Skip to content

Langfuse & Semantic Cache Setup Instructions

Issues Fixed

1. Langfuse Authentication Errors ✅

Problem: Invalid credentials errors in Langfuse logs 

langfuse-web-1 | Error: Invalid credentials
langfuse-web-1 | No key found for public key
langfuse-web-1 | storing api-key-non-existent in redis

Root Cause: Placeholder API keys in development.env were not actual Langfuse keys.

Solution: Updated configuration to support proper API key generation workflow.

2. Semantic Cache Ollama Connection ✅

Problem: Semantic cache failing to connect to Ollama 

[SEMANTIC_CACHE] ⚠ Initialization failed: Failed to connect to Ollama

Root Cause: No Ollama service defined in docker-compose setup.

Solution: Added Ollama service to docker-compose.langfuse.yml.

3. Missing S3 Bucket Configurations ✅

Problem: Langfuse v3 requires multiple S3 buckets but only events bucket was configured.

Root Cause: Incomplete S3/MinIO configuration per Langfuse documentation.

Solution: Added configurations for: - LANGFUSE_S3_MEDIA_UPLOAD_* (required) - LANGFUSE_S3_BATCH_EXPORT_* (optional) - Updated MinIO to create all buckets on startup

Changes Made

1. development/development.env

  • ✅ Updated Langfuse API key configuration with instructions
  • ✅ Added missing S3 access key credentials
  • ✅ Added all required S3 bucket configurations
  • ✅ Updated Ollama endpoint configuration

2. development/docker-compose.langfuse.yml

  • ✅ Added Ollama service for embeddings and semantic cache
  • ✅ Updated MinIO to create 3 buckets: langfuse-events, langfuse-media, langfuse-batch
  • ✅ Added ollama_data volume

3. development/creds.example.env

  • ✅ Added detailed Langfuse API key setup instructions
  • ✅ Added support for headless initialization (optional)

4. network-agent/docker-compose.yml (bonus fix)

  • ✅ Added missing media and batch export S3 configurations
  • ✅ Updated MinIO bucket creation

Next Steps

  1. Copy the example creds file (if you haven't already): 

    cd development
cp creds.example.env creds.env

  2. Leave Langfuse keys empty initially in creds.env: 

    LANGFUSE_PUBLIC_KEY=""
LANGFUSE_SECRET_KEY=""

  3. Start the services: 

    invoke debug

  4. Pull Ollama embedding model (after services are up): 

    docker exec -it <ollama-container-name> ollama pull mxbai-embed-large

To find the Ollama container name: 

docker ps | grep ollama

  1. Generate Langfuse API Keys:
  2. Access Langfuse UI: http://localhost:8000
  3. Create an account or login
  4. Go to: Settings → API Keys
  5. Click "Create new key"

  6. Copy the keys

  7. Update creds.env with the generated keys: 

    LANGFUSE_PUBLIC_KEY="pk-lf-xxxxxxxxxxxxxxxx"
LANGFUSE_SECRET_KEY="sk-lf-xxxxxxxxxxxxxxxx"

  8. Restart Nautobot to apply the keys: 

    docker-compose restart nautobot worker beat

Option B: Headless Initialization (For Automation/CI)

Uncomment and configure these variables in creds.env:

# Langfuse Headless Initialization
LANGFUSE_INIT_ORG_NAME="Nautobot AI Ops"
LANGFUSE_INIT_ORG_ID="nautobot-ai-ops"
LANGFUSE_INIT_PROJECT_NAME="Development"
LANGFUSE_INIT_PROJECT_ID="dev"
LANGFUSE_INIT_PROJECT_PUBLIC_KEY=""  # Auto-generated
LANGFUSE_INIT_PROJECT_SECRET_KEY=""  # Auto-generated
LANGFUSE_INIT_USER_EMAIL="admin@example.com"
LANGFUSE_INIT_USER_NAME="Admin"
LANGFUSE_INIT_USER_PASSWORD="changeme123!"

Then copy the auto-generated keys from logs to LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY.

Verification

1. Check Langfuse Connection

Look for this log in Nautobot: 

[deep_agent] ✓ Langfuse observability enabled

Instead of: 

[deep_agent] ⚠ Failed to initialize Langfuse: <error>

2. Check Semantic Cache

Look for this log: 

[deep_agent] Semantic caching enabled (ttl=3600s, threshold=0.05)

Instead of: 

[SEMANTIC_CACHE] ⚠ Initialization failed: Failed to connect to Ollama

3. Check MinIO Buckets

Access MinIO console: http://localhost:9003 - Username: minio - Password: miniosecret

Verify 3 buckets exist: - langfuse-events - langfuse-media - langfuse-batch

Troubleshooting

Issue: Langfuse still shows "Invalid credentials"

Solution: 1. Make sure you've generated keys from the UI (Step 5 above) 2. Verify keys are set in creds.env 3. Restart the services that use Langfuse

Issue: Semantic cache still failing

Solutions: 1. Check Ollama is running: 

docker ps | grep ollama
curl http://localhost:11434/api/tags

  1. Pull the embedding model: 

    docker exec -it <ollama-container> ollama pull mxbai-embed-large

  2. Check Redis is accessible: 

    docker exec -it redis redis-cli ping
# Should return: PONG

Issue: MinIO buckets not created

Solution: 

# Recreate MinIO service
docker-compose down minio
docker volume rm <project>_minio_data
docker-compose up -d minio

# Verify buckets
docker exec -it minio-<container> ls -la /data/

Architecture Overview

┌─────────────────────────────────────────────────────────┐
│                    Nautobot AI Ops                      │
│  ┌──────────────┐     ┌──────────────────────────┐     │
│  │ Deep Agent   │────▶│  Langfuse (Port 8000)    │     │
│  │              │     │  - Web UI & API          │     │
│  │ - LLM calls  │     │  - Trace collection      │     │
│  │ - Semantic   │     └──────────┬───────────────┘     │
│  │   Cache      │                │                      │
│  └───────┬──────┘                │                      │
│          │                       │                      │
│          ▼                       ▼                      │
│  ┌──────────────┐     ┌────────────────────┐           │
│  │   Ollama     │     │    MinIO (S3)      │           │
│  │ Port: 11434  │     │    Port: 9002/9003 │           │
│  │              │     │                    │           │
│  │ - Embeddings │     │ - langfuse-events  │           │
│  │ - mxbai-     │     │ - langfuse-media   │           │
│  │   embed-     │     │ - langfuse-batch   │           │
│  │   large      │     └────────────────────┘           │
│  └──────────────┘                                       │
│                                                          │
│  ┌──────────────┐     ┌────────────────────┐           │
│  │   Redis      │────▶│   PostgreSQL       │           │
│  │ Port: 6379   │     │   Port: 5432       │           │
│  │              │     │                    │           │
│  │ - Cache      │     │ - Nautobot DB      │           │
│  │ - Semantic   │     │ - Langfuse DB      │           │
│  │   Cache      │     └────────────────────┘           │
│  └──────────────┘                                       │
│                       ┌────────────────────┐           │
│                       │   ClickHouse       │           │
│                       │   Port: 8123/9000  │           │
│                       │                    │           │
│                       │ - Analytics DB     │           │
│                       │ - Traces storage   │           │
│                       └────────────────────┘           │
└─────────────────────────────────────────────────────────┘

References

Summary

All configuration issues have been resolved: - ✅ Langfuse authentication properly configured - ✅ Ollama service added for semantic cache - ✅ All required S3/MinIO buckets configured - ✅ Missing access keys added - ✅ Documentation updated

Next: Follow Option A or B above to complete the setup!