How to Host Flowise on a VPS

Flowise handles sensitive data: OpenAI/Anthropic API keys, vectorised internal documents, conversation history, the business logic of your agents. On the shared cloud version, these elements pass through infrastructure you do not control. On a dedicated VPS, the Flowise canvas, the database and the embeddings all stay with you. You keep control of versions (flows evolve fast between releases), of the connectors you enable, and of model billing since you plug in your own keys. A VPS also lets you expose flows as a REST API to integrate them into your own applications, with no quota imposed by a vendor.

Another practical reason: Flowise Cloud plans limit the number of active flows, monthly executions and access to advanced nodes. On your VPS, none of these limits apply — you define the allocated resources yourself.

Flowise itself is lightweight, but real-world usage depends on embeddings and vector stores. Budget at least 2 vCPUs and 2 GB of RAM for testing, and 4 vCPUs / 4 to 8 GB of RAM if you index large documents or run several concurrent flows. Plan for 20 GB of disk for the containers, vector stores and any local models.

On the software side: Docker and Docker Compose installed, a domain or subdomain (e.g. flowise.your-domain.com) pointing to the VPS IP, and port 443 open. If you plan to run a local LLM via Ollama, increase RAM to at least 8 GB (16 GB for 7B+ models at 4-bit quantisation). Prepare the API keys of the LLM providers you intend to use ahead of time.

For high-performance RAG flows with no external API cost, connect Flowise to an Ollama container running on the same VPS through the internal Docker network (http://ollama:11434). You thus combine embeddings and generation locally: no data leaves the server, and latency drops since everything travels over the host's private network.

In your docker-compose.yml, add the Ollama service on the same Docker network as Flowise:

services:
  ollama:
    image: ollama/ollama
    volumes:
      - ollama_data:/root/.ollama
    networks:
      - flowise_net

  flowise:
    image: flowiseai/flowise
    networks:
      - flowise_net
    ...

In Flowise, when you add an Ollama node (Chat Model or Embeddings), enter http://ollama:11434 as the base URL. Pull a model from inside the container: docker exec -it <ollama_id> ollama pull llama3.2. For embeddings, nomic-embed-text is a good starting point on 4 GB of RAM.

Flowise exposes around twenty configuration variables. The most useful ones in production:

FLOWISE_USERNAME and FLOWISE_PASSWORD — authentication for the admin interface. Mandatory as soon as the public port is open.

FLOWISE_SECRETKEY_OVERWRITE — encryption key for credentials stored in the database. If you change this value after first startup, existing credentials become unreadable.

DATABASE_TYPE — defaults to sqlite. For heavier load or a redundant deployment, switch to postgres and fill in DATABASE_HOST, DATABASE_PORT, DATABASE_USER, DATABASE_PASSWORD, DATABASE_NAME.

CORS_ORIGINS — list of origins allowed to call the prediction API from a browser. Example: https://your-app.com. Without this variable, CORS calls from another origin will be blocked.

BLOB_STORAGE_PATH — storage path for uploaded files (RAG documents). Defaults to ~/.flowise/storage; mount a dedicated volume if you index large corpora.

LOG_LEVEL — values: error, warn, info, verbose, debug. In production, warn is enough to keep logs manageable.

Flowise releases regular updates that add nodes and fix flow behaviour. The update is non-destructive if your data is in a named volume.

docker compose pull
docker compose up -d

Flowise automatically migrates the SQLite database on startup. Before each update, export your important flows via the canvas Export button (JSON file) — it is the simplest safety net.

If you use PostgreSQL, check the release notes before pulling the image: some versions introduce irreversible schema migrations. Keep a database backup (pg_dump) before updating.

The canvas does not load or nodes disappear. Check that your reverse proxy properly passes WebSockets. With nginx, the Upgrade and Connection headers must be present. With Caddy, they are set automatically.

Error: FLOWISE_SECRETKEY_OVERWRITE is required. This variable was not required in earlier versions but has become so. Add it to your compose and restart — credentials already entered remain valid as long as the value matches the one used when they were stored.

The vector store loses its documents after restart. Check that the Docker volume path matches both BLOB_STORAGE_PATH and DATABASE_PATH. A misconfigured bind-mount stores data in an ephemeral container folder.

Calls to Ollama fail with ECONNREFUSED. The Ollama service must be on the same Docker network as Flowise. If Ollama runs outside Compose (systemd service, for instance), use the Docker gateway IP (172.17.0.1:11434) rather than localhost.

OpenAI/Anthropic credentials are rejected after migration. If FLOWISE_SECRETKEY_OVERWRITE changed between two deployments, secrets encrypted with the old key can no longer be decrypted. Delete and re-enter the credentials in the interface.