About LobeHub Desktop 2.2.17 — AI Agent Workspace App
LobeHub is an AI workspace organised around agents. You create agents, give them models and skills, put several into a group to work on one task, and schedule runs for when you are away. LobeHub Desktop is its Electron app for Windows, macOS and Linux.
This page links to the official 2.2.17 Windows installer and the matching macOS and Linux builds, published by LobeHub LLC on GitHub. Bineret did not build LobeHub and is not affiliated with it.
What it does
- Agent Builder. Describe what you need once, and it sets up an agent you can use straight away.
- Many models. Over a dozen providers, including GPT-5, Claude, Gemini, DeepSeek, Llama and Qwen, and you can switch mid-conversation. Local models are supported when data has to stay on the device.
- Skills and plugins. Agents connect to tools and MCP-compatible plugins; the project puts its library at over 10,000.
- Work with several agents. Agent Groups split a task across agents, Pages let several of them write on a shared context, Schedule runs them at set times, and Projects and Workspaces keep it organised.
- Memory you can read. Agents keep structured, editable memory, so you can see what they remember and change it.
What the desktop app adds
According to its own README: native menus, shortcuts and notifications, a built-in updater, encrypted token storage with an OAuth sign-in flow, a multi-window layout, and sync with a remote LobeHub instance. That instance can be LobeHub’s own account service, or a server you host yourself using the project’s self-hosting guide.
What it costs to run
The app costs nothing to download. The models are a separate matter: hosted models such as OpenAI’s need an API key from that provider, and those calls are billed by the provider. Local models cost only your hardware.
The licence, briefly
LobeHub uses the LobeHub Community License, which is Apache 2.0 with added conditions. Commercial use is allowed, including as a front end and back end service, as long as you do not modify the source. Developing and distributing a derivative work needs a commercial licence from LobeHub. Installing the app and using it yourself needs nothing extra.
About 2.2.17
Released on 11 September 2026. On the desktop side it adds Computer Use through auv, thumbnails when you hover over tabs, settings for the completion sound, and clearer recovery when sign-in fails. For people running their own LobeHub server, it also adds database migrations 0158 to 0161; the release notes ask you to back up the database before upgrading, and say rollback has not yet been verified.
Which file to download
Windows: LobeHub-2.2.17-setup.exe, 134.6 MB. macOS: separate builds for Apple Silicon and Intel. Linux: an AppImage and a .deb here, with .rpm, .snap and a tar.gz on the release page.
When to use something else
If you want one chat window talking to one model, this is a lot of app for the job. If your team needs a server everyone signs in to, and you want a licence with no conditions on modifying it, LibreChat under MIT is closer. And LobeHub describes itself as under active development, so expect things to move between releases.
Key capabilities
Agents as the unit of work
Describe what you need and the Agent Builder configures an agent you can use at once. Agents connect to skills and MCP-compatible plugins, so the same agent can search, write and call tools.
Any model, including local ones
GPT-5, Claude, Gemini, DeepSeek, Llama, Qwen and more from over a dozen providers, switchable mid-conversation. When data has to stay on the machine, run a local model instead.
Groups, pages and schedules
Agent Groups split one task across several agents. Pages let them write on a shared context, Schedule runs them at set times, and Projects and Workspaces keep the work organised.
Memory you can read
Agents build structured memory from how you work, and it stays editable. You can see what an agent remembers about you and change it, instead of trusting a hidden profile.
A native desktop app
Native menus, shortcuts and notifications, a built-in updater, encrypted token storage, several windows at once, and sync with LobeHub's service or your own self-hosted server.
The licence in one paragraph
Apache 2.0 with conditions. Using it, including commercially and unmodified as a service, is allowed. Building and distributing a modified version needs a commercial licence from LobeHub.
Questions & Answers
Is it free?
The app is free to download and use. Hosted models are billed by their providers through your own API keys, and running local models costs only your hardware.
Is this the same LobeHub as the web version?
Yes. The desktop app is an Electron build of the same product, with native menus, notifications, a built-in updater and multi-window support added.
Can I keep my data on my own machine?
You can use local models when data has to stay on the device. Syncing between devices goes through a remote LobeHub instance, which can be a server you run yourself instead of LobeHub's service.
Which macOS file do I need?
Apple menu → About This Mac. If the chip says Apple M-something, take the arm64 build. If it says Intel, take x64.
Can I modify it and ship my own version?
Not without a commercial licence. The LobeHub Community License allows commercial use of the unmodified software, including as a service, but developing and distributing a derivative work needs a licence from LobeHub.
I run a self-hosted LobeHub server. Anything to know about 2.2.17?
It adds database migrations 0158 to 0161. Back up the database first; the release notes say downgrade and rollback have not yet been verified.
Is Bineret connected to LobeHub?
No. This page describes LobeHub Desktop and links to the builds LobeHub LLC publishes on GitHub.
Tutorials
Windows
Download LobeHub-2.2.17-setup.exe and run it. If Windows SmartScreen shows a warning, confirm the file came from the GitHub link on this page before choosing to run it.
Check the Windows download
The release publishes the installer's SHA-512 in latest.yml, encoded as base64. With Python installed, this prints the same value for an intact download:
python -c "import hashlib,base64;print(base64.b64encode(hashlib.sha512(open('LobeHub-2.2.17-setup.exe','rb').read()).digest()).decode())"For 2.2.17 it should print:
PGPuqnvcjIQSGF8+RCGJKqT5U+9A5i/AJxkerF9I14Rn8aFzqFEJWb1M/4YFtVl38vwJOKrrRXpUenG3u3I5aw==macOS
Macs with an M-series chip take LobeHub-2.2.17-arm64.dmg; Intel Macs take LobeHub-2.2.17-x64.dmg. Open the disk image and drag LobeHub into Applications. The desktop README states the macOS builds are notarized.
Linux
chmod +x LobeHub-2.2.17.AppImage
./LobeHub-2.2.17.AppImage
# or on Debian and Ubuntu
sudo apt install ./lobehub-desktop_2.2.17_amd64.debFirst steps
Sign in, or connect the app to a LobeHub server you host. To chat with a hosted model you need an API key from that provider; the LobeHub README walks through getting one from OpenAI as its example. Then describe an agent to the Agent Builder and start from there.
Updates
The app has a built-in updater, so you do not need to come back here for each release.
Support
Where support comes from
LobeHub LLC and its contributors maintain LobeHub. Start with the documentation, and report bugs as GitHub issues with your operating system, the app version and what you did before the problem. The project also runs a Discord community, linked from its README.
What Bineret covers
This page. If a download link is broken or something written here is wrong, tell us and we will fix it. We do not provide model API keys, accounts, or set-up help for LobeHub.
Commercial licensing
Questions about derivative works or licensing go to LobeHub directly at hello@lobehub.com, the address given in its licence.