davemessew/AgentFlux

Agentic AI platform for visual workflows and autonomous agents. Run it with Docker on your own computer or server.

What it solves

AgentFlux removes the friction of connecting documents, data sources, and language models into a single automated pipeline. Instead of writing code or juggling multiple tools, you visually arrange steps and let an AI agent turn a plain-task description into a working workflow. All data stays on your own machine or server, so you keep control over private files and model calls.

How it works

You use a browser interface to build workflows on a canvas: connect input sources (files, folders, HTTP endpoints, approved programs) to processing steps, then to language models or report outputs. There are 17 preconfigured steps, eight templates, branching, and undo/redo. An AI agent can also create a workflow from a short description, ask for missing details, and reuse saved plans for recurring tasks. A special video feature records your screen and speech, then automatically generates a workflow that replicates the actions you demonstrated. The application runs in Docker, stores data in named volumes, and delivers results as JSON, CSV, or Markdown files to a local outbox.

Who it's for

AgentFlux is designed for a single user who wants a self-hosted AI assistant for automating data handling, document processing, and report generation. It suits analysts, researchers, or developers who prefer a visual environment, want to keep data local, and are comfortable running a Docker container. There is no login or multi-user support, so it's intended for personal or trusted-network use.

Highlights

  • Visual workflow editor with 17 preconfigured steps, eight templates, branching, and undo/redo
  • AI agents that generate workflows from task descriptions and handle recurring jobs
  • Video-to-workflow: screen recording plus speech analysis creates executable workflows automatically
  • Local knowledge base for documents and observations, with reusable skills and instructions
  • Connect your own models via Ollama, LM Studio, vLLM, or any OpenAI-compatible endpoint
  • Traceable runs: processes only changed inputs and outputs structured reports to a local folder
  • No login, single-user focus, with optional approval for cloud calls and external HTTP steps
  • Docker release includes FFmpeg and a local speech service with Whisper base model

関連

  • プロジェクト
  • プロジェクト
  • プロジェクト
  • プロジェクト