Original Summary

Hi, (/edit: I wrote a wall of text and I'm sorry. But it's kind of an important topic to me) I have been working on a side project for the largest part of the last 18 months. Free to use, open source, and a ton of effort that went into it. I'm literally working to improve it every minute I have. It's an AI harness and workflow builder. Kinda in the same nieche as Dify, Flowise, n8n, activepieces - but also a bit claude-code, codex, opencode and similar. My problem: No traction, barely any users. Nobody seems to use this as their daily driver, but me. Nobody talking about it, but me. We're at 660 stars on github, so it's not "nothing" - but we all know a Github Star doesn't say anything about the actual adoption. I see it as "660 people liked the idea and bookmarked it for later". I need to understand what's wrong with my sideproject. Is the app actually useless? (Flowise was already sunset; looking at the growing capabilities of frontier models) Is the name sh*t? (" FLUJO ") Is it the in-app UI? (I'm trying to keep it clean, easy, non-technical. But I am not an UX specialist) Is it the Website/Github Repo? (Readme bloated? Should I compare against the alternatives?) Is the app itself too complicated to use? (I know it inside-out, but I can't tell how complicated it is to wrap your head around.) Is it missing features? (probably not; I stopped adding new features already and I'm focusing on polish and getting it really really enterprise-ready) Is it because we're not appearing in any of the "lists" online? (a death-spiral: can't add it to the lists without it being famous enogh - cant get famous enough without being on the lists) There are many smart people in this subreddit and a lot of smart LLM bots, so that's why I am posting here. To get a second (or third) opinion. The project is called FLUJO; spanish for Flow. As said earlier, it's an multi-agent-/workflow-harness revolving around the MCP protocol. It's a NextJs+React app that runs on your computer for (skip if you dont care): Connecting models (OpenAI, Anthropic, [...], Ollama, Claude Subscription, Codex Subscription) Installing/running MCP servers from the official MCP registry, GitHub URL, remote-url, or a local folder Building agents as either a simple "recipe" or a visual graph (branch, loop, subagents) Chatting with live execution, breakpoints, and tool-approval gates Triggering agents through automation (cron, webhook, file watch, URL watch, MCP poll) Re-exposing agents as an OpenAI-compatible API - and installed MCP-servers through an mcp-proxy for reuse Cursor/Claude Desktop/Cline and a lot of things more. Agents with long-term goals, durable memory, meetings for agents, recover on failures, etc. Currently, I am not collecting any telemetry. Neither on the Website, nor through the app's UI. I ask people to leave their feedback or create Github issues, but I'm no

  • 情报分类：技术学习与提效
  • 分类依据：内容涉及技术、AI、软件工具或工程实践
  • 信息来源：Reddit · SideProject
  • 发布时间：2026/10/15 08:25:07