚠️ CRITICAL BUGS & FIXES¶

Bug #1: CORS Error - Cannot Connect to Ollama¶

Symptom: - Browser error: Failed to connect to LLM server at http://192.168.0.100:11434/v1. Network Error - Console: CORS policy: No 'Access-Control-Allow-Origin' header - Fetch Available Models button fails - curl from host works, but browser blocks

Cause: Ollama default config doesn't allow cross-origin requests from browser.

Solution: See Ollama CORS Configuration section above.

Bug #2: Sonarr v4 Authentication Bypass¶

Symptom: - 401 Unauthorized errors when connecting to Sonarr - API key works in curl but not in Recommendarr - Sonarr UI forces authentication method selection

Solution: See Sonarr v4 Integration section above (External auth mode).

Bug #3: Session/Storage Corruption¶

Symptoms: - "Invalid or expired session" errors (401) - 500 Internal Server Error on all settings saves - "Unsupported protocol ttp:" (missing 'h' in URLs) - Cannot save credentials or settings - UI becomes unresponsive

Cause: Corrupt Recommendarr storage/database files.

Solution: Reset Storage

# Nuclear option - wipes all settings
docker stop recommendarr
rm -rf /srv/docker-data/recommendarr/*
docker start recommendarr

# Fresh browser tab (hard refresh: Ctrl+Shift+R)
# Login again (admin/1234)
# Reconfigure all services

Prevention: - Ensure URLs are correct (http:// not ttp://) - Don't spam save button repeatedly - Wait for saves to complete before navigating away

Bug #4: Wrong Model Selected¶

Symptom: - Recommendations fail to generate - Errors about model incompatibility - No output from LLM

Cause: Selected nomic-embed-text or other embedding-only model.

Solution: - Embedding models (e.g., nomic-embed-text) cannot generate text - Select a generative model: - ✅ qwen2.5:7b - ✅ llama3.2:3b - ✅ mistral:7b - ❌ NOT nomic-embed-text

Operational Procedures¶

Generate Recommendations¶

Navigate to Recommendations tab
Select user (if multi-user setup)
Click "Generate Recommendations"
Wait for LLM processing (GPU: 10-30 seconds)
Review generated suggestions
Click "Request in Sonarr/Radarr" to add

Performance Monitoring¶

GPU Usage (Nobara):

# Real-time monitoring
watch -n 1 nvidia-smi

# Check Ollama process
ps aux | grep ollama

# If GPU not utilized - model might be CPU-only
# Re-pull model to force GPU: ollama pull qwen2.5:7b

Ollama Process Management:

# If Ollama consuming excessive resources
systemctl restart ollama

# Or kill runner processes
kill -9 $(pgrep -f ollama)
systemctl restart ollama

Session Management¶

If session expires: 1. Logout 2. Login again (admin + password) 3. Settings should persist (unless storage corrupted)

If settings lost: - Storage corruption likely (Bug #3) - Reset storage and reconfigure

Troubleshooting¶

Cannot Fetch Ollama Models¶

Check: 1. Ollama service running: systemctl status ollama 2. CORS enabled: sudo systemctl cat ollama - check for OLLAMA_ORIGINS 3. Network connectivity: curl http://192.168.0.100:11434/v1/models 4. Firewall: sudo firewall-cmd --list-all (if enabled)

Fix:

# Restart Ollama with CORS
sudo systemctl restart ollama

# Verify
curl -I -H "Origin: http://192.168.0.110:3003" \
  http://192.168.0.100:11434/v1/models | grep Access-Control

Sonarr Connection Failed¶

Check: 1. Sonarr running: docker ps | grep sonarr 2. Auth mode: grep AuthenticationMethod /srv/docker-data/sonarr/config.xml 3. API key valid: curl http://192.168.0.110:8989/api/v3/system/status -H "X-Api-Key: KEY"

Fix: - Set External auth mode (see Bug #2) - Verify API key copied correctly - Test connection from Recommendarr Settings

Recommendations Not Generating¶

Check: 1. Ollama model selected (not embedding model) 2. GPU being utilized: nvidia-smi 3. Jellyfin watch history exists for user 4. LLM accessible: curl http://192.168.0.100:11434/v1/models

Debug:

# Recommendarr logs
docker logs recommendarr -f

# Ollama logs
journalctl -u ollama -f

# Test generation manually
ollama run qwen2.5:7b "Recommend movies similar to Inception"

500 Internal Server Errors¶

Likely: Storage corruption (Bug #3)

Solution:

docker stop recommendarr
rm -rf /srv/docker-data/recommendarr/*
docker start recommendarr
# Reconfigure

Container Won't Start¶

# Logs
docker logs recommendarr

# Common issues:
# - Port 3003 conflict: change to 3004:3000
# - Volume permissions
# - Corrupt storage (rm -rf and restart)

Integration URLs¶

Recommendarr: http://192.168.0.110:3003
Ollama (Nobara): http://192.168.0.100:11434
Sonarr: http://192.168.0.110:8989
Radarr: http://192.168.0.110:7878
Jellyfin: http://192.168.0.110:8096
SuggestArr: http://192.168.0.110:5000

Performance Notes¶

GPU vs CPU (Ollama)¶

RTX 2060 SUPER (8GB VRAM): - qwen2.5:7b: 36-40 tokens/s ✅ - llama3.2:3b: 50+ tokens/s ✅ - llama3.1:8b: 7-9 tokens/s ❌ (too slow, 80% VRAM)

CPU-only (Proxmox LXC): - Any 7B+ model: <5 tokens/s ❌ (unusable)

Recommendation: Always use GPU for LLM inference.

Known Issues (Feb 2026)¶

CORS requirement - Ollama needs explicit CORS configuration for browser access
Sonarr v4 auth - Forces authentication, External mode only workaround
Storage corruption - UI/settings can become corrupted, requires reset
Session expiry - Periodic logouts, no permanent session option
Model confusion - Embedding models don't work for generation

Bug Reports: Submit to https://github.com/qdread/recommendarr/issues

Advanced Configuration¶

Multi-User Setup¶

Configure multiple Jellyfin users
Generate personalized recommendations per user
Separate watch histories analyzed

Custom Prompts (if supported)¶

Modify LLM prompt templates
Adjust recommendation criteria
Fine-tune generation parameters

Backup Strategy¶

# Backup settings (before making changes)
cp -r /srv/docker-data/recommendarr /srv/docker-backups/recommendarr-$(date +%Y%m%d)

# Restore
docker stop recommendarr
rm -rf /srv/docker-data/recommendarr
cp -r /srv/docker-backups/recommendarr-YYYYMMDD /srv/docker-data/recommendarr
docker start recommendarr