Deploy and Host PipesHub on Railway

About Hosting PipesHub

PipesHub is an open-source AI context layer for enterprise search, retrieval, connectors, and agentic workflows. This template deploys the stable v0.6.0 release with MongoDB, Redis Streams, Qdrant, and Neo4j. The PipesHub service owns the public HTTPS domain on port 3000; all databases remain private.

On first deployment, enter PIPESHUB_ADMIN_EMAIL. The template generates the administrator password and all cross-service secrets. A bootstrap adapter creates the single initial organization before the public health gate succeeds. Sign in with the email you entered and the PIPESHUB_ADMIN_PASSWORD value shown on the PipesHub service.

Common Use Cases

Unified search and retrieval across workplace data
Knowledge-base ingestion and semantic search
Explainable, graph-backed AI answers
Connector-driven indexing and scheduled synchronization
No-code agents that do not require a local code-execution sandbox

Dependencies for PipesHub Hosting

Deployment Dependencies

MongoDB 8.0.17 for application and identity data
Redis 8.4.0 for key-value data and Redis Streams
Qdrant 1.14.1 for vectors, kept within the bundled client’s supported version window
Neo4j Community 5.26.0 with APOC for the knowledge graph
One persistent volume per stateful service
Outbound HTTPS on first boot for the embedding-model download
A user-supplied LLM provider configured after deployment

Implementation Details

The upstream all-in-one image serves the web application on port 3000. Railway service references wire private hostnames and shared secrets; changing those references can break the stack. HOME, HF_HOME, and the application volume are rooted at /data/pipeshub, preserving both local uploads and the approximately 2.5 GB model cache. Daily backups are enabled for irreplaceable volumes.

Neo4j Community logs one-time ConstraintCreationFailed messages when PipesHub probes Enterprise-only property-existence constraints. PipesHub v0.6.0 catches these as optional and continues with supported unique constraints and indexes.

Railway cannot provide the Docker socket required by PipesHub's production sandbox. The template intentionally leaves SANDBOX_MODE=docker without a daemon, so safe code-execution and artifact-generation tools are unavailable. Never use SANDBOX_MODE=local for an internet-facing deployment because it runs generated code inside the main service.

PipesHub recommends at least 4 CPU cores, 15 GB RAM, and 20 GB disk. This is not a free-tier stack. LAN-only connectors require a secure public or tunneled route, and SMTP-dependent features require post-deploy SMTP configuration.

Why Deploy PipesHub on Railway?

Railway supplies managed HTTPS, private service DNS, generated secrets, isolated persistent volumes, logs, health checks, and repeatable service wiring. This template also pins every image by stable version and digest and protects the initial-admin handoff from a public first-user race.