Deploy WeKnora RAG + ParadeDB

Deploy and Host WeKnora RAG + ParadeDB on Railway

WeKnora is Tencent's open source LLM knowledge platform. Upload PDFs, Office files, web pages and notes, and it parses them, builds a hybrid knowledge base (dense vectors plus BM25 keyword search), answers questions with citations back to the source, runs reasoning agents over your documents, and can keep a wiki up to date from them. Workspaces, a four level role model and API keys make it a team tool, and it can publish each knowledge base as an MCP server.

About Hosting WeKnora RAG + ParadeDB

Self-hosting WeKnora means five containers: a web front door, the Go application, a document parsing service, Postgres and Redis. This template deploys all five from Tencent's official images pinned to v0.8.2, wires them over Railway's private network, generates every secret, and gives you one public URL for the UI and API.

Postgres is ParadeDB, which is what WeKnora's default retrieval engine needs: vectors (pgvector) and keyword indexes (pg_search BM25) live in the same database, so there is no separate vector store to run or pay for. Uploaded files are kept on a volume attached to the application.

The template closes public signup and creates your admin account on the first boot, on an instance that never listens beyond its own container. Upstream ships signup open with no default account, which on a public URL means the first stranger to register owns the instance.

Common Use Cases

• Chat with your documents: contracts, manuals, papers and reports, answered with citations
• Team knowledge base: workspaces per team, Owner, Admin, Contributor and Viewer roles, invite links
• RAG API for your own app: workspace API keys for search and chat endpoints
• Knowledge as MCP tools: expose a knowledge base to Cursor, Claude or other MCP clients
• Self-maintaining wiki: let the agent turn a document collection into linked wiki pages

Dependencies for WeKnora RAG + ParadeDB Hosting

• ParadeDB Postgres 17: users, workspaces, documents, pgvector embeddings and pg_search BM25 (included)
• Redis: background task queue and chat streams (included)
• DocReader: document parsing service with LibreOffice and a headless browser (included)
• An LLM provider: not needed to boot or log in, but a chat model and an embedding model must be added in the UI before anything can be indexed or answered. OpenAI compatible endpoints, 27 built-in vendors and Ollama are supported.

Deployment Dependencies

• WeKnora GitHub repository
• Configuration reference
• Workspaces, roles and authentication
• API reference

First steps after deploy

1. Wait for all five services to go green. The first boot runs the database migrations and creates your admin account, so give the application a few minutes.
2. Copy WEKNORA_ADMIN_PASSWORD from the WeKnora App service variables. Your login email is WEKNORA_ADMIN_EMAIL (default admin@weknora.local).
3. Open the WeKnora public domain, log in, and change the password from your profile.
4. Create a knowledge base and add a chat model and an embedding model in the setup wizard.
5. Upload documents, wait until they show as completed, and ask questions.
6. Invite teammates from Workspace settings, Members. Signup stays closed; invite links still work.

Key environment variables

WeKnora App service:

WEKNORA_ADMIN_EMAIL=admin@weknora.local
WEKNORA_ADMIN_PASSWORD=Wk1${{secret(29)}}
DISABLE_REGISTRATION=true
JWT_SECRET=${{secret(64,"abcdef0123456789")}}
SYSTEM_AES_KEY=${{secret(32)}}
RETRIEVE_DRIVER=postgres
STORAGE_TYPE=local
DB_HOST=${{Postgres.RAILWAY_PRIVATE_DOMAIN}}
REDIS_ADDR=${{Redis.RAILWAY_PRIVATE_DOMAIN}}:6379
DOCREADER_ADDR=${{DocReader.RAILWAY_PRIVATE_DOMAIN}}:50051
WEKNORA_SANDBOX_DOCKER_ENABLED=false

SYSTEM_AES_KEY encrypts the model keys you store, so never change it after the first boot. The Docker sandbox for agent skills stays off because Railway has no Docker socket; knowledge base chat, search and agents without skills do not need it, and E2B or CubeSandbox can be configured per workspace.

Why Deploy WeKnora RAG + ParadeDB on Railway?

Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.

WeKnora is a five service stack with a few traps when moved off Docker Compose: it needs ParadeDB rather than plain Postgres, its nginx pins the backend address, and its first admin flow expects open signup. This template handles all of that: every service runs on Railway's private network, every secret is generated per deployment, the front door keeps working when the backend redeploys, signup is closed from the first second, and your admin account is waiting for you. You get managed TLS and one public URL, and nothing else is exposed to the internet.