OpenHuman is an open-source desktop AI assistant from tinyhumansai. It installs locally on your laptop, connects to 118+ services like Gmail, GitHub, Slack, Notion, and Calendar via OAuth (powered by Composio), and builds a personal memory tree from your data. It went from 2K to nearly 20K GitHub stars in days. Single-user, local-first, with native voice and a desktop mascot that can join Google Meets.

Can I host OpenHuman on OpenClaw Launch?

No. OpenHuman is a desktop app, not a containerized server agent. Its value depends on running on your local machine — mascot UI, native voice, Obsidian vault, Google Meet participation, single-user privacy story. OpenClaw Launch hosts multi-tenant audience-facing bots; OpenHuman is your personal single-user assistant.

OpenHuman vs OpenClaw Launch — which should I pick?

They solve different problems. OpenHuman is for you (your laptop, your personal data, only you talk to it). OpenClaw Launch is for your audience (managed bots on Telegram, Discord, WhatsApp, WeChat, web chat that other people message). Both products use Composio for OAuth into Gmail, GitHub, Slack, and 90+ other apps. Most serious setups run both.

OpenHuman vs Hermes Agent — what is the difference?

Hermes Agent is a self-hosted general agent runtime with multi-level memory and 40+ tools — you wire it up however you want. OpenHuman is a fixed desktop product focused specifically on building personal memory from your SaaS stack. Different layers: Hermes is the runtime, OpenHuman is the personal-memory product.

Will OpenClaw Launch add OpenHuman-style memory?

OpenClaw Launch already shares OpenHuman integration backbone — both use Composio for OAuth into Gmail, GitHub, Slack, Notion, and 90+ other apps. The difference today is that OpenClaw Launch exposes connections as live tools the agent calls on demand, while OpenHuman pre-indexes them into a memory tree. A hosted memory-sidecar that adds passive pre-indexing on top of the existing /integrations product is on the OpenClaw Launch roadmap. Email support@openclawlaunch.com to be notified when it ships.

Does OpenHuman support Chinese SaaS tools like Feishu or WeChat Work?

Not yet. OpenHuman supports English-stack tools (Gmail, GitHub, Slack, Notion, Google Calendar, Drive, Linear, Jira, Stripe). Feishu, DingTalk, and WeChat Work are not in the supported list as of May 2026.

Is OpenHuman really running on my machine?

Yes. The memory tree, SQLite database, Obsidian vault, and TokenJuice preprocessor all run locally on your laptop. Raw data from your connected SaaS tools never leaves your machine until it is sent to an LLM you have configured (and that step can be skipped entirely by running a local model via Ollama).

← All Comparisons

Comparison

OpenHuman vs OpenClaw Launch vs Hermes Agent

OpenHuman went from 2K to nearly 20K GitHub stars in days — the most-talked-about AI agent launch of 2026 so far. But it's not a competitor to OpenClaw Launch or Hermes Agent. It's a different layer of the stack. Here's what each one actually does, when to pick which, and why most serious setups end up using more than one.

Quick Comparison

	OpenHuman	OpenClaw Launch	Hermes Agent
Type	Desktop app	Managed platform	Self-hosted framework
Where it runs	Your laptop	Our managed cloud	Your VPS
Primary job	Learn you from Gmail, GitHub, Slack…	Run bots on Telegram, Discord, WhatsApp, web	Self-improving agent runtime
Who messages it	Just you (single-user)	You + your audience (multi-user)	You + your audience (multi-user)
Data storage	Local SQLite + Obsidian vault	Encrypted on our cloud	On your VPS
OAuth target	Your personal Gmail, GitHub, Slack, Notion, Calendar…	Bot platforms + 90+ apps via Composio	Bot platforms + your tools
Setup time	~5 min (install + OAuth)	30 seconds	30+ min (server, Docker, SSL)
Telegram / Discord / WhatsApp	Not really	Built-in	Built-in
Desktop mascot & voice	Yes (native STT, ElevenLabs TTS, lip-sync)	No	No
Joins Google Meet as participant	Yes	No	No
Persistent memory of you	Memory tree from your real data	Per-instance sessions; integrations as live tools	Multi-level long-term memory
Token compression	TokenJuice (~80% claim)	Standard	Standard
Pricing	Free (you bring model API)	$3/mo* (credits included)	Free + VPS cost
Open source	Yes	Yes (OpenClaw)	Yes (MIT)
Local AI (Ollama)	Yes	Self-host only	Yes

*First month $3, then $6/mo on Lite.

The Real Distinction: Form Factor

The cleanest way to understand the three is by where they run and who talks to them.

OpenHuman is a desktop app. You install a DMG or EXE. It runs on your laptop with a mascot on screen, native voice, and direct access to your local Obsidian vault. Only you talk to it. The whole pitch is local-first: your personal Gmail, GitHub, and Slack data never leaves your machine.
OpenClaw Launch is a managed bot platform. You configure an agent in a web form, hit deploy, and 30 seconds later there's a bot answering on Telegram, Discord, WhatsApp, WeChat, or a web widget on your site. Other people talk to it — your community, your customers, your audience.
Hermes Agent is a self-hosted runtime. You spin up a VPS, install Hermes via Docker, and run an agent with persistent multi-level memory and 40+ built-in tools. You own the infrastructure end-to-end.

OpenHuman serves you. OpenClaw Launch and Hermes serve your users. That's the line.

OpenHuman

OpenHuman (tinyhumansai) is positioned as a “personal AI super intelligence, private, simple, and extremely powerful.” The pitch that drove the viral GitHub run: instead of you teaching the AI through prompts and skills, the AI learns you passively from the data already in your daily SaaS stack.

How it works:

118+ OAuth integrations — one-click connect to Gmail, GitHub, Slack, Notion, Calendar, Drive, Linear, Jira, Stripe, and more. The integration plumbing is powered by Composio, the same OAuth + tool-invocation platform OpenClaw Launch already uses on its /integrations page.
20-minute polling — every active connection is walked on a fixed cadence; you don't write cron jobs.
Memory tree — data is canonicalized into ≤3K-token Markdown chunks, scored, and folded into hierarchical summary trees. Stored locally in SQLite plus an Obsidian-compatible vault.
TokenJuice — preprocessing layer that converts HTML to Markdown, shortens URLs, and dedupes content before anything touches an LLM. Claims ~80% token savings.
Native voice and Meet participation — STT input, ElevenLabs TTS output, mascot lip-sync, and an agent that can join Google Meets as a real participant.
Optional local AI via Ollama for on-device workloads with no cloud round trip.

What to consider:

Single-user only — OpenHuman is your personal assistant on your machine. There's no managed multi-tenant version, no Telegram bot people can DM, no audience-facing surface.
Desktop-bound — if your laptop is off, the agent is off. There's no background server doing work for you while you sleep.
Local-only memory — the memory tree lives on your machine. No official multi-device sync yet.
Early beta — under active development as of May 2026. Expect rough edges.
Chinese SaaS gap — the 118 integrations are mostly English-stack tools. Feishu, DingTalk, WeChat Work aren't supported.

OpenClaw Launch

OpenClaw Launch deploys multi-channel AI agents in 30 seconds. Pick a model, pick the platforms (Telegram, Discord, WhatsApp, WeChat, web chat), hit deploy. The bot is live, with SSL, monitoring, and updates handled.

Where it fits:

Audience-facing bots — your community on Telegram, your customers on WhatsApp, your support widget on a website. OpenHuman doesn't do this.
30-second deploys — warm pool technology, no server provisioning, no Docker.
Flat pricing — $3/mo first month then $6/mo on Lite, or $20/mo on Pro, with AI credits included on every tier.
3,200+ skills — install from the ClawHub registry in one click.
20+ models — Claude, GPT, Gemini, Llama, Hermes, all via OpenRouter or BYOK.
Same Composio backbone as OpenHuman — the /integrations page connects your bot to Gmail, GitHub, Slack, Notion, and 90+ other apps as live tools.

Hermes Agent

Hermes Agent by Nous Research is a self-hosted, MIT-licensed agent framework with persistent multi-level memory, 40+ built-in tools, and the Hermes Function Calling standard. It's optimized for users who want to own the runtime.

Hermes is the closest of the three to a “general agent runtime.” OpenHuman is a fixed product with a specific UI; Hermes is a framework you wire up however you want. The trade-off is setup — VPS, Docker, SSL, monitoring all on you.

See the full OpenClaw Launch vs Hermes Agent comparison for pricing, deploy paths, and the managed Hermes hosting waitlist.

Can OpenClaw Launch Host OpenHuman?

Short answer: no, not the way OpenClaw Launch hosts OpenClaw or Hermes agents. OpenHuman is a desktop app. Its value — the on-screen mascot, native voice, Google Meet participation, local Obsidian vault, single-user privacy story — depends on running on your machine. Move it to a server and the differentiators evaporate.

That said, OpenClaw Launch already shares OpenHuman's integration backbone. Both products use Composio for one-click OAuth into Gmail, GitHub, Slack, Notion, and 90+ other apps. The difference today is that OpenClaw Launch exposes those connections to your bot as live tools (the agent calls Gmail when needed), while OpenHuman pre-indexes them into a persistent memory tree (the agent already knows). A hosted “memory of you” sidecar that any OpenClaw or Hermes agent can call as a tool is on the OpenClaw Launch roadmap. If you want to be notified when that ships, send a note to [email protected].

How They Work Together

These aren't three competing products. They're three layers of a personal AI stack:

OpenHuman — the personal-memory layer. Knows your inbox, your calendar, your repos, your channels. Talks only to you.
OpenClaw Launch — the channels and ops layer. Puts an always-on bot in front of your community, customers, or audience without any infra work.
Hermes Agent — the self-hosted reasoning layer. Multi-level memory, planning, subagent delegation, full control.

A realistic 2026 setup: OpenHuman on your laptop quietly indexes your work; OpenClaw Launch runs your Telegram support bot and Discord community bot; Hermes (self-hosted, or via the managed Hermes hosting waitlist) handles long-running internal agents that need persistent memory. Different problems, different layers, different machines.

What OpenHuman Is Not

The viral framing of OpenHuman has caused some confusion. To be precise:

Not a hosted SaaS. You install it locally. There's no “sign up at openhuman.com.”
Not a bot platform. It doesn't put an agent on Telegram or Discord for your users to message. It's your personal assistant.
Not a Hermes replacement. Hermes is a general agent runtime. OpenHuman is a specific product with a fixed UI and a personal-memory product thesis.
Not an OpenClaw alternative. They solve different problems. You can run both at the same time without overlap.

Which Should You Choose?

Choose OpenHuman if you want a desktop AI that learns your personal workflow from your Gmail, GitHub, Slack, and Calendar — without you writing prompts or skills. Best for solo developers and knowledge workers who live in those tools and want a private, local-first assistant.

Choose OpenClaw Launch if you need an always-on AI bot that other people message — on Telegram, Discord, WhatsApp, WeChat, or a website widget. 30-second deploy, $3/mo first month, AI credits included. The fastest path from idea to live bot.

Choose Hermes Agent if you want full control over a self-hosted agent runtime with persistent multi-level memory and 40+ built-in tools. Best for developers comfortable with Docker who need long-running internal agents.

Use more than one. The three layers don't overlap. OpenHuman on your laptop + OpenClaw Launch for audience-facing bots + Hermes for self-hosted internals is a coherent stack, not a duplication.

Bottom Line

OpenHuman is the first agent product to seriously compete on usage friction rather than capability ceiling — it learns you instead of asking you to teach it. That's a real category. But the form factor (desktop app, single-user, local-first) means it lives next to, not on top of, products like OpenClaw Launch and Hermes Agent.

If you want a personal assistant on your laptop, install OpenHuman. If you want a bot your audience messages, deploy on OpenClaw Launch. If you want a self-hosted agent runtime, run Hermes — or join the managed Hermes hosting waitlist to skip the VPS step.

Also see: OpenClaw Launch vs Hermes Agent for the managed-vs-self-hosted comparison, OpenClaw Launch vs Composio for the integration-platform comparison, or all comparisons for the full list of OpenClaw alternatives.