Lobe Chat now points to an agent workspace
The Lobe Chat repository now redirects to LobeHub, and release v2.2.19 makes the change in scope hard to miss. You can organize agents into projects, schedule work, and share a workspace. Someone looking for a chat window to put over an API should pause here. Much of this product's appeal depends on wanting agents to keep doing work after the conversation ends.
That scope has a physical cost: our checkout contained 19,375 files and occupied 266.4 MB before installation. The README describes agent groups and editable personal memory, with MCP-compatible plugins connecting outside tools. Those features could save a team from moving context between separate chats. They also give the operator more behavior to understand before granting an agent access to a working account.
What happened when we ran it
Our npm install of commit e2dd39d failed after 359 seconds on October 14, 2026. The sandbox used the lab-node:22 image with 3 CPUs and 8 GB of RAM in an unprivileged container without secrets. The final error was EUNSUPPORTEDPROTOCOL, followed by Unsupported URL Type "workspace:": workspace:*. That is the installation result we can report. The supplied measurements contain no successful build or test result.
The log tail also points to npm's dependency-resolution report and includes a warning that force mode disables recommended protections. It does not identify every earlier dependency conflict. Blaming a particular library, Debian package, or Node defect would go beyond the evidence. Our run stopped before we could judge the application itself, so this review's product assessment rests on the documented behavior and current project activity.
The repository scan found 31 CI workflow files, a Dockerfile, a tests directory, and monorepo workspaces across roughly 2,837,167 source lines. Those are useful signs of an established development process. They do not turn a failed local install into a passing one. Teams planning a fork should reproduce their own supported development path before committing to maintaining a product of this size.
The documented development path uses pnpm
The current development instructions call for pnpm install and pnpm dev. A separate Bun command starts the SPA frontend. Our 359-second npm failure therefore describes the harness path, rather than proving that the documented pnpm path is broken. Package-manager choice matters in this workspace repository, but we did not rerun installation with another manager and cannot promise that doing so resolves everything.
For deployment, the README supplies hosted buttons and a Docker Compose route that initializes infrastructure with a downloaded setup script. Its example requires an OpenAI API key and allows a proxy endpoint through environment variables. Read that script and the linked deployment guide before execution. A running web interface still needs a configured model service, and provider credentials create an operating responsibility that a deployment button cannot handle for you.
Scheduled tasks need supervision when their runtime disappears
Open issue 20663 concerns self-hosted v2.2.19 specifically: an asynchronous agent task can remain marked running after its live runtime becomes unavailable. The reporter describes completion and cancellation commands returning HTTP 409. That is relevant if you want scheduled work to finish unattended. A stuck status can leave an operator unsure whether to retry a task or wait for work that has already stopped.
The report is separate from our sandbox failure, which never reached that runtime. The 31 workflow files show upstream automation exists, while this issue shows a concrete operating case still needs attention. I would start with work whose outcome a person can inspect, such as a draft report, before giving scheduled agents authority over external changes. A shared workspace makes results visible, but visibility alone cannot establish whether an operation completed correctly.
The community license deserves a commercial review
The LobeHub Community License adds conditions to Apache 2.0, including a commercial-license requirement for developing and distributing a derivative work. The README names that license even though a badge still refers to Apache. If your plan is to modify LobeHub into a commercial product, resolve those terms before spending engineering time on a fork. Access to source does not settle redistribution rights.
That decision comes before absorbing a 266.4 MB checkout into your product workflow. For an internal agent workspace, the feature set may justify the codebase and deployment machinery. For a company shopping for a permissively licensed chat foundation, the extra license conditions are a specific reason to compare alternatives. LibreChat is worth examining for multi-provider conversations and agents. Open WebUI is another candidate when an interface for Ollama or compatible APIs is the central requirement.
October activity shows maintenance alongside unfinished fixes
GitHub recorded a push on October 15, 2026, and 1,081 open issues and PRs in the repository metadata we fetched. Release v2.2.19 shipped on October 7. Recent pull requests address workspace credential detection and incorrect completion reporting for parked child agents. Those changes, together with the recent issue activity, show active maintenance. An open pull request describes proposed work, so its existence does not establish that an installed release contains the fix.
LobeHub earns a closer look when shared agents and scheduled work are requirements you already have. Our 359-second install failure means I would make a working pnpm development setup an adoption condition for any team planning source changes. For production use, require a demonstrated way to recognize and recover a stalled task. The first unattended job should have an outcome you can verify without trusting its status label.