What you get
- Models, all through OpenRouter: Claude Sonnet 5.5 answers, Claude Haiku 5.5 rephrases questions and summarises documents, OpenAI
text-embedding-3-smallembeds them. - Ragen Brain is switched on from the first boot.
- Every secret is generated by the template — session secrets, the encryption key, database and Redis passwords.
web,mcpandadminget public Railway domains. Everything else, Qdrant included, is reachable only on Railway’s private network.
Steps
1
Get an OpenRouter key
Create one at openrouter.ai/keys. One key serves both chat and embeddings, so the knowledge base works from your first upload.
2
Deploy the template
Click Deploy on Railway above, paste the key into
OPENROUTER_API_KEY, and click Deploy. Docling is the one service built from source; the rest are published images, so the first deploy takes a few minutes.3
Create the first account
Open the
web service’s public domain. The first account you create becomes the platform administrator, and the template ships no demo credentials.4
Ask your first question
Upload a document under Knowledge base, wait until it is processed, and ask about it in a new chat. Answers cite the passages they came from.
Before you rely on it
What the template deliberately leaves out:- PII masking. There is no Presidio in the template, so personal data is not masked before it reaches the model. See PII masking to add it.
- Content moderation. It needs an OpenAI key (
OPENAI_MODERATION_KEY). Without one, the content-moderation guardrail logs that it is not configured and lets turns through. - Mail. Verification links and invitations are written to the log until you set
RESEND_API_KEYor theSMTP_*variables onweb. - Connectors. The MCP connector services (Google, ClickUp, HubSpot) are not part of it; the token vault they need is.
local — a key in an environment variable. For production, consider a KMS; see Encryption.
Next steps
Ragen Brain
Turn your documents into reviewed, published knowledge pages.
MCP server
Connect Claude Desktop or Cursor to your assistants.
Models
Use a different provider or model.
Self-hosting
Run the same stack on your own infrastructure.