AgentFlux

by davemessewAppGet organised

Runs document workflows and AI agents on your own computer, with results saved locally

The problem it solves

Every week you copy the same fields out of incoming files into a spreadsheet, and you don't want to upload those files to a cloud tool.

For People automating document choresTeams keeping data in-houseOllama and LM Studio users

Before: You open each file, copy the fields by hand and paste them into a sheet or report.
After: A workflow picks up new files, cleans the records and saves JSON, CSV or Markdown reports to a local folder on a schedule.

Stars: ★ 235
License: Apache-2.0
Cost: Free
Platforms: Linux · Win
Runs: Self-hosted

Screenshots

From the project's own repository · 4 images, swipe or use the arrows.

The workflow editor with connected steps that read an inbox, redact data and save the results — Image: davemessew (Apache-2.0). Original file

Overview

AgentFlux is a self-hosted app for building workflows that import documents, clean data and produce reports, either step by step on a visual canvas or by describing the task to an agent. You start it with Docker and open it in your browser; it connects to local model servers such as Ollama or LM Studio and can turn a screen recording with spoken explanation into a workflow. Results arrive as JSON, CSV or Markdown files in a local folder. It suits people who want to automate document chores while keeping files on their own machine.

Some workflows run with no AI model at all, and cloud model calls and external HTTP POST steps need your approval in settings and in the workflow.

Try these first

Start with the first one: it is a small task you can check easily.

In Agents, create 'Clean and export records without a model', choose Local inbox and run it once.First try · no model needed
Build a workflow that reads new files from my inbox, redacts sensitive information and saves the results to the knowledge base.
Every Monday, extract tasks with owners and dates from new documents and write a Markdown report.needs a model
In Agents, use Analyse video to turn this screen recording of my weekly report routine into a workflow.needs a vision model

Example prompts written by Agent Flows Hub.

Features

From the creator's documentation.

Feature	What it does	Runs
Visual workflows	17 preconfigured steps, eight templates, branching and JSON import and export	Your computer
Agents	Describe a task to get a workflow, and set recurring runs on a schedule	Your computer
Knowledge base	Stores imported PDF, DOCX, XLSX, CSV and text files locally and makes them searchable	Your computer
Reports	Delivers JSON, CSV and Markdown files to a local folder	Your computer
Your own models	Connects Ollama, LM Studio, vLLM and OpenAI-compatible endpoints	Model server
Video to workflow	Builds a workflow from a screen recording of up to 30 minutes with spoken explanation	Vision model

Good to know

Clean a folder of incoming documents and export records as JSON, CSV or Markdown on a schedule, without sending files anywhere. Connect a local model later to extract tasks or write reports.

Limits

The prebuilt image targets Linux/amd64; other architectures need a build from source and were not part of the release checks.
AgentFlux has no login and is designed for one user, so keep it bound to localhost unless you add your own access control.

Notes by Agent Flows Hub (original project name: AgentFlux), written from the creator's documentation, not from running it ourselves. AgentFlux README: Docker setup and first workflow · AgentFlux: install the release

Stuck setting it up?

Tell us what happened. We take a look first; if it needs paid work, you get an estimate before anything starts.

Get help

How to start

Get the appDownload it from the official page, or paste the install request into an AI agent you already use.Linuxmkdir agentflux cd agentflux curl -fLO https://github.com/davemessew/AgentFlux/releases/download/v0.3.0/agentflux-0.3.0-linux-amd64.tar.gz curl -fLO https://github.com/davemessew/AgentFlux/releases/download/v0.3.0/compose.yaml docker load -i agentflux-0.3.0-linux-amd64.tar.gz docker compose up -d --wait
WindowsDownload the installerWindowsdocker load -i agentflux-0.3.0-linux-amd64.tar.gz docker compose up -d --wait
Open itFollow the first-run steps in the app.
Give it a first taskCreate the agent 'Clean and export records without a model' with the local inbox, run it once and open the delivered files in the outbox.

Steps follow the creator's instructions. The request tells your agent to read them first and wait for your OK.

Before you start

A computer with Linux or Windows.
Docker with Compose (Docker Desktop in Linux container mode on Windows)
For AI steps: a model server such as Ollama or LM Studio, or a provider key
Free. Nothing to pay.

Details

Made by: davemessew
License: Apache-2.0 Terms
Platforms: Linux, Windows
Runs on: Self-hosted server

View source Official docs Report a problem

For AI agents

This page as a Markdown install brief:

/en/library/agentflux/install.md

Other apps

Amical Turns what you say into typed text in any app, with speech recognition on your computer
Anarlog Takes notes in your meetings without a bot joining the call, then writes the summary
AnythingLLM Lets you chat with your own documents and run AI agents in a private desktop app
ArcReel Turns a novel, script or product photos into storyboards and short videos on your server
ArtCraft Lets you stage a scene in 2D or 3D, then render it as an AI image or video
AstrBot Puts an AI assistant into QQ, WeCom, Feishu, Telegram, Slack and other chat apps

See all 113 apps

GitHub