Sushi Kitchen
Temaki
Temaki

Flowise

temaki.flowise

A Docker Compose file for Flowise alone, generated from the catalog and checked by the safety inspector. No account needed.

Visual builder for language-model applications, wiring retrieval, memory, tools and model nodes on a canvas and exposing the finished flow as an API endpoint or embeddable chat widget.

Flowise represents a language-model application as a graph of nodes — loaders, splitters, embedding models, vector stores, memory, tools, chat models — connected on a canvas, which makes the structure of a retrieval or agent pipeline visible rather than buried in code. Each saved flow is served immediately as an API endpoint with a generated embeddable chat widget, so a prototype becomes callable without a deployment step. Credentials are stored per connection and reused across flows, and execution traces show what each node received and returned. Agent flows add tool calling and multi-step reasoning over the same node model.

You know it worked when

  • The canvas loads and authentication is required to reach it.
  • A flow built from a model node and a prompt returns a completion in the test pane.
  • A document indexed into a vector store node is retrieved as context.
  • The saved flow answers a request to its generated API endpoint.
  • Flows and credentials are present after a container restart.

Known sharp edges

  • Flow definitions, credentials and chat history live in the configured database, and the default embedded store on a non-persistent volume loses everything on recreation.
  • There is no authentication unless username and password are set through the environment, and an unprotected instance exposes every stored credential.
  • A node can execute arbitrary code and reach any endpoint the container can, so edit access to a flow is effectively execution access.
  • Enterprise-licensed features appear in the interface and fail only when activated, which makes the licence boundary easy to cross unintentionally.
agent-builderllmworkflow