Flowise Is Dead at 55k Stars, and Its Founders Blame Coding Agents

TL;DR

Flowise, one of the earliest and best-known open-source visual builders for LLM apps and agents, is shutting down. In a letter titled "The Future of Flowise" that hit Hacker News overnight, the team says active development already stopped on July 29, the GitHub repository (55.2k stars, 24.8k forks) gets archived on August 10, and official support ends August 31. The stated reason is not funding and not burnout: developers moved to coding agents like Claude Code and OpenClaw, and the drag-and-drop canvas "quickly hits the limit when it comes to complexity." The tool for building AI agents got outbuilt by AI agents. One more wrinkle: Workday acquired Flowise less than a year ago, and the goodbye letter does not mention Workday once.

What was announced

The letter, signed "Henry & Flowise Team," is short and unusually honest. Flowise launched in 2023 as an Apache-2.0 visual builder: you dragged nodes onto a canvas, wired an LLM to a vector store and some tools, and shipped a chatbot or RAG pipeline without writing the glue code yourself. It rode the 2023-2024 LLM-app wave to 55,200 stars and 24,800 forks, numbers most infrastructure projects never see.

The wind-down is dated and specific:

• July 29, 2026: active development stopped; pull requests no longer accepted.
• August 10, 2026: the GitHub repo is archived; npm packages and Docker images are deprecated.
• August 31, 2026: official team support ends.

The code is not disappearing. "Flowise source code will still remain on Github and the Apache 2.0 licensed code is yours to keep building on," the letter says, and it explicitly encourages users to fork. What ends is maintenance: no more releases, no more security patches, no one on the other end of the issue tracker.

The stated cause of death: coding agents

Post-mortems for dev tools usually blame the market, the macro, or "focus." Flowise's letter names its replacement. As models got better at reasoning, it says, developers began "increasingly relying on new coding agents such as Claude Code/OpenClaw to handle complex tasks," while the rigid low-code workflow approach "quickly hits the limit when it comes to complexity."

That tracks with what the category was for. Low-code existed because code was expensive to produce: you accepted the canvas's constraints the way you accept flat-pack furniture, tolerating the Allen key and the wobble because a carpenter costs too much. A coding agent is a carpenter who works for twenty dollars a month, and the market for wobble collapsed. When generating the real, expressive version of your pipeline costs one prompt, a visual abstraction over that pipeline stops being a shortcut and starts being a ceiling.

The irony is not lost on anyone: Flowise spent three years helping people ship agents, and the agents shipped back. It is one of the cleanest data points yet for a thesis this blog has tracked all year, from Google reorganizing its SDLC around agents to harness-first development patterns: the abstraction layer that wins is the one the model can write, read, and refactor. A JSON canvas the model cannot easily diff is a dead end.

The Workday-shaped hole in the letter

On August 14, 2025, Workday announced it had acquired Flowise, then at 42,000+ stars and, per the press release, processing "millions of chats and workflows." The pitch was that Flowise's "industry-leading agent builder" would accelerate agent development across Workday's HR and finance platform.

Less than twelve months later, the public project is being archived, and the word Workday appears nowhere in the letter, an omission doing a lot of quiet work. Hacker News commenters reached for the "Our Incredible Journey" file within minutes, and the pattern fits: enterprise acquirer buys beloved open-source tool, open-source tool enters hospice. To be fair to the thesis, the letter's technical argument stands on its own; both things can be true. The builder likely lives on inside Workday's walls. The one you could self-host does not.

The category is not dying evenly

Before writing low-code's obituary, look at the neighbors. n8n, the workflow automation platform that bolted AI onto a general-purpose integration engine, sits at 199,270 GitHub stars. Langflow, Flowise's most direct visual-canvas rival, is at 152.8k and still shipping. The pure LLM-chain canvas is what died; workflow tools with a broader remit, and escape hatches into real code, are still growing.

The HN thread also surfaced the strongest counterpoint to "agents ate low-code": determinism. Compliance-heavy platforms still want a rigid, auditable graph that runs the same way every time, not a probabilistic agent improvising a pipeline per request. That niche is real. It just was not big enough, or Flowise-shaped enough, to carry a general-purpose canvas whose core audience learned to type "build me a RAG pipeline" into a terminal instead.

If you run Flowise, your clock is August 10

This is the actionable part. Archived GitHub repos stay cloneable, but deprecated npm packages and Docker images are how a working deployment quietly rots out from under you.

• Fork now. Archive status freezes issues and PRs on the upstream repo. If you have local patches, get your fork up before August 10, not after.
• Pin your artifacts. Mirror the exact Docker image digest and npm versions you run into your own registry. Do not depend on the deprecated tags resolving forever.
• Pick a lane. Maintaining a fork means inheriting dependency updates and security patches for an Express/Node surface indefinitely, a point HN commenters made with feeling. If you will not staff that, migrate the workflows: Langflow is the closest canvas-for-canvas swap, n8n the broader automation play, or take the letter's advice and let a coding agent rewrite the flow as plain code you own.
• Audit exposure. No more security patches after August 31. A self-hosted Flowise on a public URL becomes a slowly aging attack surface; move it behind a VPN or retire it.

Key Takeaways

• Flowise, the 55.2k-star open-source visual LLM app builder, is sunsetting: development stopped July 29, the repo is archived and npm/Docker artifacts deprecated on August 10, and team support ends August 31, 2026.
• The founders' stated cause is structural, not financial: developers moved to coding agents like Claude Code and OpenClaw, and the low-code canvas "quickly hits the limit when it comes to complexity."
• Workday acquired Flowise in August 2025, promising to accelerate agent building on its platform; the shutdown letter does not mention Workday at all.
• The visual-workflow category is not dying uniformly: n8n (199.3k stars) and Langflow (152.8k) keep growing, while the pure LLM-chain canvas lost its reason to exist.
• The code stays Apache 2.0 and forkable, but forking means owning security patches forever; self-hosters should fork and pin artifacts before August 10, then decide between maintaining, migrating, or letting an agent rewrite the flows as code.
• File this as one of the clearest casualties yet of the coding-agent era: when generating real code costs one prompt, abstractions built to avoid writing code become ceilings.

Sources: Flowise sunset announcement, Flowise GitHub repository, Workday newsroom, Workday press release (PR Newswire), SiliconANGLE, Hacker News discussion