LobeHub

What it is

LobeHub organises AI agents into ongoing operation rather than one-off chats. The framing the project uses is a "Chief Agent Operator": you hire agents, schedule them, and read their reports, while the agents run around the clock. The unit of work is an agent with a persistent role, not a conversation you have to babysit.

It grew out of LobeChat, the widely used open-source chat interface, and the repository has since been rebranded to match the broader product.

What that looks like in practice

Agents are long-term teammates with continuity, not executors you re-brief each session. They can be composed into groups that share context on a project, so several specialised agents — a market analyst, a designer, a product operator — can work the same brief and each contribute from their own angle. Scheduled operation means work continues between your sessions, and you catch up through reports instead of watching a stream.

The platform speaks MCP, so agents can pull in tools from the wider Model Context Protocol ecosystem, and there is a plugin system and agent marketplace on top of that. Model support is broad and provider-agnostic — you can bring your own API keys or use LobeHub's hosted credits.

Self-hosted or cloud

This is the unusual part of LobeHub's positioning: the whole thing is Apache 2.0 and genuinely self-hostable. One-click deploys are documented for Vercel, Zeabur, Sealos and Alibaba Cloud, and there is a Docker path for running it on your own infrastructure. Self-hosting costs nothing beyond your own model API spend.

The hosted service is metered in credits. The free tier gives 500,000 monthly credits and 10 MB of file storage with no card required — roughly 1,600 messages on a fast model — which is enough to evaluate it properly. Paid tiers run $9.90/month (Starter, 5M credits, 1 GB files, plus access to frontier models and early access to new ones), $19.90/month (Premium, 15M credits, 2 GB, priority email support) and $39.90/month (Ultimate, 35M credits, 4 GB, priority chat support). An enterprise edition adds a commercial licence, private deployment and custom integration at negotiated pricing.

Worth knowing

The credit model means your effective cost depends on which models you route to, so the tier prices are not directly comparable to a flat per-seat assistant subscription. If you already hold model API keys, the self-hosted route sidesteps credits entirely.