What is OpenHuman in 2026, and how does it differ from OpenClaw or ChatGPT?
OpenHuman is the GPL-3.0 desktop agent from TinyHumans AI that continuously ingests Gmail, calendars, GitHub, local files, and chat history into a markdown Memory Tree on disk. Unlike a one-off ChatGPT browser tab that forgets you when the session ends, OpenHuman behaves more like a personal knowledge layer living on your machine—routing tasks to cloud or local models while keeping context under your control.
Compared with OpenClaw, OpenClaw centers on message-channel gateways and workspace files around the openclaw CLI for Telegram, WhatsApp, and Discord. OpenHuman emphasizes a desktop GUI, Obsidian-compatible Memory Tree, voice, and Google Meet participation. ChatGPT alone gives fast answers but no durable local ingest, no signed package install path, and no Ollama routing without extra wiring. This guide covers the official desktop install channels—not a Python pip clone or source build from the repo README.
Expect twenty to forty minutes for a first clean pass on hardware you already own: pick an install channel, complete OAuth onboarding, connect one data source, and confirm a Memory Tree write. Add another hour if you are indexing a large Gmail archive, tuning local-vs-hosted model boundaries, or pairing Ollama on the same box.
OpenHuman stores durable state under your user profile—SQLite, markdown trees, and connector tokens. Treat that directory like application state, not disposable cache, especially once Gmail sync has run overnight.
curl pipe security: The official install.sh is served from raw GitHub without a separate signature today; prefer Homebrew, apt, npm, or signed MSI when integrity matters.
Laptop sleep: Closing the lid pauses background ingest; Memory Tree growth flatlines while OAuth tokens may need refresh on wake.
RAM for mailbox plus Ollama: Official docs cite 4 GB minimum but large Gmail plus a local model wants 16 GB or more unified memory.
OAuth and hosted boundary: Connectors need browser OAuth; some model routes still call hosted APIs—read the onboarding toggles before assuming fully offline operation.
Early beta: The README labels the project early beta; connector behavior and install channels change between releases—pin a version when you go production.
Pick a deployment plane first, then walk the six-step sequence in section 04.
Where should OpenHuman run? MacBook, VPS, or monthly Mac Mini?
OpenHuman ships native binaries for macOS, Windows, and Linux amd64. Homebrew and apt ride OS signing chains; npm downloads a verified SHA-256 binary on first install. GUI onboarding and OAuth flows are smoothest on a machine with a real desktop session. API-only cloud inference without ingest fits 4 GB RAM; Gmail indexing plus Ollama local-ai usually wants 16 GB or more.
| Deployment plane | 24/7 ingest | Install path completeness | Memory Tree growth | Monthly cost feel |
|---|---|---|---|---|
| Personal MacBook | Stops when lid closes (~60%) | High (Homebrew, GUI OAuth) | Frequent interruptions | $0 hardware + ops anxiety |
| x86 VPS 4GB | ~99.5% uptime, no GUI | Linux apt, headless friction | OAuth and desktop gaps | Usage-based billing |
| Monthly Mac Mini M4 | Datacenter SLA, ~100% | Highest (launchd + UMA) | Continuous indexing | Fixed OpEx, return hardware anytime |
If you evaluate cloud placement, weigh Gmail sync latency and OAuth redirect URLs against where Memory Tree files will live. A VPS in another continent can be cheap yet add friction to desktop OAuth; a macOS node with Screen Sharing often completes onboarding faster even at similar monthly cost.
Install tutorials answer whether OpenHuman launches; platform choice answers how long Memory Tree ingest runs and whether Gmail stays current—production personal agents usually deserve a dedicated macOS node.
Install channels compared: Homebrew, apt, npm, curl, and Windows MSI
Official docs rank native package managers first because they verify artifacts through Homebrew bottle hashes, a signed apt repo, npm SHA-256 postinstall checks, or a signed Windows MSI from GitHub Releases. The curl one-liner remains convenient but is explicitly unverified until release-signed script assets ship.
| Channel | Integrity | Best for | Upgrade path |
|---|---|---|---|
| Homebrew | Bottle hash | macOS and Linux dev machines | brew upgrade openhuman |
| apt | Signed repo | Debian and Ubuntu servers | sudo apt upgrade openhuman |
| npm -g | SHA-256 at install | Node shops, cross-platform CI | npm update -g openhuman |
| curl install.sh | None today | Quick try on trusted networks | Re-run script or switch channel |
| Windows MSI | Release signature | Corporate Windows desktops | New MSI from Releases |
Recommended macOS and Linux path:
brew install tinyhumansai/openhuman/openhuman openhuman --version
Debian and Ubuntu signed apt repo:
sudo apt-get install -y gnupg2 curl ca-certificates curl -fsSL https://tinyhumansai.github.io/openhuman/apt/KEY.gpg \ | sudo gpg --dearmor -o /etc/apt/keyrings/openhuman.gpg echo "deb [signed-by=/etc/apt/keyrings/openhuman.gpg arch=amd64] \ https://tinyhumansai.github.io/openhuman/apt stable main" \ | sudo tee /etc/apt/sources.list.d/openhuman.list sudo apt-get update sudo apt-get install -y openhuman
Fast but unverified curl path—prefer native packages in production:
curl -fsSL https://raw.githubusercontent.com/tinyhumansai/openhuman/main/scripts/install.sh | bash source ~/.zshrc openhuman --version
On Windows, download the MSI from GitHub Releases or tinyhumans.ai/openhuman. For npm globally: npm install -g openhuman with Node 18 or newer—the wrapper fetches the platform binary on first install. Corporate networks sometimes block raw GitHub URLs; mirror the apt repo or MSI internally, or install from a machine with direct access then rsync the user data directory to the target host.
Tip: Do not follow outdated blog posts that suggest pip install or cloning the repo for a daily driver—that path is for contributors building from source, not the supported 2026 desktop install.
Six steps from install to your first Memory Tree request
Pick a channel and install: Use Homebrew or apt in production; curl or npm for a quick lab. Confirm openhuman --version before opening the GUI.
Sign in and complete onboarding: Launch the desktop app, create or sign in to your TinyHumans account, and accept the early-beta notices—note which features stay local vs hosted.
Connect Gmail: Start with one mailbox in Settings, finish Google OAuth in the system browser, and wait for the first sync batch before expecting Memory Tree search to work.
Draw the local vs hosted boundary: Keep Memory Tree files and SQLite on disk locally; choose cloud models for heavy reasoning or flip to local-only routes when privacy policy requires it.
Run the first AI request: Ask a question that requires ingested mail or calendar context; confirm the answer cites Memory Tree nodes, not generic web knowledge.
Optional Ollama local-ai: Install Ollama, pull a 7B–13B model sized to your RAM, point OpenHuman model routing at http://127.0.0.1:11434, and retest the same prompt offline.
When moving to a cloud Mac for 24/7 ingest, follow the dual-stack rental checklist: back up Memory Tree directories and connector state, reinstall via the same Homebrew tap over SSH, restore files, and add a LaunchAgent so sync survives reboot. Reboot once on purpose after setup: if Gmail ingest resumes without manual relaunch, you have a real always-on path—not merely a GUI you forgot to keep open.
Restrict OAuth scopes to mailboxes and calendars you actually need. A wide connector surface on a sleeping laptop leaks convenience without delivering continuous Memory Tree value; tighten sources before leaving ingest enabled overnight on shared hardware.
Sizing, licensing, and quick error reference
Three numbers worth citing: ① official docs list 4 GB RAM minimum and 16 GB or more when indexing large mailboxes alongside a resident Ollama model. ② OpenHuman is released under GPL-3.0—fork-friendly, with copyleft obligations if you redistribute modified builds. ③ Plan about twenty minutes to read this guide and execute a first-channel install on familiar hardware; full Gmail backfill can run hours longer in the background.
| Error | Likely cause | Fix |
|---|---|---|
command not found | PATH after brew/npm/curl | source ~/.zshrc or reinstall channel |
| OAuth loop or blank browser | Headless or blocked redirect | Use desktop session or cloud Mac GUI |
| Memory Tree empty after hours | Laptop sleep or sync paused | Keep host awake; check connector status |
| Ollama timeout | RAM too low for model | Smaller model or 16 GB+ host |
| npm binary missing | Failed postinstall download | npm uninstall -g openhuman && npm install -g openhuman |
Honest alternatives: a sleeping MacBook installs cleanly but cannot guard overnight Gmail ingest; a cheap Linux VPS runs apt but fights GUI OAuth and Memory Tree desktop integration; ChatGPT alone answers fast without durable local context. Renting a dedicated Mac Mini M4 from KVMNODE gives launchd supervision, unified memory for mailbox indexing plus optional on-box Ollama, and the same Homebrew workflow over SSH—monthly OpEx lets you validate thirty days before buying hardware. That trade usually beats a laptop you open occasionally when OpenHuman is production personal infrastructure, not a weekend experiment.
Operational rhythm that works in the field: weekly check on connector status after upstream releases, snapshot Memory Tree directories before major upgrades, and rotate Google OAuth tokens if the host was imaged on shared hardware. Disk growth lives beside markdown trees and SQLite—plan snapshot space, not just the initial installer footprint.
See pricing for tiers and help center for delivery, networking, and backup playbooks. When you are ready to commit uptime without capital expense, the order flow provisions an isolated M4 you can SSH the same afternoon.