Onyx v4 Enterprise AI Search

Deploys the full Onyx v4 stack for AI-powered enterprise search and chat: API server, web app, background workers, two local embedding model servers, OpenSearch, Postgres, Redis and S3-compatible file storage. Connect sources such as Slack, Google Drive or Confluence and bring your own LLM provider.

What you get

  • Onyx services (API, background Celery workers, web, inference and indexing model servers) pinned to v4.9.0, behind nginx
  • OpenSearch 3.6.0 as the hybrid keyword and vector index, on a volume
  • Postgres 18 and Redis 8.2 on volumes, plus SeaweedFS as the S3-compatible file store
  • Generated secrets, healthchecks and private networking between all ten services
  • Enterprise features left disabled

Good for

  • Ask questions across Slack, Google Drive, Confluence, Notion, Jira or GitHub and get cited answers
  • Run a private chat assistant that can use internal documents and custom agents
  • Build assistants scoped to document sets for support, sales enablement or engineering
  • Keep indexed company data and chat history on infrastructure you control

Services in this template

  • Redis · image redis:8.2
  • Postgres · image ghcr.io/railwayapp-templates/postgres-ssl:18
  • nginx · from baranberkay96/onyx-railway
  • opensearch · from baranberkay96/onyx-railway
  • indexing-model-server · image onyxdotapp/onyx-model-server:v4.9.0
  • api-server · image onyxdotapp/onyx-backend:v4.9.0
  • inference-model-server · image onyxdotapp/onyx-model-server:v4.9.0
  • SeaweedFS · image chrislusf/seaweedfs:4.48
  • web-server · image onyxdotapp/onyx-web-server:v4.9.0
  • background · image onyxdotapp/onyx-backend:v4.9.0

Questions people ask

How do I log in?

Wait for api-server to become healthy (first start runs migrations and can take up to 15 minutes), open the nginx public URL and create an account. The first account becomes the administrator. Then add an LLM provider and connectors in the admin panel.

What drives the cost?

Memory. The template is resource-intensive: its author plans for roughly 8 to 14 GB of RAM across the services (an estimate, not a measurement). The background service, both model servers and OpenSearch are the largest. Hobby is the minimum plan; Pro is recommended for production. LLM usage is billed by your provider.

Is data persistent?

OpenSearch, Postgres, Redis and SeaweedFS each use a volume. To use fewer services, set FILE_STORE_BACKEND=postgres on api-server and background before first use and remove SeaweedFS.

How do I restrict sign-ups or upgrade?

Set VALID_EMAIL_DOMAINS on api-server or configure SSO in the admin panel. For upgrades, move all Onyx services (api-server, background, web-server and both model servers) to the same new tag and read the release notes first; migrations run when api-server starts.

Which licence applies?

Onyx is MIT outside its ee directories; those parts use the Onyx Enterprise License and stay disabled here. OpenSearch, nginx and SeaweedFS are permissively licensed. This is a community template, not affiliated with Onyx.

Full template README

Deploy and Host Onyx with Railway

Onyx is an open-source AI platform for enterprise search and chat: it connects to tools like Slack, Google Drive, Confluence, Jira and GitHub, indexes their content and answers questions with citations using the LLM of your choice. This community template deploys the full Onyx v4 stack on Railway, including connectors, document indexing and local embedding models.

About Hosting Onyx

A complete Onyx deployment has many moving parts. An API server and a Next.js web app sit behind nginx; a background service runs Celery workers that pull documents from connectors, split and embed them, sync permissions and keep the index fresh. Two model servers run the embedding models locally, OpenSearch stores the hybrid keyword and vector index, Postgres keeps users, connectors and chat history, Redis brokers the task queues and holds sessions, and an S3-compatible store keeps uploaded files. This template wires all of it together over Railway's private network with generated secrets, persistent volumes and healthchecks. It is resource-intensive: plan for roughly 8–14 GB of RAM.

Common Use Cases

  • Ask natural-language questions across company knowledge in Slack, Google Drive, Confluence, Notion, Jira, GitHub and other sources, with cited answers.
  • Give teams a private ChatGPT-style assistant that can use internal documents, web search and custom agents.
  • Build agents and assistants scoped to document sets for support, sales enablement or engineering.
  • Keep indexed company data and chat history on infrastructure you control.

Dependencies for Onyx Hosting

  • Onyx images onyxdotapp/onyx-backend, onyx-web-server and onyx-model-server, version v4.9.0
  • OpenSearch 3.6.0 (document index)
  • PostgreSQL (Railway Postgres 18) and Redis (Railway Redis 8.2)
  • S3-compatible storage (SeaweedFS 4.48)
  • nginx 1.30 as the front door
  • An LLM provider (OpenAI, Anthropic, Azure, Bedrock, Ollama, a LiteLLM gateway, ...), configured in the admin panel

Deployment Dependencies

Implementation Details

Service Role Public Storage
nginx Routes /api to the API and the rest to the web app Yes (port 80) –
api-server REST/streaming API, runs migrations No –
background Celery workers + beat (connectors, indexing, pruning, permission sync) No –
web-server Next.js UI No –
inference-model-server Query embeddings, reranking No –
indexing-model-server Document embeddings No –
opensearch Hybrid search index No Volume
Postgres Users, connectors, chat history No Volume
Redis Task broker, cache, sessions No Volume
SeaweedFS S3 file store for uploads No Volume

First login

  1. Wait for api-server to become healthy (the first start runs database migrations and can take up to 15 minutes).
  2. Open the nginx public URL and create your account. The first account becomes the administrator.
  3. In the admin panel, add an LLM provider, then add connectors (Web, Slack, Google Drive, Confluence, ...) and let them index.
  4. To restrict sign-ups, set VALID_EMAIL_DOMAINS on api-server or configure SSO under Admin Panel > Organization > SSO Providers.

Scaling: indexing throughput depends on background and indexing-model-server CPU/RAM; give them more resources for large sources. Search and chat load scales with api-server and inference-model-server. OpenSearch scales vertically (raise OPENSEARCH_JAVA_OPTS together with available RAM).

Pinning and upgrades: all Onyx services use the same tag (v4.9.0); upgrade api-server, background, web-server and both model servers together and read the release notes first. Migrations run automatically when api-server starts. OpenSearch is pinned in services/opensearch/Dockerfile to the version Onyx's compose file uses.

File storage: uploads use SeaweedFS because Onyx's S3 client requires path-style requests. If you prefer fewer services, set FILE_STORE_BACKEND=postgres on api-server and background before first use and remove SeaweedFS.

Why Deploy Onyx on Railway?

Railway runs Onyx's ten services as one project with private networking, persistent volumes, generated credentials and usage-based billing, so you get the full connector-and-indexing stack without operating Kubernetes. Resources can be raised per service as your indexed content grows.

Guides for this template

View Onyx v4 Enterprise AI Search on railway.com

Similar templates

TemplateCategoryDeploymentsDeploys
9Router
Self-hosted AI router for Claude, Gemini, Cursor & OpenAI tools.
AI/ML9,734
flowise-railway
Flowise - low-code LLM apps builder
AI/ML8,512
LibreChat
Open-source ChatGPT clone: multi-model, agents, search, multi-user auth
AI/ML6,762
Hermes Agent by Nous Research
Template for Hermes Agent by Nous Research
AI/ML6,413
OpenClaw
[Homebrew support] The AI that actually does things
AI/ML5,167
Hermes Agent
Self Host Hermes Agent. Autonomous AI agent w/ self-improving, tools & more
AI/ML4,092