Railway Deployment

This page provides step-by-step instructions for deploying AutoMem on Railway, including service configuration, internal networking setup, and troubleshooting. Railway is a Platform-as-a-Service (PaaS) that provides containerized deployments with persistent volumes and internal service discovery.

For the fastest hosted setup with generated MCP config, see InstaPods Deployment. For local development setup, see the Development Guide. For backup configuration and monitoring, see Backup & Recovery and Health Monitoring. For MCP bridge setup specifically, see the MCP Integration guide.

Overview

Railway deployment provisions three services: the AutoMem Flask API (memory-service), FalkorDB graph database (falkordb), and optionally the MCP bridge for cloud AI platforms (mcp-sse-server). Railway handles container orchestration, internal networking via IPv6, persistent volumes, and health checks.

Production Deployment Architecture

graph TB
    subgraph internet["External Clients"]
        DirectAPI["Direct API Clients<br/>curl, Python, MCP package"]
        CloudAI["Cloud AI Platforms<br/>ChatGPT, Claude.ai, ElevenLabs"]
        LocalTools["Local Development Tools<br/>Cursor, Claude Desktop"]
    end

    subgraph railway["Railway Project"]
        subgraph public["Public Layer"]
            Domain1["automem.up.railway.app<br/>HTTPS/443"]
            Domain2["mcp.up.railway.app<br/>HTTPS/443"]
            LB["Railway Load Balancer<br/>SSL Termination"]
        end

        subgraph services["Service Layer"]
            subgraph api["memory-service"]
                FlaskApp["app.py<br/>Flask + Gunicorn<br/>PORT=8001"]
                Workers["Background Workers<br/>enrichment_worker<br/>embedding_worker<br/>consolidation_worker<br/>sync_worker"]
            end

            subgraph mcp["mcp-sse-server"]
                MCPBridge["server.js<br/>Express + SSE<br/>PORT=8080"]
            end

            subgraph db["falkordb"]
                FalkorDB["FalkorDB Container<br/>Port 6379<br/>FALKOR_PASSWORD"]
                Volume["Persistent Volume<br/>/var/lib/falkordb/data<br/>50GB"]
            end
        end

        subgraph networking["Internal Network"]
            DNS["*.railway.internal<br/>IPv6 DNS<br/>RAILWAY_PRIVATE_DOMAIN"]
        end

        subgraph external["External Services"]
            Qdrant["Qdrant Cloud<br/>QDRANT_URL<br/>QDRANT_API_KEY"]
            OpenAI["OpenAI API<br/>OPENAI_API_KEY"]
            Voyage["Voyage AI<br/>VOYAGE_API_KEY"]
            S3["S3 Bucket<br/>Backups"]
        end
    end

    DirectAPI -->|HTTPS| Domain1
    CloudAI -->|HTTPS/SSE| Domain2
    LocalTools -->|HTTPS| Domain1

    LB --> Domain1
    LB --> Domain2
    Domain1 --> FlaskApp
    Domain2 --> MCPBridge

    MCPBridge -->|"http://memory-service.railway.internal:8001<br/>AUTOMEM_API_URL"| FlaskApp

    FlaskApp -->|"redis://falkordb.railway.internal:6379<br/>FALKORDB_HOST"| FalkorDB
    FlaskApp -->|HTTPS| Qdrant
    FlaskApp -->|HTTPS| OpenAI
    FlaskApp -->|HTTPS| Voyage

    Workers --> FalkorDB
    Workers --> Qdrant

    FalkorDB --> Volume

    DNS -.->|Resolves| FlaskApp
    DNS -.->|Resolves| MCPBridge
    DNS -.->|Resolves| FalkorDB

Quick Start: One-Click Template

The Railway template automates service provisioning with pre-configured environment variables and networking.

What Gets Deployed

Service	Image/Source	Purpose	Public Domain
memory-service	GitHub repo root	Flask API, background workers	Yes (required)
falkordb	falkordb/falkordb:latest	Graph database with persistence	No (internal only)
mcp-sse-server	GitHub repo mcp-sse-server/	MCP bridge for cloud AI platforms	Yes (if using ChatGPT/Claude.ai)

Template Variables (Auto-Generated)

The template uses Railway’s shared variables feature to generate secrets once and reference them across services. After template deployment, add your API keys:

1. Navigate to Railway Dashboard → memory-service → Variables
2. Add required variables:
   • OPENAI_API_KEY — For embeddings and classification
   • QDRANT_URL — Qdrant Cloud endpoint (optional but recommended)
   • QDRANT_API_KEY — Qdrant authentication
3. Redeploy memory-service to apply changes

Manual Deployment

For production deployments or custom configurations, manual setup provides more control.

Step 1: FalkorDB Service

FalkorDB requires persistent volume configuration to survive restarts.

Required Environment Variables

Variable	Value	Purpose
PORT	6379	Redis protocol port
FALKOR_PASSWORD	<generated-secret>	Authentication password
REDIS_ARGS	--save 60 1 --appendonly yes --appendfsync everysec	Persistence configuration

Do not use --requirepass

Do NOT add --requirepass to REDIS_ARGS. FalkorDB handles authentication via the FALKOR_PASSWORD environment variable. Adding --requirepass causes authentication to fail.

Volume Configuration

• Mount Path: /var/lib/falkordb/data (not /data)
• Minimum Size: 1GB (adjust based on expected data)
• Purpose: Persists graph data across container restarts

Step 2: Memory Service

The memory-service runs the Flask API, background workers, and handles all memory operations.

Build Configuration

Builder: DOCKERFILE
Dockerfile Path: ./Dockerfile
Root Directory: (empty - use repo root)

PORT is required

PORT=8001 is required. Without this variable, Flask defaults to port 5000, causing connection failures from other services.

Variable Reference Patterns

Railway supports two patterns for environment variables:

Pattern	Example	Use Case	Stability
Hardcoded	FALKORDB_HOST=falkordb.railway.internal	Production, manual setup	Stable, easier to debug
Reference	FALKORDB_HOST=${{FalkorDB.RAILWAY_PRIVATE_DOMAIN}}	Templates only	Updates automatically, harder to troubleshoot

Tip

Use hardcoded values for manual deployments. Variable references only work in Railway templates.

Health Check Configuration

• Path: /health
• Timeout: 100 seconds
• Expected Response: {"status": "healthy"}

Step 3: MCP Bridge (Optional)

The mcp-sse-server service is only needed for cloud AI platforms (ChatGPT, Claude.ai, ElevenLabs). Skip this if you only use Cursor, Claude Desktop, or direct API access.

Build Configuration

Builder: DOCKERFILE
Dockerfile Path: mcp-sse-server/Dockerfile
Root Directory: mcp-sse-server

Environment Variables

Variable	Required	Default	Description
PORT	Yes	8080	HTTP server port
AUTOMEM_API_URL	Yes	None	Internal URL: http://memory-service.railway.internal:8001
AUTOMEM_API_TOKEN	Yes	None	Copy from memory-service

Networking Architecture

Railway uses IPv6-based internal networking with automatic DNS resolution.

Health Check Sequence

sequenceDiagram
    participant Client
    participant API as "app.py<br/>/health"
    participant Falkor as "FalkorDB<br/>redis_ping()"
    participant Qdrant as "Qdrant<br/>client.get_collections()"

    Client->>API: GET /health

    API->>Falkor: Test connection<br/>redis_ping()
    alt FalkorDB Available
        Falkor-->>API: PONG
        API->>API: Set falkordb: connected
    else FalkorDB Unavailable
        Falkor-->>API: Exception
        API->>API: Set falkordb: unavailable
        API-->>Client: 503 Service Unavailable
    end

    API->>Qdrant: Test connection<br/>get_collections()
    alt Qdrant Available
        Qdrant-->>API: Collections list
        API->>API: Set qdrant: connected
    else Qdrant Unavailable
        Qdrant-->>API: Exception
        API->>API: Set qdrant: unavailable<br/>(continue anyway)
    end

    API->>Falkor: Count memories<br/>MATCH (m:Memory) RETURN count(m)
    Falkor-->>API: count

    API->>API: Check enrichment queue<br/>ServiceState.enrichment_queue

    API-->>Client: 200 OK<br/>{status, falkordb, qdrant,<br/>memory_count, enrichment}

IPv6 Dual-Stack Binding

Railway’s internal networking uses IPv6 addresses. Services must bind to :: (dual-stack) instead of 0.0.0.0 (IPv4 only). AutoMem v0.7.1+ handles this automatically — Flask binds to host="::" at startup.

If you see this in startup logs, the service is correctly bound:

* Running on http://[::]:8001

If you see the following (old versions only), upgrade to v0.7.1+:

* Running on http://0.0.0.0:8001

Internal Service Discovery

Services communicate using internal hostnames:

Service Name	Internal Hostname	Port	Usage
memory-service	memory-service.railway.internal	8001	MCP bridge connects here
falkordb	falkordb.railway.internal	6379	Memory service connects here

The internal hostname is automatically set by Railway as <service-name>.railway.internal.

TCP Proxy for External Access

TCP Proxy enables external services (like GitHub Actions) to access internal Railway services.

When TCP Proxy Is Required

Use Case	Needs TCP Proxy?	Reason
Internal service-to-service	No	Use .railway.internal hostnames
GitHub Actions backups	Yes	External runners can’t access internal network
Local development testing	Yes	External connection to Railway database
Public API access	No	Use public domains

Configuration Steps

1. Railway Dashboard → falkordb service
2. Settings → Networking → Enable TCP Proxy
3. Note the generated endpoint:
   • RAILWAY_TCP_PROXY_DOMAIN: e.g., monorail.proxy.rlwy.net
   • RAILWAY_TCP_PROXY_PORT: e.g., 12345 (random high port)

Usage in GitHub Actions

For GitHub Actions backups, set these secrets:

FALKORDB_HOST = monorail.proxy.rlwy.net    # TCP Proxy domain
FALKORDB_PORT = 12345                       # TCP Proxy port (not 6379!)
FALKORDB_PASSWORD = <same-as-railway>

The backup workflow validates connectivity before running and checks that FALKORDB_HOST is not a *.railway.internal hostname (which external runners cannot reach).

Post-Deployment Verification

After deploying all services, verify connectivity and data flow.

Health Check

curl https://your-project.up.railway.app/health

Expected response:

{
  "status": "healthy",
  "falkordb": "connected",
  "qdrant": "connected",
  "memory_count": 0,
  "enrichment": {
    "status": "running",
    "queue_depth": 0
  }
}

Store Test Memory

curl -X POST https://your-project.up.railway.app/memory \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"content": "Test memory for deployment verification"}'

Verify Background Workers

Check enrichment queue processing via /health — the processed count should increase after storing memories.

Verification Checklist

• Health endpoint returns "status": "healthy"
• FalkorDB shows "connected"
• Qdrant shows "connected" (if configured)
• Test memory stores successfully
• Enrichment worker processes memories (processed count increases)
• MCP bridge responds on /health (if deployed)

Configuration Test Patterns

graph TB
    subgraph "Unit Tests (make test)"
        UnitConfig["No external services<br/>Mock configuration<br/>Fast execution"]

        MockFalkor["DummyGraph fixture<br/>In-memory stubs"]
        MockQdrant["Mock Qdrant client<br/>No network calls"]
        MockProvider["Mock embedding provider<br/>Deterministic outputs"]
    end

    subgraph "Integration Tests (make test-integration)"
        IntConfig["Real services via Docker<br/>Actual configuration<br/>Comprehensive validation"]

        DockerServices["docker-compose up<br/>FalkorDB + Qdrant"]
        RealConfig["Load from .env<br/>or environment"]
        TestAPICalls["HTTP requests to API<br/>AUTOMEM_TEST_BASE_URL"]
    end

    subgraph "Live Tests (make test-live)"
        LiveConfig["Railway deployment<br/>Production-like environment<br/>Safety checks"]

        ConfirmPrompt["Interactive confirmation<br/>AUTOMEM_ALLOW_LIVE=1"]
        RailwayAPI["https://automem.up.railway.app"]
        CleanupTests["Unique UUIDs<br/>Self-cleaning"]
    end

    UnitConfig --> MockFalkor
    UnitConfig --> MockQdrant
    UnitConfig --> MockProvider

    IntConfig --> DockerServices
    DockerServices --> RealConfig
    RealConfig --> TestAPICalls

    LiveConfig --> ConfirmPrompt
    ConfirmPrompt --> RailwayAPI
    RailwayAPI --> CleanupTests

Environment Variables Reference

memory-service Variables

Variable	Required	Default	Description
PORT	Yes (Railway)	5000	Flask listen port. Must be 8001 on Railway
FALKORDB_HOST	Yes	localhost	FalkorDB hostname. Use falkordb.railway.internal on Railway
FALKORDB_PORT	Yes	6379	FalkorDB port
FALKORDB_PASSWORD	Yes	None	FalkorDB authentication password
FALKORDB_GRAPH	No	memories	Graph name in FalkorDB
AUTOMEM_API_TOKEN	Yes	None	API authentication token
ADMIN_API_TOKEN	Yes	None	Admin endpoint authentication
OPENAI_API_KEY	Recommended	None	Required for embeddings and classification
QDRANT_URL	Recommended	None	Qdrant Cloud endpoint
QDRANT_API_KEY	Recommended	None	Qdrant authentication
QDRANT_COLLECTION	No	memories	Qdrant collection name

falkordb Variables

Variable	Required	Default	Description
PORT	Yes	6379	Redis protocol port
FALKOR_PASSWORD	Yes	None	Database authentication
REDIS_ARGS	Recommended	None	Persistence config: --save 60 1 --appendonly yes --appendfsync everysec

mcp-sse-server Variables

Variable	Required	Default	Description
PORT	Yes	8080	HTTP server port
AUTOMEM_API_URL	Yes	None	Internal URL: http://memory-service.railway.internal:8001
AUTOMEM_API_TOKEN	Yes	None	Copy from memory-service

Troubleshooting

ECONNREFUSED on Port 8001

Symptoms

Error: connect ECONNREFUSED fd12:ca03:42be:0:1000:50:1079:5b6c:8001

MCP bridge or other services cannot connect to memory-service.

Solution 1: Add PORT Variable

Most common cause. Flask defaults to port 5000 without explicit PORT variable. Add PORT=8001 to memory-service variables in Railway Dashboard, then redeploy.

Solution 2: Update IPv6 Binding

If using older AutoMem versions (before v0.7.1), the service bound only to IPv4. Update to AutoMem v0.7.1 or later — Flask now binds to host="::" for dual-stack support.

Solution 3: Verify Internal Hostname

Ensure the MCP bridge uses AUTOMEM_API_URL=http://memory-service.railway.internal:8001 (internal hostname, not the public domain).

Variable Reference Not Resolving

Symptoms

Logs show literal ${{...}} strings instead of resolved values.

Cause

Railway variable references only work in templates, not manual service configuration.

Solution

Replace variable references with hardcoded values:

# Instead of:
FALKORDB_HOST=${{FalkorDB.RAILWAY_PRIVATE_DOMAIN}}

# Use:
FALKORDB_HOST=falkordb.railway.internal

FalkorDB Data Loss After Restart

Symptoms

All memories disappear after FalkorDB service restarts. /health shows memory_count: 0.

Cause

No persistent volume configured on FalkorDB service.

Solution

1. Add Volume (prevents future data loss)
   • Railway Dashboard → falkordb service
   • Settings → Volumes → Add Volume
   • Mount Path: /var/lib/falkordb/data
   • Redeploy service
2. Recover Existing Data (if Qdrant is intact)
   • Run scripts/recover_from_qdrant.py — see Backup & Recovery

GitHub Actions Backup Fails

Symptoms

Error: Connection reset by peer (error 104)

Backup workflow cannot connect to FalkorDB.

Cause

TCP Proxy not enabled or incorrect secrets configuration.

Solution

1. Enable TCP Proxy (if not already enabled)
   • Railway Dashboard → falkordb service
   • Settings → Networking → Enable TCP Proxy
   • Note the public endpoint
2. Update GitHub Secrets
   • Go to GitHub repo → Settings → Secrets and variables → Actions
   • Set these values:

     FALKORDB_HOST = monorail.proxy.rlwy.net    # TCP Proxy domain
     FALKORDB_PORT = 12345                       # TCP Proxy port (not 6379!)
     FALKORDB_PASSWORD = <same-as-railway>

3. Verify Connectivity — the backup workflow includes pre-flight checks that validate the TCP proxy connection before attempting backup

Cost Optimization

Railway uses usage-based pricing with resource limits.

Service Sizing Recommendations

Service	RAM	vCPU	Storage	Est. Cost/Month
memory-service	512MB	0.5	-	~$5
falkordb	1GB	1.0	2GB volume	~$10
mcp-sse-server	256MB	0.25	-	~$2-3

Total Estimated Cost: $17-18/month (Railway Pro plan required: $20/month base)

Cost-Saving Options

1. Remove mcp-sse-server if only using Cursor/Claude Desktop
   • Saves ~$2-3/month
   • No impact on core API functionality
   • Delete service from Railway Dashboard
2. Use Qdrant Cloud Free Tier
   • 1GB storage free
   • Upgrade to $25/month for 10GB if needed
3. Start with Smaller Volumes
   • FalkorDB: 1GB initially (expand as needed)
   • Volume resizing is non-disruptive

Security Best Practices

1. Always set FALKOR_PASSWORD
   • Railway auto-generates this in templates
   • For manual setup, use a strong random value
2. Use Internal Networking
   • Service-to-service communication via .railway.internal hostnames
   • Never expose FalkorDB publicly (no public domain)
3. Rotate API Tokens Periodically
   • Update AUTOMEM_API_TOKEN and ADMIN_API_TOKEN via Railway Dashboard
   • Redeploy affected services
4. Disable Public Domains on FalkorDB
   • Only memory-service and mcp-sse-server need public access
   • FalkorDB should only be accessible via internal network or TCP Proxy
5. Service Naming Stability
   • Internal DNS is based on service name (<name>.railway.internal)
   • Renaming a service updates its RAILWAY_PRIVATE_DOMAIN
   • Update hardcoded hostnames in other services after renaming

Next Steps

After successful deployment:

• Configure automated backups — see Backup & Recovery
• Set up health monitoring — see Health Monitoring
• Test disaster recovery procedures — see Backup & Recovery
• Review environment configuration — see the Configuration Reference