OpenAgent

<p align="center"> <img src="assets/openagent-logo.png" alt="OpenAgent" width="360" /> </p> <p align="center"> Persistent AI agent framework with MCP tools, long-term memory, and multi-channel support. <br /> Model agnostic by design, with three independent apps: Agent Server, CLI Client, and Desktop App. </p> <p align="center"> <a href="https://openagent.uno/">Website</a> · <a href="https://openagent.uno/downloads">Downloads</a> · <a href="https://openagent.uno/guide/">Documentation</a> · <a href="https://github.com/geroale/OpenAgent/releases">Releases</a> </p> <p align="center"> <a href="https://pypi.org/project/openagent-framework/"><img alt="PyPI" src="https://img.shields.io/pypi/v/openagent-framework?style=flat-square&color=ef4136" /></a> <a href="https://github.com/geroale/OpenAgent/releases"><img alt="GitHub Release" src="https://img.shields.io/github/v/release/geroale/OpenAgent?style=flat-square&color=f26b3d" /></a> <a href="https://github.com/geroale/OpenAgent/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/github/license/geroale/OpenAgent?style=flat-square&color=fbb040" /></a> </p>

Overview

OpenAgent turns an LLM into a persistent agent that can remember, act, and stay reachable across different client surfaces. It is model agnostic by design: Claude CLI/API, Z.ai GLM, Ollama, LM Studio, vLLM, and OpenAI-compatible providers all use the same MCP tools, memory model, channels, and clients.

Why OpenAgent

• Model-agnostic execution with Claude CLI/API, Z.ai GLM, Ollama, LM Studio, vLLM, and OpenAI-compatible endpoints
• Bundled MCP tools for filesystem, editor, shell, web search, browser automation, messaging, scheduling, and vault operations
• Obsidian-compatible markdown memory with wikilinks, frontmatter, and graph-friendly notes
• Native service installation, cron scheduling, dream mode maintenance, and auto-update support
• Shared desktop app built with Electron and React Native Web for chat, configuration, MCPs, and memory exploration

Three Independent Apps

• Agent Server: the persistent runtime in openagent/, available as a standalone executable or pip package (openagent-framework)
• CLI Client: the terminal client, installed as openagent-cli
• Desktop App: the Electron UI, distributed as platform-specific binaries

Tagged GitHub releases are the shared download point for all three.

Quick Start

Option A: Standalone Executable (recommended)

Download the latest executable for your platform from GitHub Releases and run:

./openagent serve ./my-agent

This creates a self-contained agent directory at ./my-agent with default config, database, and memory vault. No Python required.

Option B: pip install

pip install openagent-framework[all]
openagent serve

Multi-Agent

Run multiple independent agents in parallel, each with its own data directory:

./openagent serve ./agent-work
./openagent serve ./agent-home

Each directory contains its own openagent.yaml, database, memories, and logs. Ports are auto-allocated to avoid conflicts.

Desktop App

Download packaged desktop builds from GitHub Releases or from the OpenAgent downloads page. To build locally:

cd app
./setup.sh
./start.sh macos

Documentation

The canonical documentation now lives on the website:

• Getting Started
• Desktop App
• Models
• MCP Tools
• Memory & Vault
• Configuration Reference

Repository Layout

OpenAgent/
├── openagent/          # Python framework runtime
├── app/                # Universal app + Electron wrapper
├── cli/                # Terminal client package
├── docs/               # VitePress website and documentation source
├── scripts/            # Build and release scripts
└── assets/             # Shared branding assets

License

MIT