LLM chains and agents on a canvas, now archived
Flowise is no longer maintained: its team archived the repository in August 2026 and ended all support on August 31. Six security advisories published on September 10, 2026, two of them critical, affect every version, and no released version fixes them.
Flowise is a web app for building LLM chains and agents by wiring nodes on a canvas: a model, a retriever, memory, tools, a JavaScript function. Every flow gets its own API endpoint (/api/v1/prediction/{id}, with optional streaming) and can sit on any website as a chat bubble through the flowise-embed script.
There are three editors. Assistant builds an agent that answers from uploaded files. Chatflow covers chains and single agents. Agentflow handles multi-agent setups and workflows with branches and loops.
Dify does the same job on a canvas and shipped version 1.17.1 on September 10, 2026, while Flowise has released nothing since July 29. The comparison of open source workflow automation tools sets it against other AI agent builders and classic workflow tools.
LangChain blocks as nodes, with an API for every flow
- LangChain and LlamaIndex as nodes: the repo's
componentspackage depends onlangchainandllamaindex, and each building block (loader, splitter, embeddings, vector store, retriever, memory) becomes a configurable node. A RAG pipeline means connecting those nodes, with no chain code to write. - The serving layer LangChain leaves to you: Flowise wraps each chain in a per-flow API, a chat widget, an SDK and a CLI, and adds execution tracing and evaluations.
- Agentflow V2 runs on native nodes: per its documentation, agent logic there relies on Flowise's own nodes rather than on an external framework like LangChain. Its fourteen node types include loops, conditions, human approval and calls to other flows.
Abandoned, with critical flaws left unpatched
Nobody maintains Flowise anymore. According to the end-of-life announcement, the team froze the code on July 29, 2026, the day 3.1.4 shipped, archived the repository in August and left Discord and GitHub on August 31. Workday had acquired Flowise in August 2025. The stated reason: rigid low-code workflows hit a wall on complex tasks, and developers are moving to coding agents such as Claude Code or OpenClaw, the same agents Paperclip arranges into an org chart with budgets. HenryHengZJ, who signed the announcement, authored 1,766 commits; the next contributor has 232.
Known vulnerabilities have no published fix. The repository lists 130 security advisories, 105 of them published in 2026, and releases 3.1.3 and 3.1.4 patched 47. On September 10, 2026, six more advisories, two of them critical and tied to SSO login, were published against every version up to 3.1.4, with no patched version listed. Fixes, several of them for access checks, landed on the main branch until August 10 but never made it into a release. SECURITY.md no longer accepts reports.
SSO and roles are not Apache 2.0. The packages/server/src/enterprise directory that holds them is under a commercial license requiring an Enterprise subscription for production use. The docs reserve SSO (Entra ID, Google, Auth0) for Enterprise and workspaces for Cloud and Enterprise. Without a subscription, a fork running in production can only rely on the Apache 2.0 part.
Node.js 20 and SQLite are enough to start. Install with npm install -g flowise. The docs recommend Postgres for production, and MySQL and MariaDB are supported too. Once prediction volume outgrows the main Node process, queue mode hands execution to workers and requires Redis. No memory figure is documented, but building the monorepo may require raising the Node heap to 4 GB.
Flowise Cloud's fate is not announced. On October 7, 2026, flowiseai.com still listed its $35 and $65 monthly plans under a "We're sunsetting Flowise" banner.