opensourceprojects.dev

A broadsheet for software that doesn't ask for your email

Build private AI agents visually, keep your keys local
GitHub RepoImpressions3

Project Description

View on GitHub

Build Private AI Agents Visually, Without Handing Over Your API Keys

You've probably got a handful of API keys scattered across environment files and browser extensions, and you've probably wondered where they actually end up. Building an AI agent usually means choosing between a hosted platform that wants your keys and a pile of glue code that you maintain yourself. FLUJO is an open-source, local-first project that tries to split that difference: build agents visually, run them yourself, and keep your secrets on your own machine.

What It Does

FLUJO is a visual builder for AI agents. You connect your AIs and apps, assemble an agent, and then talk to it, automate it, or call it from other software. The project describes two ways to build: a simple step-by-step recipe mode for straightforward agents, and an expert visual flow mode for something more involved. Once an agent exists, you can inspect every run through a built-in debugger, which matters more than it sounds when you're trying to figure out why a chain of steps produced something strange.

The architecture is local-first. You can keep FLUJO on localhost, allow devices on your private network, or configure it for a public deployment. Agents get exposed through OpenAI-compatible and MCP endpoints, so the same agent you built visually can be called from other tools. It's built on the PocketFlow Framework, and the README notes it was developed with Cline and Claude Code. It's MIT licensed and currently at version 3.46.1.

Why It's Cool

  • Secrets stay server-side. API keys and other secrets are encrypted in local storage, with an optional custom encryption password for extra protection. The frontend only ever sees a masked placeholder. If you've ever been uneasy about pasting a key into a web app, this is the part that'll land.

  • MCP-native and OpenAI-compatible endpoints. This is the quiet workhorse feature. Building an agent visually is nice, but being able to call it from other software through endpoints you already know how to talk to is what makes it useful beyond the demo. You're not locked into a proprietary interface.

  • Two builders for two kinds of people. The step-by-step recipe mode is for when you just want a thing that works. The expert flow mode is for when you need to see and shape the graph. Not every project respects both audiences.

  • A real debugger. Agent runs are opaque by default. Being able to inspect every run turns "it didn't work" into "step four returned nothing," which is the difference between a tool you keep and a tool you abandon.

  • A genuinely helpful installer. The recommended install path sets up Git, Node.js, Python, uv, and ripgrep, clones the repo, builds it, and creates a global flujo command. The README is explicit that this matters because MCP servers need those runtimes. That's a thoughtful detail most projects skip.

  • The maintainer seems reachable. The README asks for Discord messages and GitHub issues, and says they read every message and try to respond within a day. Take that with the usual grain of salt, but it's a good sign for an open-source project.

One honest caveat: the README was truncated in what I can see, so there's likely more feature surface than what's covered here. What's visible is enough to know the shape of the thing.

How to Try It

The recommended path is the installer, which handles the dependency chain for you.

Windows — download flujo-setup.exe from the latest release. Or, in PowerShell:

irm https://raw.githubusercontent.com/mario-andreschak/FLUJO/main/scripts/install.ps1 | iex

Linux / macOS — paste this into a terminal:

curl -fsSL https://raw.githubusercontent.com/mario-andreschak/FLUJO/main/scripts/install.sh | bash

Already have Node.js? You can skip installation and run a prebuilt version from npm:

npx flujo-ai

The README warns that MCP servers may still need git, python, or uv on your PATH even with the npx route, so the installer is the safer bet if you want everything to just work.

There's also a hosted version you can try at try.flujo.com.co if you'd rather poke at it before installing, and a 2:28 product film at flujo.com.co/short.

The repository is at github.com/mario-andreschak/flujo.

Final Thoughts

FLUJO is best for developers who want visual agent building without giving up control of their keys or their data, and who are comfortable with a local install rather than a browser tab. The combination of local-first secrets, MCP and OpenAI-compatible endpoints, and an actual debugger covers the parts that usually make these tools frustrating in practice. If you've been circling visual agent builders but keep bouncing off the hosted ones, this is worth an afternoon.

Back to Projects
Project ID: 74462f91-6541-4006-a6ac-039ed745652cLast updated: September 30, 2026 at 05:33 AM