# OpenHuman

> A desktop personal AI that keeps its memory on your own machine

- Category: Agents & Automation / Assistants & Chatbots
- Language: Rust
- Licence: GPL-3.0
- Repository: https://github.com/tinyhumansai/openhuman
- Website: https://tinyhumans.ai/openhuman
- Canonical page: https://ai-tldr.dev/tools/openhuman/

## In short

OpenHuman is a desktop personal AI that keeps its memory on your own machine. It sits in the Assistants & Chatbots category of the Agents & Automation tools and is open-source, written in Rust and licensed GPL-3.0. To get started: download an installer.

## Overview

OpenHuman is a desktop personal AI for macOS, Windows and Linux, built by Tiny Humans and released as open source under the GNU GPL. The project frames it as three things at once: "a brain that remembers everything, a fantastic orchestrator, a deep researcher" — an assistant whose defining choice is that the memory and the orchestration run on your machine rather than in someone else's account.

The memory layer is the centrepiece. Data is stored locally as a Memory Tree with an Obsidian wiki view, compressed rather than kept raw, and refreshed by an auto-fetch pass every twenty minutes across connected sources. More than a hundred OAuth integrations, a large catalogue of MCP servers and an even larger skills library feed into it, and goals and todos sit alongside the notes so the assistant has something to act on rather than only recall.

On top of that sits the orchestrator: checkpointed graph-based workflows built on the project's tinyflows and tinyagents components, encrypted agent-to-agent messaging, and automation that proposes a plan and waits for approval before executing it. Research and tool use cover web search and scraping, coding tools, browser automation, native Whisper voice, and image and video generation. Seventeen messaging channels including native email give it places to reach you. A Privacy Mode keeps everything local. The project is explicit that it is in early beta and under active development, so treat it as software you are trying rather than depending on.

## What it does

- Local-first Memory Tree with an Obsidian wiki view and compressed storage, auto-refreshed every 20 minutes
- 100+ OAuth integrations, thousands of MCP servers and a large installable skills catalogue
- Checkpointed graph workflows on tinyflows and tinyagents, with approval-gated automation
- Encrypted agent-to-agent messaging between the assistant's own agents
- Deep research: web search, scraping, coding tools and browser automation
- Native Whisper voice, plus image and video generation
- Privacy Mode for local-only operation; no configuration files to hand-edit

## Getting started

OpenHuman is distributed as a desktop application. Installers are published on the project site and on GitHub Releases; terminal installation routes are documented in the repository's INSTALL.md.

1. Download an installer

   Get the build for your platform from the project site or the GitHub releases page.

```text
https://tinyhumans.ai/openhuman
https://github.com/tinyhumansai/openhuman/releases
```

2. Or install from a terminal

   Homebrew, Debian/Ubuntu packages and an AUR package are covered in INSTALL.md in the repository.

```text
https://github.com/tinyhumansai/openhuman/blob/main/INSTALL.md
```

3. Connect your sources

   Authorise the OAuth integrations you want the brain to read. The auto-fetch pass then pulls from them every twenty minutes and folds the results into the local memory tree.

```text
Settings → Integrations → connect accounts
```

4. Turn on Privacy Mode if you need local-only operation

   Privacy Mode confines the assistant to local execution, which is the setting to use before pointing it at anything sensitive.

```text
Settings → Privacy Mode
```

5. Give it a goal

   Goals and todos live beside the memory tree; workflows are checkpointed and automation is approval-gated, so a proposed plan is shown to you before it runs.

```text
Goals → New goal → review the proposed workflow → approve
```

## When to use it

- Reach for it when you want a personal assistant whose memory stays on your own machine
- Reach for it when the assistant needs to read across many accounts and keep a durable, browsable note store
- Reach for it when automation should propose a plan and wait for your approval rather than acting immediately
- Reach for it when you want one assistant reachable from email and a range of messaging apps